ChatGPT still leads the AI model race, but Gemini and Claude are changing the scoreboard

ChatGPT is still the name most people attach to generative AI, but the market share question is now too complex for a single percentage. A consumer app chart, a website traffic ranking, a business-spend index, a developer API report, a search-referral study and a model leaderboard all describe different markets. They overlap, but they do not measure the same thing.

The AI model market no longer has one scoreboard

That distinction matters because the leading AI systems are no longer just chatbots. ChatGPT is becoming a general consumer and work platform. Gemini is becoming a Google-wide assistant layer. Claude is strongest where professionals, coders and teams spend long sessions. Perplexity is an AI answer engine built around cited web research. Grok is tied to xAI’s models, the standalone Grok app and the social distribution of X. Each product competes for attention, but also for different types of work.

The freshest public data points tell the story. OpenAI says ChatGPT has more than 900 million weekly active users and more than 50 million consumer subscribers. Reuters, citing Sensor Tower, reported that ChatGPT crossed 1 billion global monthly active app users in May 2026, while Claude reached 56 million global monthly active app users and was growing faster year over year from a far smaller base. Google says the Gemini app has surpassed 900 million monthly users, more than doubling in a year. Similarweb’s May 2026 worldwide web ranking places chatgpt.com first in the AI Chatbots and Tools category, followed by gemini.google.com and claude.ai. Statcounter’s May 2026 AI chatbot chart gives ChatGPT 79.08%, Perplexity 7.67%, Google Gemini 7.03%, Microsoft Copilot 3.23% and Claude 2.98%.

Those figures appear to conflict because they track different things. ChatGPT can lead app usage, Gemini can surge in web visits, Claude can rise in enterprise spending, Perplexity can punch above its size in answer search, and Grok can look bigger when usage inside X is counted. None of these readings is fake. None is complete.

The market is moving from a simple adoption race to a layered power contest. The biggest question is no longer whether ChatGPT leads. It does. The harder question is where that lead is strongest, where it is narrowing, and which measurement layer will matter most for revenue, search behavior, brand visibility and enterprise software budgets.

Market share depends on the layer being measured

The phrase “AI model market share” sounds precise, but it hides at least six separate markets. Consumer assistant share measures people using products such as ChatGPT, Gemini, Claude, Grok and Perplexity through apps or websites. Web traffic share measures visits to domains. Mobile app share measures installs, active users or engagement inside apps. Enterprise adoption measures companies paying for or using tools. API usage measures tokens, requests or developer spending. AI search share measures how often answer engines send users to websites or replace traditional search behavior.

These layers produce different winners because the products enter users’ lives through different doors. ChatGPT enters through habit and brand memory. Gemini enters through Google’s existing accounts, Android, Search, Workspace and Chrome. Claude enters through professional trust, coding workflows and long-form work. Perplexity enters through research and cited answers. Grok enters through X and an audience already trained to use public conversation as a discovery feed.

The cleanest way to read the market is to avoid treating any single data provider as the scoreboard. Similarweb is useful for web visits and engagement on public domains. Statcounter gives a high-level AI chatbot share chart, but its public page does not expose every methodological detail on the visible page. Sensor Tower is stronger for mobile app usage estimates. Ramp sees business spending among companies using Ramp. OpenRouter sees model use routed through its own multi-model inference platform. Anthropic and OpenAI publish first-party usage studies, but each sees only its own product.

This is why market-share claims often sound more dramatic than the underlying evidence supports. A headline saying Gemini has taken share from ChatGPT on the web may be true inside a seven-app web-visit set. It does not prove Gemini has passed ChatGPT in total active users. A chart showing Claude gaining enterprise spend does not mean Claude has matched ChatGPT in consumer reach. A benchmark win by a Claude, Gemini, GPT or Grok model does not automatically create distribution.

The AI market is split between reach, frequency, depth, revenue and trust. Reach asks how many people touch the product. Frequency asks how often they return. Depth asks how much work they hand over. Revenue asks who pays and how much. Trust asks whether users and companies will rely on the output for work that carries risk. ChatGPT is strongest on reach. Claude is increasingly visible on depth. Gemini is advancing through distribution. Perplexity has a research identity. Grok has an attention channel and an unusually public usage pattern.

ChatGPT remains the consumer default

ChatGPT’s market position starts with a simple fact: it became the default word for consumer AI before most rivals had mature products. That early lead still matters. OpenAI’s own 2026 page says ChatGPT has more than 900 million weekly active users and more than 50 million consumer subscribers. Reuters, citing Sensor Tower, reported that ChatGPT reached 1 billion global monthly active app users in May 2026, roughly three years after launch, making it the fastest app to hit that threshold by Sensor Tower’s estimate.

The scale is not only a marketing statistic. It changes the behavior of companies, schools, publishers, software vendors and ordinary users. A product with that many weekly users becomes a default surface for questions, drafting, coding help, tutoring, shopping research, trip planning, data interpretation, image generation and workplace tasks. For many people, the first question is not “Which model is best?” It is “Can ChatGPT do this?”

OpenAI’s usage paper with Harvard economist David Deming gives a rare internal view of how this adoption spread. By July 2025, the paper says ChatGPT had more than 700 million weekly active users sending 18 billion messages each week, or about 2.5 billion messages per day. It also found that non-work messages had grown faster than work messages, accounting for more than 70% of all usage by June 2025, while “Practical Guidance,” “Seeking Information” and “Writing” together accounted for nearly 80% of conversations.

That usage mix explains why ChatGPT’s lead is hard to dislodge. It is not only a workplace assistant or a search tool. It is a broad, messy, everyday utility. People use it for decisions, drafts, explanations, recipes, emotional framing, software help, schoolwork, travel, health questions, financial questions, image prompts and personal planning. Some of those uses are serious. Some are casual. The breadth is part of the moat.

ChatGPT also has a subscription base most rivals have not matched publicly. More than 50 million consumer subscribers means OpenAI has turned a free consumer habit into a paid software business at unusual speed. That payment layer matters because model serving is expensive. Consumer attention is useful, but paid usage gives OpenAI room to fund new models, infrastructure, product features and enterprise sales.

The threat is not that ChatGPT suddenly disappears as the leader. The threat is slower and more practical. Users may keep ChatGPT as the default while adding Claude for coding, Gemini for Google-connected work, Perplexity for cited research and Grok for X-native questions. In that future, ChatGPT keeps the largest audience but loses some of the highest-value tasks.

Gemini turns Google distribution into usage

Gemini’s rise is the clearest reminder that AI market share is not only about model quality. It is also about distribution. Google has Search, Android, Chrome, Gmail, Docs, YouTube, Maps, Photos, Workspace and Cloud. No AI-native startup can copy that footprint. The challenge for Google has been converting those surfaces into a product people actively choose, rather than AI features they passively encounter.

Google says the Gemini app had over 400 million monthly active users at I/O 2025, more than 450 million by Alphabet’s Q2 2025 earnings call, over 750 million by Q4 2025, and more than 900 million monthly users by I/O 2026. Google also said daily requests grew more than seven times across that year.

That growth puts Gemini in a different class from most ChatGPT challengers. It is not simply a “second-place chatbot.” It is becoming Google’s consumer AI brand across apps, devices and search behavior. When Gemini appears in Workspace, Chrome or Android, it lowers the cost of trying the assistant. When it is linked to Google accounts and user data with opt-in personal features, it can compete on context that standalone tools must request or import.

Similarweb’s May 2026 ranking already places gemini.google.com second among AI Chatbots and Tools websites worldwide, behind chatgpt.com and ahead of claude.ai. A separate Similarweb-based analysis by Momentic estimated that, within a seven-app web-visit set, ChatGPT’s share fell from 76.5% in February 2025 to 54.7% in April 2026, while Gemini rose from 5.6% to 27.4%. That Momentic figure is not total global AI usage, but it captures the direction of web traffic among major standalone assistant domains.

Gemini’s biggest advantage is also a source of ambiguity. Much of Google’s AI usage sits inside existing products, not always inside the Gemini app or gemini.google.com. AI Overviews, AI Mode, Workspace features and Gemini-powered Android experiences may drive AI adoption without looking like standalone chatbot usage. That makes Gemini difficult to compare with ChatGPT on a clean product-to-product basis.

If the market is measured by explicit visits to AI assistant websites, ChatGPT still leads. If it is measured by AI embedded across default computing surfaces, Google’s position is far larger than a chatbot chart suggests. This is the strategic tension. Google does not need every user to open Gemini first if Gemini becomes the intelligence layer inside products people already use all day.

Claude shows the power of high-intent adoption

Claude is much smaller than ChatGPT and Gemini by broad consumer reach, but it has become one of the strongest signals that market share is not only about headcount. Claude’s brand is tied to long-form reasoning, careful writing, coding, document work and professional trust. That gives Anthropic a narrower but more intense audience.

Reuters, citing Sensor Tower, reported that Claude had 56 million global monthly active app users in the second quarter to date in 2026, while its year-over-year monthly active user growth was about 640%, far above ChatGPT’s 62% growth from a much larger base. The same Reuters report said U.S. ChatGPT users who installed Claude in the first quarter of 2026 spent 5% less time on ChatGPT one month after installation compared with their prior average.

That 5% time shift is more interesting than it first appears. A user does not have to abandon ChatGPT for Claude to matter. A small time shift among high-value users can redirect writing projects, code sessions, business analysis, legal drafts, policy memos and software work. In enterprise AI, minutes are not equal. A casual recipe prompt and a five-hour coding session both count as use, but they do not carry the same revenue or switching signal.

Anthropic’s own Economic Index shows a platform with work-heavy usage patterns. Its March 2026 report studied Claude usage in February 2026 and found that Claude.ai use cases had diversified: the top 10 tasks made up 19% of traffic, down from 24% in November. It also said coding tasks continued to move from augmentative Claude.ai usage toward more automated workflows in Anthropic’s first-party API traffic.

Claude Code is a major reason Claude’s market presence looks larger inside professional communities than in broad consumer app charts. Anthropic describes Claude Code as an agentic coding system that reads a codebase, makes changes across files, runs tests and delivers committed code. In May 2026, Anthropic announced higher usage limits, including doubled five-hour rate limits for Claude Code on Pro, Max, Team and seat-based Enterprise plans, plus higher API rate limits for Claude Opus models.

Ramp’s March 2026 AI Index gives another angle on Claude’s business traction. Ramp reported that overall business AI adoption rose to 47.6% of businesses in February 2026, while 24.4% of businesses were using Anthropic. The same report said Anthropic adoption grew 4.9% month over month, OpenAI adoption fell 1.5%, Google adoption grew slightly to 4.7%, and xAI remained under 2% among businesses in Ramp’s data.

Claude’s market-share story is a depth story. It does not need to match ChatGPT’s mass audience to reshape the high-value end of the market. If developers, analysts, founders, consultants, writers and enterprise teams spend their hardest work sessions in Claude, Anthropic can gain economic share faster than consumer-share charts imply.

Perplexity competes through search intent rather than general chat

Perplexity is often placed beside ChatGPT, Claude, Gemini and Grok, but its identity is different. It is not mainly a companion chatbot or an all-purpose writing assistant. Perplexity describes itself as an AI answer engine that researches the open web in real time, returns concise cited answers and routes queries across multiple frontier models.

That positioning gives Perplexity a different type of market share. It competes for high-intent research sessions: “Find the answer,” “compare the options,” “source this claim,” “summarize the market,” “tell me what changed,” “show me citations.” A Perplexity user may also use ChatGPT or Claude for drafting, but use Perplexity when source grounding matters.

Statcounter’s May 2026 AI chatbot chart puts Perplexity at 7.67%, slightly above Google Gemini’s 7.03% on that chart and far above Claude’s 2.98%. That figure should not be treated as total user share, because Statcounter’s public chart is one measurement system, not a full census of all AI usage. It is still notable because Perplexity appears stronger in some search-oriented measurements than in raw app-user rankings.

Perplexity’s strategic problem is scale. It has a clear product identity and strong brand association with citations, but it does not have OpenAI’s consumer default position, Google’s distribution, Anthropic’s coding mindshare or xAI’s ownership link to a major social network. It must win by making research, shopping, browsing and answer verification more useful than a general assistant’s web mode.

The company is acting like a serious long-term competitor. Reuters reported on June 9, 2026, that Perplexity plans to pursue an IPO in 2028 regardless of how Anthropic or OpenAI listings perform, citing CEO Aravind Srinivas. The report also quoted Perplexity’s chief business officer saying the company has kept 2028 as its earliest IPO date while building a healthy, high-growth business.

Perplexity’s share is best understood as answer-engine share, not chatbot share. Its product is strongest when users want checkable facts, source links and live web context. That makes it especially relevant for publishers, brands, analysts and SEO teams tracking AI-driven discovery. Perplexity may never need to become the most-used assistant if it becomes one of the most-used places for cited answers.

Grok is an attention product as much as an assistant

Grok’s market position is difficult to measure because it exists across multiple surfaces: the standalone Grok app, grok.com, xAI’s API and usage inside X. That makes Grok different from Perplexity, which has a search identity, and Claude, which has a professional-work identity. Grok is tied to public conversation, memes, real-time events, creator culture and Elon Musk’s platform network.

xAI presents Grok as a frontier AI system for reasoning, code, voice, images and video, with API access for developers. Similarweb’s May 2026 AI Chatbots and Tools ranking puts grok.com fifth worldwide, behind chatgpt.com, gemini.google.com, claude.ai and chat.deepseek.com, with an average visit duration of 11 minutes and 11 seconds and 12.93 pages per visit. That engagement figure is unusually high among the top general AI sites listed by Similarweb.

Grok’s advantage is that X can turn AI usage into public interaction. Users do not only ask Grok private questions. They summon it into debates, ask it to interpret posts, use it to generate media, or treat it as a public referee. A 2026 paper on Grok in the wild found that Grok is called by millions of people each week on X, and that public social usage creates roles different from private chatbot use, including information provider, truth arbiter, advocate and adversary.

That public role gives Grok visibility, but it also raises safety and trust risks. Grok’s growth has been accompanied by intense scrutiny around image generation, moderation and public misuse. In a market where enterprise buyers care about compliance, that reputation can cut both ways. Some users value Grok’s looser public persona. Some organizations may see that same quality as risk.

The a16z consumer AI review noted that Grok launched as a standalone app in January 2025, added companion experiences and image and video models during the year, and reached 9.5 million daily active users and 38 million monthly active users by mid-December 2025 according to Sensor Tower data cited in the report. That is not the same as total 2026 usage, but it shows a steep ramp from zero standalone-app users at the start of 2025.

Grok’s share is partly hidden inside X, and partly amplified by X. If measured only by standalone AI assistant apps, Grok may look smaller. If measured by the number of public conversations where an AI model becomes part of the social feed, Grok becomes more important.

Web traffic shows ChatGPT ahead with Gemini closing part of the gap

Public web traffic is one of the easiest layers to observe, but it is also one of the easiest to overread. Similarweb’s May 2026 worldwide AI Chatbots and Tools ranking shows chatgpt.com first, gemini.google.com second, claude.ai third, chat.deepseek.com fourth, grok.com fifth and perplexity.ai eighth. That ranking captures visits to public domains, not total usage across mobile apps, embedded product surfaces, APIs or enterprise deployments.

Still, the ranking is useful because web usage is where many users compare assistants directly. A user opening chatgpt.com, gemini.google.com, claude.ai or perplexity.ai is making a relatively explicit choice. Similarweb also reports engagement metrics: in May 2026, gemini.google.com had an average visit duration of 6 minutes and 59 seconds, claude.ai 6 minutes and 2 seconds, chatgpt.com 5 minutes and 58 seconds, perplexity.ai 4 minutes and 28 seconds and grok.com 11 minutes and 11 seconds.

The ranking tells a more competitive story than a broad active-user chart. ChatGPT is first, but Gemini and Claude are visible immediately behind it. Grok appears in the top five. Perplexity is not first-tier by total AI chatbot web visits, but still sits in the top ten and has a distinct research use case.

Momentic’s Similarweb-based seven-app analysis gives a sharper time series. It estimated that ChatGPT’s worldwide web-visit share among ChatGPT, Gemini, Claude, DeepSeek, Grok, Perplexity and Microsoft Copilot fell from 76.5% in February 2025 to 54.7% in April 2026. Gemini rose from 5.6% to 27.4%. Claude rose to 8.2%. DeepSeek was 4.1%, Grok 2.8%, Perplexity 1.5% and Copilot 1.3% in April 2026.

Those numbers do not mean ChatGPT is losing the whole market. They mean its share of measured web visits among a defined set of seven assistant sites has narrowed. This is exactly the kind of metric that gets distorted when reduced to a headline. The underlying signal is real: Gemini and Claude gained web attention. The broader inference needs care.

Web traffic also undercounts products embedded elsewhere. Gemini’s Google surfaces, Copilot in Microsoft products, Grok inside X, Claude in enterprise systems and ChatGPT inside third-party workflows may not show up fully in domain rankings. Web share is a visibility metric, not a full market census.

Current public market-share signals by measurement layer

This table should be read as a map of measurement systems, not as a unified ranking. The winner changes when the unit changes from app users to web visits, business spend, assistant search, or developer routing.

App usage still favors ChatGPT by a wide margin

Mobile app usage makes ChatGPT’s lead look larger than web traffic alone suggests. Reuters reported that ChatGPT crossed 1 billion global monthly active app users in May 2026, based on Sensor Tower estimates. In the same report, Claude’s global monthly active app users were 56 million in the second quarter to date. That is rapid growth for Claude, but it still places ChatGPT in a separate scale category.

App usage matters because phones are where default consumer habits become durable. A chatbot used through a browser tab can be compared more easily. An app with notifications, voice mode, images, files, memory, widgets and account history becomes part of daily behavior. Once a user has months of conversations, custom instructions, saved projects, files and subscription history in one app, switching becomes less casual.

Google is the main exception because Gemini’s mobile footprint is not only a standalone app question. On Android, Gemini can be the assistant layer, the search layer and the device-level AI layer. That creates a different type of app distribution. The user may not think “I am opening an AI chatbot app.” The user may experience Gemini as the intelligence inside a device action, a search result, a photo task or a Workspace document.

Claude has a different challenge. Its user base appears smaller but more concentrated among high-intent users. If Claude takes five percent of ChatGPT time among users who install it, as Sensor Tower data cited by Reuters suggests, the revenue impact could be larger than the share shift sounds. A high-income professional moving serious work sessions from ChatGPT to Claude is more valuable than a casual user testing Claude once.

Perplexity’s app position sits between research and browsing. Its mobile experience is not simply a chatbot; it is an alternate search front end. That makes Perplexity vulnerable to platform defaults but attractive to users who dislike search-result clutter and want cited answers.

Grok’s app trajectory is tightly tied to X. If Grok is used as a standalone AI app, it competes in the same app-store charts as ChatGPT and Claude. If it is used inside X, it behaves more like a social feature. That makes comparisons messy, but strategically useful for xAI because distribution inside a social feed can create repeated exposure.

The app market still says ChatGPT is the mass-market winner. The change is that app usage no longer captures the whole AI market. Serious work is moving into coding agents, browsers, search layers, office suites, developer APIs and embedded enterprise tools.

Search and answer share make Perplexity more visible

Search share is not the same as chatbot share. A user who asks Perplexity a sourced question is behaving differently from a user who asks Claude to rewrite a legal memo or ChatGPT to generate an image. Search intent is narrower, but it can be commercially powerful because it often sits near decisions: buying, comparing, citing, learning, planning or verifying.

Perplexity benefits from this distinction. Its official description stresses real-time web research, concise cited answers and multi-model routing. That gives it a clean answer-engine identity.

The Statcounter AI chatbot chart for May 2026 places Perplexity at 7.67%, ahead of Google Gemini’s 7.03% and Claude’s 2.98%, with ChatGPT at 79.08%. That chart is not the same as monthly active users, but it shows that Perplexity can appear stronger in measurement systems connected to AI chatbot share and search behavior than it appears in broad app rankings.

This matters for publishers and brands. If Perplexity has a smaller user base but a higher share of research-heavy sessions, it can shape which sources are cited, which brands are visible, and which articles are used to answer commercial questions. A single Perplexity answer may contain citations that affect a purchasing or business decision. A casual ChatGPT exchange may never send traffic anywhere.

ChatGPT and Gemini are also moving deeper into search. ChatGPT Search, Google AI Mode and AI Overviews all put general assistants closer to traditional search behavior. Google has the biggest existing search distribution, but OpenAI has the strongest AI-native consumer brand. Perplexity’s task is to defend a specialized identity while the largest platforms absorb more search functions.

The search layer is likely to be more plural than the assistant layer. Users may keep one general assistant but switch among answer engines depending on the query. They may ask Google for local results, Perplexity for cited research, ChatGPT for synthesized planning, Claude for document analysis and Grok for real-time social context.

For SEO and GEO strategy, that means brand visibility can no longer be reduced to Google rankings. The practical question is whether a brand is understandable, citeable and trusted across AI answer systems. Perplexity’s smaller size does not make it irrelevant. It may matter precisely because its users arrive with higher research intent.

Enterprise spend gives Claude a stronger story than consumer charts

Enterprise adoption changes the market-share picture because businesses do not buy AI the way consumers try apps. They care about data handling, admin controls, security, procurement, compliance, reliability, integration, cost predictability and support. In that setting, a smaller consumer product can become a larger economic force.

Ramp’s March 2026 AI Index is one of the most useful public signals because it tracks business spending through Ramp’s own customer base. Ramp reported that 47.6% of businesses in its data used AI in February 2026, with 24.4% using Anthropic. It said Anthropic adoption grew 4.9% month over month, while OpenAI adoption fell 1.5%, Google adoption grew slightly to 4.7%, and xAI remained under 2% among businesses on Ramp.

This does not mean Anthropic is larger than OpenAI across all enterprises. Ramp itself says OpenAI is still the AI model company used by the most businesses in its data. It does mean Anthropic is gaining business adoption quickly, especially among companies paying for AI services for the first time. Ramp also reported that Anthropic was winning about 70% of head-to-head matchups against OpenAI among first-time AI-service buyers in its business-spend data.

Claude’s enterprise appeal is tied to the type of work it is used for. Anthropic’s Economic Index reports show coding, automation and professional tasks as major usage categories. Claude Code’s ability to inspect codebases, edit files, run commands and connect to development workflows turns the model into work infrastructure rather than a chat surface.

OpenAI is not weak in enterprise. It has massive consumer familiarity, ChatGPT Enterprise, business seats, API usage and a broad product roadmap. OpenAI’s enterprise report says consumer adoption has accelerated workplace use, and that ChatGPT for Work had more than 7 million total seats in 2025 while Enterprise seats had grown nine times year over year.

Google’s enterprise position is different again. Gemini can arrive through Workspace, Cloud, Vertex AI and existing contracts. Microsoft Copilot has its own embedded enterprise distribution through Office and Windows. Anthropic’s challenge is that it must win account by account, but its strength is that buyers who choose Claude often do so for specific workflows, not because it was bundled by default.

In enterprise AI, share of wallet may matter more than share of users. A million casual consumers are impressive. A smaller group of developers running expensive coding agents all day may produce more revenue and a stronger product moat.

API usage is fragmented and partly hidden

The API market is the hardest layer to measure publicly. OpenAI, Anthropic, Google, xAI, Mistral, Meta, DeepSeek, Cohere and many other providers sell access through direct APIs, cloud platforms, model routers and enterprise agreements. A developer may use Claude through Anthropic’s API, Amazon Bedrock or Google Cloud Vertex AI. A company may use GPT models through OpenAI directly, Azure, internal wrappers or third-party applications. A user may use Gemini through Google AI Studio, Vertex AI, Android, Search or Workspace.

This fragmentation makes market-share claims fragile. Token volume, API revenue, number of developers, number of business accounts and number of applications are not interchangeable. A cheap model with huge token volume can look large by tokens but smaller by revenue. A premium model with lower volume can produce more revenue. A model used inside a popular app can be invisible as a separate API provider.

OpenRouter’s State of AI report is useful because it studies real usage across a multi-model inference platform, analyzing more than 100 trillion tokens. The authors describe OpenRouter as a hub across many closed-source APIs and open-weight deployments, giving a window into how developers and end users invoke models across tasks, geography and time. The report highlights open-weight adoption, creative roleplay, coding assistance and agentic inference.

Yet OpenRouter is still one platform. It cannot represent all API usage because many customers use OpenAI, Anthropic, Google or xAI directly. The same caution applies to any model-router ranking. It reveals behavior among users who choose the router; it does not reveal the whole market.

Artificial Analysis tracks provider performance across hundreds of AI model endpoints, including price, output speed, latency and context window. Its provider leaderboard is not a market-share chart, but it shows another reason API choice is becoming more granular. Developers compare not only intelligence but also speed, context window, first-token latency and blended token price.

The API market is moving toward portfolio behavior. Developers no longer need to choose one model provider for every task. They can route cheap tasks to a fast model, coding tasks to Claude, consumer chat to GPT, long-context analysis to Gemini, real-time social queries to Grok, and cited search to Perplexity or a retrieval stack. This weakens the idea of a single AI market share winner.

The hidden API layer also explains why consumer share and economic share can diverge. ChatGPT may dominate user count. Claude may win valuable coding workloads. Gemini may process immense token volume inside Google products. Open-weight models may take share in cost-sensitive deployments. Market share becomes a portfolio of workloads.

Coding agents changed Claude’s competitive position

Coding has become one of the most commercially important AI use cases because it can produce direct productivity gains, consumes large numbers of tokens and fits well into paid professional workflows. Claude’s rise is closely tied to this shift. Anthropic’s branding, model releases and product design have leaned into coding and agentic software work.

Claude Code is the clearest example. Anthropic says the system reads codebases, makes changes across files, runs tests and delivers committed code. Its documentation describes Claude Code as available in the terminal, IDE, desktop app and browser, with the ability to work across files and development tools.

This is not the same as a chatbot answering a programming question. A coding agent becomes part of the production workflow. It can touch files, run commands, interpret errors and iterate. That increases switching costs because the tool learns the project context and the team builds habits around it. It also increases usage volume because agentic coding requires many model calls, not one answer.

Anthropic’s Economic Index says coding tasks continue to move from augmentative Claude.ai usage into more automated first-party API workflows. That distinction is central. Early coding assistance meant asking a model for help. Agentic coding means delegating tasks inside a codebase.

OpenAI is not conceding coding. Codex and ChatGPT’s developer features are central to OpenAI’s business strategy, and Reuters has reported OpenAI’s plans to overhaul ChatGPT toward a broader “superapp” experience with coding tools and agents. But Claude’s brand among developers has become strong enough that it can pressure OpenAI in one of the highest-value usage categories.

Google also has a serious coding path through Gemini, Vertex AI, AI Studio, Android development, Cloud and its own coding tools. The question for Google is not whether it has model capability. It is whether developers treat Gemini as a first-choice coding partner rather than a tool that sits inside a broader Google workflow.

Coding share is likely to be more contested than general chat share. Developers compare output quality, repo understanding, tool use, latency, cost, context handling and trust under pressure. They also switch faster than ordinary consumers when a tool feels better. That makes coding an early-warning system for frontier model competition.

Google’s embedded AI advantage is hard to measure but hard to ignore

Google’s biggest AI advantage rarely appears cleanly in chatbot market-share charts. It owns many of the surfaces where people already search, write, watch, navigate, photograph, schedule and communicate. Gemini can enter those workflows as a feature, an assistant, an app, a search mode, a device layer or a developer model.

Alphabet’s June 2026 investor presentation says Gemini has more than 900 million monthly users and that Google has started rolling out Personal Intelligence across Search and the Gemini app, connecting to Google apps for personalized suggestions in more than 190 countries. It also describes Search and Gemini entering an agent era, with Gemini Spark designed to work in the background and connect with Google tools and third parties through the model context protocol.

This embedded strategy creates a measurement problem. If a user receives an AI-generated answer in Search, asks Gemini in Android, uses “take notes for me” in Google Meet or gets help writing in Gmail, should that count as Gemini market share? From a product standpoint, yes. From a standalone chatbot chart, often no.

Google’s I/O 2025 and earnings data show how quickly it can move usage across surfaces. Google said it processed 480 trillion monthly tokens across products and APIs at I/O 2025; by Q2 2025, Alphabet said that had doubled to more than 980 trillion monthly tokens. Those token figures are not consumer market share, but they show AI moving through Google’s infrastructure at a scale that is difficult for standalone apps to match.

The strategic question is whether embedded AI creates loyalty or just exposure. A user may accept Gemini features in Gmail while still preferring ChatGPT for creative tasks and Claude for code. Google’s default reach is enormous, but active preference is harder. This is why Gemini’s standalone app growth matters. More than 900 million monthly users signals that Gemini is not only background infrastructure; it is becoming a destination.

Google’s AI share is probably understated by standalone chatbot charts and overstated by broad “AI feature reach” claims. The truth sits between them. Gemini is gaining real assistant usage, while Google’s broader AI footprint extends far beyond the Gemini app.

For competitors, this is uncomfortable. ChatGPT’s brand advantage is stronger than Google’s assistant brand, but Google’s distribution is wider. Claude’s professional trust is strong, but Google owns many workplace surfaces. Perplexity’s cited search identity is sharp, but Google owns search behavior. Grok owns social attention on X, but Google owns the default web entry point for billions.

OpenAI is turning ChatGPT from chatbot into platform

OpenAI’s market-share defense is no longer just model releases. It is product expansion. ChatGPT is becoming a platform for work, search, image generation, voice, coding, agents, third-party apps, memory and enterprise use. This is the natural move for a company with a massive consumer lead: turn habit into an operating layer.

OpenAI says ChatGPT is where people start with AI, with more than 900 million weekly active users and more than 50 million consumer subscribers. That phrase points to the strategy. If ChatGPT is the starting point, OpenAI can route users into paid features, business workflows, coding agents, shopping, image generation and app integrations.

Reuters reported in June 2026 that OpenAI was preparing a major ChatGPT overhaul to make it more like a multifunctional “superapp,” with advanced coding tools and AI agents, including interface changes that would guide users toward coding, image generation and integrations. The report also said enterprises accounted for about 40% of OpenAI’s revenue, with expectations for that share to rise.

This platform strategy has a clear defensive purpose. If users leave ChatGPT to code in Claude, search in Perplexity, use Gemini inside Google products or ask Grok inside X, OpenAI loses high-value sessions. By bringing more functions into ChatGPT, OpenAI tries to keep the user inside one environment.

The risk is product sprawl. A chatbot that becomes a superapp can become more powerful, but also more complex. Users who love ChatGPT for its simplicity may not want a crowded interface. Enterprise buyers may want deep integrations but also control. Developers may want coding tools that feel native to their environment, not merely features inside a general assistant.

OpenAI’s first-party enterprise report argues that consumer familiarity accelerates workplace adoption because employees already know ChatGPT. That is a real advantage. A company adopting ChatGPT does not need to teach employees what generative AI is. But familiarity alone is not enough when procurement teams compare model reliability, data handling, pricing and workflow fit.

OpenAI’s strongest moat is habit. Its challenge is turning habit into durable workflow ownership before rivals peel away specialized tasks. ChatGPT may remain the front door to AI. The fight is over what happens after users walk through it.

Perplexity’s citation layer is a threat to search, not only chatbots

Perplexity’s core product claim is not that it has the warmest assistant personality or the strongest coding model. Its claim is that answers should be grounded, cited and checkable. That makes it a more direct challenge to search engines, publisher discovery and research workflows than a generic chatbot chart suggests.

Perplexity says every answer is grounded in real-time web sources and carries inline citations, and that multi-model routing means the user does not need to pick which model handles a given task.

This design addresses a weakness in early chatbots: fluent answers with unclear sourcing. For readers, students, analysts and professionals, citations are not decoration. They are a way to verify whether the answer rests on current, credible material. A cited answer engine can replace a stack of search-result clicks for many queries.

The business risk is that citation-based AI search can reduce publisher traffic while still depending on publisher content. That tension is already central to the broader AI search debate. Perplexity’s strength with users can become a friction point with media companies if answers satisfy queries without sending enough traffic back to sources.

For brands, Perplexity creates a new discovery task. Classic SEO asked whether a page ranked in Google. AI answer visibility asks whether a brand is included in an answer, whether it is described correctly, whether the cited sources are authoritative, and whether the answer engine can distinguish current facts from outdated pages. A brand may rank well in search but be absent from AI answers if its pages are thin, ambiguous or hard to cite.

Perplexity’s smaller user count does not erase that impact. High-intent users ask high-value questions. A procurement manager researching vendors, a journalist checking a claim, an investor comparing markets or a student learning a field may rely heavily on cited answers. In those sessions, Perplexity’s influence can exceed its raw share.

Perplexity is less likely to become the biggest general assistant than to become one of the default places for sourced AI research. That makes it strategically relevant even when ChatGPT and Gemini dominate total reach.

Grok’s connection to X creates a different kind of AI usage

Most chatbot interactions are private. Grok’s role inside X makes a large share of its usage public or semi-public. People tag Grok in arguments, ask it to interpret viral posts, use it as a fact-checking or dunking tool, and bring AI into social conflict. That changes the model’s role.

A 2026 academic study of Grok on X found that public social usage creates roles beyond private assistance, including truth arbiter, advocate and adversary. It also found that Grok responded to 62% of requests in its sampled data, that 51% of interactions were in English, and that half of Grok responses received 20 or fewer views after 48 hours.

Those findings cut through hype. Grok is highly visible, but many interactions still have low reach. It can influence public discourse, but not every Grok response becomes viral. The value for xAI is that X gives Grok a native testing ground, public distribution and real-time data context.

Grok’s product identity is also more provocative than ChatGPT, Gemini, Claude or Perplexity. That can attract users who dislike cautious answers or heavy moderation. It can also make governments, advertisers and enterprises more wary. Public incidents around generated content, including legal and regulatory scrutiny, show how fast a looser product identity can become a liability.

Similarweb’s May 2026 ranking shows grok.com fifth in AI Chatbots and Tools, with longer average visit duration than the top three assistant sites. That does not prove Grok has more engaged users overall, but it suggests a distinctive web-use pattern among visitors to grok.com.

The future of Grok’s share depends on whether xAI can turn attention into trusted utility. Attention is powerful but volatile. Enterprise AI buyers tend to reward reliability, data control and predictable behavior. Consumer users may reward personality, speed and media features. Grok has a path in both markets, but the two paths require different trust signals.

Grok is not only competing to answer questions. It is competing to become the AI layer of a social network. That makes its market share hard to compare, but impossible to dismiss.

User overlap is lower than AI power users assume

People who follow AI closely often use many models. They compare Claude to ChatGPT, Gemini to Grok, Perplexity to Google, and open-weight models through routers. That behavior can create a misleading picture of the mass market. Most users are not constantly switching.

a16z’s State of Consumer AI 2025 report, citing Yipit data, said that for most of 2025 fewer than 10% of ChatGPT weekly users even visited another major model provider, and only 9% of consumers paid for more than one subscription across ChatGPT, Gemini, Claude and Cursor. The report described the assistant race as not strictly winner-take-all, but possibly “winner take most.”

This is a crucial point for market share. Multi-model behavior is common among developers, researchers, founders, SEO specialists, AI teams and journalists. It is much less common among ordinary users. A person who uses ChatGPT for homework help, recipes, emails and explanations may not care that Claude is better at a particular benchmark or that Gemini has a stronger Android integration.

At the same time, low overlap can change quickly when distribution shifts. Google can push Gemini to millions through existing surfaces. Anthropic can ride word-of-mouth among developers. Perplexity can win users through AI search habits. Grok can appear inside X interactions. Low overlap today does not guarantee low overlap tomorrow.

Subscription overlap matters even more than visit overlap. A user can try three tools for free, but paying for multiple AI subscriptions is harder to justify. At roughly $20 per month for many premium plans, four subscriptions become a real household or professional expense. That forces users to choose a primary paid assistant even if they sample others.

This is where free tiers and bundles matter. Gemini in Google One or Workspace, Copilot in Microsoft 365, ChatGPT Plus or Pro, Claude Pro or Max, Perplexity Pro and Grok through X-related plans all compete not only on quality but on perceived package value.

The mass market is less plural than the expert market. AI insiders see a portfolio. Ordinary users choose a default. The battle is over becoming that default, or becoming the specialist tool worth adding next to it.

Benchmarks influence perception but do not decide market share

Model benchmarks matter, but they do not equal market share. A model can top a leaderboard and still have limited distribution. A product can dominate usage with a model that is not ranked first on every test. Users choose products based on a mix of quality, speed, memory, interface, availability, trust, price, integrations and habit.

LMArena, formerly known as Chatbot Arena, became influential because it uses pairwise comparisons and crowdsourced human voting. The original Chatbot Arena paper described an open platform for evaluating LLMs by human preference, with users comparing anonymous model responses and voting on which is better. The authors said the platform had amassed more than 240,000 votes at the time of the paper and had become one of the most referenced LLM leaderboards.

That method is useful because it captures human preference rather than only static benchmark scores. It also has limits. Arena users are not a perfect sample of all users. Model providers may care deeply about leaderboard performance. Benchmarks can drift as models are tuned for them. A leaderboard win can signal capability, but it does not tell us whether a product has distribution, revenue or retention.

Artificial Analysis adds another view by tracking provider performance across endpoints, including output speed, latency, context window and price. These metrics matter because a model that is slightly smarter but slower or more expensive may lose in production.

Academic work also warns against treating benchmarks as market truth. A 2026 survey paper on how users evaluate AI chat assistants found that among active AI chat users, satisfaction with Claude, ChatGPT and DeepSeek was statistically indistinguishable, that more than 80% of surveyed users used two or more platforms, and that users were attracted to different products for different reasons. The sample was small and focused on active users, but it reinforces the point that usage behavior cannot be inferred from model rankings alone.

Benchmarks shape reputation. Distribution shapes market share. Workflow fit shapes revenue. The strongest AI company will need all three, but no single leaderboard can tell us who has won.

Geography fragments the race

AI model share varies by country, language, device mix, regulation, payment infrastructure and product access. A global average can hide sharp regional differences. ChatGPT, Gemini, Claude, Perplexity and Grok do not have equal reach in China, Russia, India, Brazil, the United States, the European Union or Southeast Asia.

a16z’s 2026 Top 100 Gen AI Consumer Apps report noted that Western AI tools such as ChatGPT, Claude, Gemini and Perplexity draw their top markets from a similar pool: the United States, India, Brazil, the United Kingdom and Indonesia, in varying order. The report also said none has meaningful usage in China or Russia.

This matters because local platforms fill gaps where Western AI assistants are unavailable, limited or less culturally fitted. In China, products from Baidu, Alibaba, Tencent, ByteDance and others operate under domestic rules and distribution channels. In Russia, access patterns, sanctions, local services and language preferences shift the competitive field. In India, price sensitivity, mobile-first usage and English-plus-local-language behavior shape adoption.

Google’s Gemini may have an advantage in Android-heavy countries, while ChatGPT benefits from global brand recognition. Claude may be stronger in professional English-language markets than in mass mobile markets. Perplexity may attract research-heavy users in countries where English web content is a large part of professional life. Grok may follow X’s regional user base and political culture.

Language quality is another hidden share variable. A model that performs well in English may not be equally trusted in Hindi, Portuguese, Arabic, Spanish, Indonesian, Japanese or Polish. Multilingual performance is not only translation. It includes local facts, cultural norms, names, institutions, idioms, legal references and current events.

The global AI market will not settle into one universal ranking. ChatGPT may lead globally, Gemini may gain where Google surfaces dominate, Claude may overperform in developer-heavy and enterprise-heavy markets, Perplexity may win research niches, and local AI systems may lead where Western tools lack access or compliance.

Android distribution gives Gemini a structural path

Android is one of Gemini’s strongest paths to adoption. A standalone app has to earn a download. A device-level assistant can become part of the phone. That difference matters because AI usage often begins with low-friction moments: voice commands, search boxes, camera tasks, screenshots, translation, email replies, calendar actions and photo interpretation.

a16z’s August 2025 Top Gen AI Consumer Apps report said Gemini ranked second behind ChatGPT on mobile by monthly active users, with nearly half as many MAUs, and that nearly 90% of Gemini’s mobile MAU base was on Android, compared with 60% for ChatGPT.

That Android skew gives Google a path no rival can fully copy. OpenAI can build a popular app on Android and iOS. Anthropic can do the same. Perplexity can build a browser and apps. Grok can use X. But only Google can weave Gemini into the operating system, Search, Play services, Pixel devices and Android partner experiences at that depth.

The constraint is user choice. Default placement can create trial, but it does not guarantee preference. Users may still install ChatGPT because friends mention it, because they already pay for it, because they prefer its responses, or because work requires it. Gemini must convert default exposure into repeated active use.

Google is addressing that through model quality, multimodal features, personal context and Workspace integration. Gemini’s growth from 400 million to more than 900 million monthly users in a year shows that it is not merely riding passive exposure. But usage depth remains the harder question. A monthly active user who tries Gemini once is not equivalent to a user who spends hours per week inside ChatGPT or Claude.

Android gives Gemini reach. Product satisfaction must turn that reach into share. If Google succeeds, Gemini can become the default AI layer for hundreds of millions of mobile users without needing to beat ChatGPT app-store by app-store.

Paid subscriptions reveal economic share

User count is only one side of market share. Paid subscriptions reveal who has converted attention into revenue. OpenAI’s public claim of more than 50 million consumer subscribers is a major economic signal. Even before enterprise revenue and APIs, that subscription base gives ChatGPT a paid consumer engine few AI products can match.

Paid share matters because AI usage has real marginal cost. Inference requires compute, energy, model serving, memory, retrieval and infrastructure. Unlike many social networks, where an extra user can be monetized through ads at low marginal cost, AI assistant usage can be expensive when users generate long responses, images, videos or code-agent loops. A free user who asks heavy questions can cost money. A paid user funds capacity.

Google has another path through bundled subscriptions. Alphabet’s Q4 2025 remarks said Google had more than 325 million paid subscriptions across consumer services and had sold more than eight million paid seats of Gemini Enterprise four months after launch. The same remarks said the Gemini app had over 750 million monthly active users by that point.

Anthropic’s paid base is less publicly quantified, but Claude’s pricing tiers, Claude Code demand and enterprise seats point to a high-intent revenue model. Ramp’s business-spend data supports the idea that Anthropic is gaining paid business adoption even without ChatGPT’s consumer reach.

Perplexity’s paid model is tied to Pro, Max, Enterprise and API use. Its official page lists Perplexity Pro at $20 per month or $200 per year, with Pro Search, Spaces, file uploads, image generation and higher limits.

Grok’s paid path runs through xAI products, API usage and possible X-linked subscriptions. Its challenge is to convert attention and media features into durable paid usage, not only viral use.

The economic race may diverge from the user race. ChatGPT leads paid consumer AI. Google can bundle AI into huge subscription and enterprise bases. Claude may win high-value professional seats. Perplexity can monetize research-heavy users. Grok can monetize through xAI, X and media-heavy AI features. The largest audience will not automatically capture the highest profit.

Inference costs pressure every provider

AI market share is expensive. Every share gain has a compute bill behind it. When models answer longer, reason more, use tools, create images, generate video or run code agents, costs rise. The economics of the market are not only about attracting users. They are about serving them without burning unsustainable amounts of capital.

This cost pressure explains why subscriptions, rate limits, API pricing, model routing and cheaper models matter. OpenAI, Google, Anthropic and xAI all need enough capacity to serve demand while keeping latency acceptable. When a product becomes popular, usage spikes can reduce quality or force stricter limits. Anthropic’s May 2026 announcement raising Claude Code and API limits shows demand pressure directly.

The cost issue also helps explain the rise of smaller and faster models. Not every prompt needs the most expensive frontier model. Many tasks can be routed to cheaper models with acceptable quality: classification, summarization, extraction, simple Q&A, formatting, translation, short drafting or internal workflow steps. The winner in enterprise may not be the company with the single smartest model, but the one that delivers the right mix of quality, cost and control.

Artificial Analysis tracks output speed, latency and price because production developers care about these tradeoffs. A model can be brilliant but too slow for a customer-service agent, too expensive for batch processing, or too unpredictable for a regulated workflow.

OpenRouter’s 100-trillion-token study also points to the rise of agentic inference. Agentic systems can consume many more tokens than single-turn chat because the model plans, acts, observes, revises and uses tools. This makes the unit economics of coding agents, browsing agents and workflow agents more demanding than ordinary chat.

Market share in AI is not free scale. Each gained user can bring cost. Each gained enterprise workload can bring heavy token use. The strongest providers will need not only users, but pricing power, infrastructure, model efficiency and routing discipline.

Enterprise buyers will resist single-provider lock-in

The early consumer AI market rewarded defaults. Enterprise AI will reward control. Companies do not want their entire knowledge-work stack dependent on one model provider if prices rise, quality drops, policies change or outages occur. This creates space for multi-model architectures even when one provider leads.

A business may use ChatGPT Enterprise for general employee access, Claude for coding and document work, Gemini through Workspace, Perplexity for research, open-weight models for sensitive internal workloads and a router to manage cost. That arrangement weakens clean market-share boundaries. It also makes procurement more complex.

Anthropic’s Claude Code and OpenAI’s Codex-style tools both create potential lock-in because they connect to codebases and workflows. Google’s Workspace integration creates lock-in through documents, meetings and email. Perplexity’s Spaces and Enterprise features create lock-in around research collections. Grok’s tie to X creates lock-in around social context. The more useful these tools become, the more buyers will ask how portable their workflows are.

The model context protocol, multi-model routers, cloud marketplaces and internal abstraction layers are all responses to this risk. Enterprises want the ability to switch models without rebuilding every application. They also want audit logs, data controls, role permissions and cost caps.

Ramp’s business-spend data suggests companies are already making provider-level choices rather than blindly adopting the best-known name. Anthropic’s gains among businesses in Ramp’s sample show that enterprise share can move when buyers believe a tool fits their workflows better.

The enterprise market will be less winner-take-most than the consumer market. A company can standardize on one assistant for employees while using several models behind the scenes. This means consumer market share may overstate future enterprise dominance.

Model routers make provider share harder to see

Model routers and inference platforms change how AI share is allocated. Instead of a developer choosing “OpenAI” or “Anthropic” for a whole product, a router can send each task to a different model. The user may never know which model answered. The application may switch providers based on price, latency, availability, context length or capability.

OpenRouter is the best-known example in public data. Its State of AI report uses more than 100 trillion tokens from its platform to study real-world LLM usage across tasks and model types. The report highlights open-weight adoption and agentic inference, showing a market where many users already think in terms of model choice rather than provider loyalty.

This routing layer complicates market share in two ways. First, it creates a separate market for aggregators and infrastructure providers. Second, it hides the provider from the end user. A consumer might use an app powered by Claude for one task, Gemini for another and an open-weight model for a third without seeing the model name.

Perplexity’s own product description says it routes queries across multiple frontier models so each question is handled by the model best suited for it. That means Perplexity competes as an AI answer product while also using other model providers under the hood.

The rise of routers favors task-level competition. If Claude is best for code, it gets code traffic. If Gemini is cheaper for long-context tasks, it gets long-context traffic. If GPT is strongest for general chat, it gets chat traffic. If an open-weight model is good enough and far cheaper, it gets cost-sensitive traffic. Share becomes fluid.

In a routed AI market, the brand on the screen and the model doing the work may be different. That makes public market-share charts less complete and makes developer ecosystems more powerful.

Brands need to treat AI assistants as discovery channels

For businesses, the market-share race is not an abstract technology story. AI assistants are becoming discovery channels. People ask them which tools to buy, which vendors to compare, which articles to trust, which restaurants to book, which software fits a need and which claims are true. The assistants answer by drawing on training data, web retrieval, structured sources, user context and partner integrations.

ChatGPT’s huge user base makes it a consumer discovery surface. Gemini’s integration into Search and Google products makes it a search-discovery surface. Perplexity’s cited answers make it a source-discovery surface. Claude’s professional use makes it a work-decision surface. Grok’s X tie makes it a social-discussion surface.

This changes SEO and digital strategy. A brand needs crawlable, current, authoritative pages. It needs clear entity information, pricing pages, documentation, comparison pages, reviews, structured data, media coverage and trustworthy third-party mentions. It also needs consistency across its site and public profiles. AI systems are less forgiving of vague positioning because they synthesize from many sources.

Similarweb’s ranking of AI assistant sites and Statcounter’s AI chatbot chart both point to a fragmented discovery environment. ChatGPT dominates some layers, Gemini rises in web visits, Perplexity appears strong in chatbot share, and Claude is gaining professional attention.

The practical implication is clear: brands should not write only for Google’s classic blue links. They need to be understandable to AI systems that summarize, compare and cite. That means stronger source pages, better factual consistency, clearer authorship, updated product data and fewer thin marketing claims.

AI visibility will not replace SEO; it will sit on top of it. Search engines, answer engines and assistants all need reliable source material. The brands that publish precise, current and citeable content will have an advantage across ChatGPT, Gemini, Perplexity, Claude and future AI interfaces.

Regulation and copyright pressure can reshape share

AI market share is not decided only by users and models. Regulation, copyright disputes, safety rules, privacy enforcement and antitrust cases can shift distribution. The companies with the biggest reach will face the most scrutiny. The companies built around web retrieval will face pressure from publishers. The companies tied to social platforms will face moderation and safety questions.

Google’s AI position sits inside active antitrust scrutiny over search and browser distribution. Any remedy that changes Chrome, Search defaults or Android bundling could affect Gemini’s future reach. OpenAI faces scrutiny over data, copyright, safety, competition and its corporate structure. Anthropic faces questions about frontier-model safety and enterprise deployment. Perplexity faces criticism from publishers around content use and crawling. xAI faces scrutiny tied to Grok’s content generation and X distribution.

Perplexity’s position is especially exposed to publisher relations because citations and answer extraction are central to the product. If publishers restrict access, demand licensing, or seek legal remedies, answer engines may need new commercial arrangements. If they build partnerships, Perplexity can become a more formal distribution channel.

Grok’s risks are different. Its public social role makes misuse visible and politically charged. Image and video generation raise legal concerns around deepfakes, non-consensual content and defamation. Restrictions may reduce misuse but also change product appeal.

Regulation can also help incumbents. Large companies can afford compliance teams, audits, policy work and enterprise certifications. Smaller competitors may struggle with fragmented rules. Google, Microsoft, OpenAI and Anthropic are better placed than most startups to meet enterprise and regulatory demands, though size also makes them larger targets.

Trust, rights and compliance are market-share variables. A model that gains users through aggressive capabilities can lose business adoption if customers fear legal or reputational risk. A cautious product can lose consumer excitement but gain enterprise trust. The balance will differ by market.

Safety and factual reliability affect retention

Users will tolerate occasional mistakes in casual chat. They are less forgiving when AI systems answer health questions, financial questions, legal questions, news questions or business questions with confidence and errors. Factual reliability is now part of market share because it affects whether users trust a system enough to return for serious work.

OpenAI’s usage paper found that “Practical Guidance,” “Seeking Information” and “Writing” are the top ChatGPT conversation categories, together accounting for nearly 80% of conversations. That means many users rely on AI for advice, information and drafted output. If quality falls, market share is exposed.

A 2026 paper evaluating commercial AI chatbots as news intermediaries tested six chatbots on same-day BBC News questions across languages and regions. It found that top systems could exceed 90% multiple-choice accuracy on questions about events reported hours earlier, but lost accuracy under free-response evaluation and across the broader cohort. The paper is a reminder that retrieval-synthesis systems can look strong under one test and weaker under another.

Perplexity tries to address this with citations. Claude’s brand leans on carefulness and long-form reasoning. ChatGPT benefits from broad usage and product polish but must manage hallucination risk at massive scale. Gemini’s challenge is quality across many Google surfaces and languages. Grok’s challenge is reliability inside fast-moving social contexts.

Safety also includes model behavior, not only facts. Users want systems that do not fabricate sources, expose private data, produce harmful content, or take actions without consent. Enterprises want auditability and control. Regulators want accountability.

The market will reward models that users can trust with higher-stakes tasks. Entertainment use can drive growth, but durable economic share comes from work that users and companies are willing to delegate.

The competitor map is becoming clearer

The leading AI products now have distinct identities. They still compete directly, but their strongest positions are different enough that the market is unlikely to collapse into one uniform winner.

ChatGPT is the mass-market default. It has the strongest consumer brand, the largest public user base, a large subscription business, and a broad product roadmap. Its risk is losing specialized high-value sessions to rivals while trying to turn ChatGPT into a platform without making it too crowded.

Gemini is the distribution challenger. It has Google’s ecosystem, fast app growth, Search and Android reach, Workspace and Cloud paths, and deep infrastructure. Its risk is that default exposure does not always translate into active preference.

Claude is the professional-depth challenger. It is smaller by user count but strong in coding, writing, long-form work and enterprise adoption signals. Its risk is capacity, pricing and narrower mass-market reach.

Perplexity is the answer-engine specialist. It has a clear cited-research identity and search-intent strength. Its risk is scale, publisher friction and competition from ChatGPT Search and Google AI Mode.

Grok is the social-attention challenger. It has X distribution, fast product iteration, media features and a distinct persona. Its risk is trust, safety, moderation and enterprise reluctance.

Microsoft Copilot, DeepSeek, Meta AI, Mistral, Qwen, Kimi and open-weight models also matter, even though the user’s prompt focused on Claude, ChatGPT, Gemini, Grok and Perplexity. Microsoft has enterprise distribution. DeepSeek and other Chinese models affect global price and capability expectations. Meta AI has massive potential distribution across WhatsApp, Instagram and Facebook. Open-weight models pressure pricing and give enterprises more control.

The AI model market is becoming a system of defaults and specialists. ChatGPT is the default for many consumers. Gemini is the default candidate inside Google surfaces. Claude is a specialist that is becoming a default for some professionals. Perplexity is a specialist for cited answers. Grok is a specialist for social AI and media-rich interaction.

Strategic position of the main AI assistant brands

The market is not flattening into a single assistant category. Each brand is trying to own a different route into user behavior: habit, default distribution, professional trust, cited research, social context or enterprise bundling.

ChatGPT’s lead is strongest at the top of the funnel

ChatGPT’s strongest asset is that users start there. For consumer AI, that is powerful. The first assistant a person uses becomes the place where they learn prompt habits, store memories, subscribe, upload files and build confidence. That top-of-funnel lead helps OpenAI launch new features faster because it has a huge audience ready to try them.

OpenAI’s scale claims are unmatched in public first-party data: more than 900 million weekly active users and more than 50 million consumer subscribers. The Reuters-Sensor Tower milestone of 1 billion monthly active app users reinforces that ChatGPT is the mass-market leader even as rivals grow.

The weakness of a top-of-funnel lead is that it can be shallow. Users may open ChatGPT first but move elsewhere for specific tasks. The AI market is full of task-specific dissatisfaction: one model writes better, another codes better, another cites better, another integrates with work data, another has cheaper API pricing, another handles context better.

OpenAI’s response is to broaden ChatGPT until fewer sessions leak out. That means better search, stronger coding, more agents, deeper enterprise features, app integrations and multimodal tools. The company wants ChatGPT to be the place where users not only ask, but act.

The challenge is focus. ChatGPT must serve casual users, students, paid power users, developers, enterprises, governments and creators without becoming too complex for any of them. A mass-market default has to feel simple. A professional platform has to feel powerful. Balancing those demands is difficult.

ChatGPT’s lead is real, but it is no longer enough to win every layer. It can remain the biggest AI product while losing specific high-value workflows. The difference between those two outcomes will shape OpenAI’s revenue mix and strategic urgency.

Gemini’s growth is strongest where Google can reduce friction

Gemini’s market gains come from lowering the friction of AI adoption. Users already have Google accounts. They already search. They already use Android, Gmail, Docs, Meet, Chrome or YouTube. Each of those surfaces gives Gemini a chance to appear where work or curiosity already begins.

Google’s user milestones show a steep curve: 400 million monthly users at I/O 2025, more than 450 million by Q2 2025, more than 750 million by Q4 2025 and more than 900 million by I/O 2026.

This growth is not only from chatbot curiosity. Google has tied Gemini to practical surfaces: writing, meetings, photos, browser context, search, personal assistance and enterprise workflows. The more Gemini can answer inside the place where the user already is, the less it has to win a separate app-opening habit.

For publishers and brands, Gemini’s path is especially important because it intersects with Google Search. AI Overviews and AI Mode can change how users discover information before they ever reach a website. If Gemini becomes more agentic inside Search, it may influence not only answers but actions: booking, buying, comparing, organizing and completing tasks.

The competitive risk for Google is user perception. Google has to prove its AI is not merely inserted into products but genuinely preferred. Users who feel Gemini is forced into Search may resist it. Users who find it useful in Gmail or Docs may adopt it without thinking. The line between convenience and intrusion matters.

Gemini’s share will grow fastest where AI feels like a natural extension of Google behavior. Search, Android and Workspace are the central battlegrounds.

Claude’s growth is strongest where quality is felt over long sessions

Claude’s market strength is easiest to understand when watching long sessions. A user who asks one short factual question may not notice much difference between frontier assistants. A user who spends three hours drafting a strategy memo, refactoring code, analyzing a PDF set or debugging a system will notice differences in context handling, tone, patience and error recovery.

This is why Claude’s smaller user base can have an outsized reputation. Among power users, quality differences are felt through sustained work. Claude’s brand has become linked to writing feel, long-context collaboration and coding workflows. Those are not always captured in app MAU charts.

Anthropic’s Economic Index shows that Claude use is spreading across many tasks and that coding tasks are moving into automated API workflows. The top 10 Claude.ai tasks accounted for 19% of traffic in February 2026, down from 24% in November 2025, suggesting broader use rather than dependence on only a few tasks.

Claude Code deepens this pattern. Once a coding agent is allowed to operate inside a repo, the user is not merely chatting. They are letting the model interact with production-adjacent work. That raises both value and trust requirements. A tool that performs well there can win loyalty quickly.

Ramp’s business-spend signal reinforces Claude’s professional momentum. Nearly one in four businesses in Ramp’s data were using Anthropic in February 2026, and Anthropic adoption grew quickly month over month.

Claude’s market share is likely understated by broad consumer rankings and overstated by AI insider enthusiasm. The balanced view is that Claude is not close to ChatGPT’s mass reach, but it is one of the strongest challengers in the workflows that matter most for paid professional adoption.

Perplexity’s future depends on answer trust and distribution

Perplexity’s product is clear. Its distribution challenge is harder. To become a daily habit, it must convince users that cited AI answers are better than a Google search, a ChatGPT search response, a Gemini answer or a direct source visit.

Its official positioning as an answer engine gives it a clean message. Users know why they are there: to ask a question and see sources. That clarity is rare in AI, where many products blur into general assistance.

The challenge is that the giants can copy parts of the experience. ChatGPT can cite web sources. Gemini can cite search results. Google can place AI answers above classic links. Microsoft can integrate AI into Bing and Copilot. Perplexity must remain faster, clearer, more trusted or more specialized.

Its IPO timeline, reported by Reuters as targeting 2028, suggests the company is not positioning itself as a quick feature company. It wants to be a durable business.

For SEO and media, Perplexity is already influential because it trains users to expect answers with citations. Even if the platform remains smaller than ChatGPT, it pushes the market toward source-visible AI. That pressure affects Google and OpenAI as much as Perplexity.

Perplexity’s path is not to become the biggest chatbot. Its path is to make cited AI search feel safer and faster than ordinary search. If it owns that mental category, it can remain important in a market led by larger platforms.

Grok’s future depends on turning attention into trust

Grok has attention. It has a distribution channel through X. It has a brand personality that stands apart from more cautious assistants. It has multimodal ambitions through xAI. The question is whether attention becomes trust.

The social role is powerful because it creates public usage. People see Grok replies, screenshots and generated media. They encounter the product even before choosing to use it. That kind of viral exposure can build awareness faster than paid advertising.

But trust in social AI is fragile. If a model is used as a public arbiter, mistakes are public. If it generates harmful images or misleading claims, backlash is public. If moderation feels inconsistent, critics can document it in real time. Grok’s link to X makes it visible and exposed.

Academic work on Grok in the wild shows both the novelty and complexity of public social AI. Grok is not only answering private user prompts; it is mediating disputes and participating in public information flows.

For enterprise adoption, xAI will need to separate Grok’s consumer persona from reliability and compliance expectations. A company may enjoy Grok’s speed and capabilities but hesitate if the brand feels unpredictable. xAI’s API and enterprise tooling will have to prove that Grok can operate under business controls.

Grok’s market share may grow quickly in attention-heavy consumer contexts and more slowly in trust-heavy enterprise contexts. That split could define its 2026 and 2027 trajectory.

AI search will force new source economics

As AI assistants answer more questions directly, source economics become a market-share issue. If users get answers without clicking, publishers lose traffic. If AI systems lack reliable sources, answer quality falls. The market needs source material, but the payment and attribution models remain unsettled.

Perplexity puts the issue on the surface because citations are part of the product. ChatGPT, Gemini and Copilot also use retrieval and browsing. Google’s AI Overviews and AI Mode sit directly on top of the search economy. Claude’s web search features bring source questions into professional workflows. Grok’s real-time answers draw from web and X context.

For brands, the answer is not to hide content from AI by default. Some publishers may choose legal or technical restrictions, but most commercial brands need AI visibility. If AI assistants become recommendation engines, invisibility is costly. The better strategy is to publish source material that is accurate, structured, current and worth citing.

The commercial tension will be sharpest for media, reference sites, reviews and specialized publishers. Their content may power answers that reduce visits. Licensing deals, traffic-sharing arrangements, publisher controls and citation standards will become part of AI competition.

A search engine that respects sources may win trust. An answer engine that is seen as extracting value without fair exchange may face legal and reputational pressure. This can influence market share because users, regulators and partners react to perceived fairness.

Source relationships are becoming part of AI product quality. The best answer engine is not only the model with the best synthesis. It is the system with the best access to current, trusted and legally durable information.

AI model share will affect advertising and commerce

AI assistants are moving closer to transactions. Users ask for products, prices, trips, restaurants, software tools, financial explanations and shopping advice. Once assistants recommend, rank, compare or book, market share becomes commerce power.

Google has the clearest existing commerce infrastructure through Search, Shopping, Ads and merchant data. Gemini can sit inside that network. OpenAI is building ChatGPT into a place where users can discover apps, research purchases and complete tasks. Perplexity has explored shopping and answer-based commerce. Grok can connect recommendations to social attention on X. Claude may influence business procurement and professional tool choices through research and document work.

This will change advertising. Classic search ads target keywords. AI commerce may target intent inside conversational sessions. The assistant can ask follow-up questions, compare options, summarize reviews and reduce the number of brands shown. Being one of three recommended options in an AI answer may become more valuable than ranking eighth in search.

The risk is opacity. If users do not know why an assistant recommended a product, or whether paid placement influenced it, trust can suffer. Regulators may push disclosure rules. Brands will demand visibility into how AI systems rank and cite them. Platforms will seek monetization without damaging user trust.

AI market share is becoming distribution share for commerce. ChatGPT’s consumer reach, Google’s shopping infrastructure, Perplexity’s research intent, Claude’s business trust and Grok’s social influence all point to different commerce paths.

For companies, the practical move is to build content and product data that AI systems can use: clear pricing, comparison pages, documentation, availability, reviews, support content and structured entity information. Thin brand slogans will not survive synthesis.

Open-weight models put price pressure on the leaders

The prompt focuses on Claude, ChatGPT, Gemini, Grok and Perplexity, but open-weight models matter because they pressure price and control. Meta’s Llama, Mistral, Qwen, DeepSeek and other open or semi-open models give developers alternatives when closed-model APIs are too expensive, restricted or opaque.

OpenRouter’s 100-trillion-token study found substantial adoption of open-weight models on its platform, alongside closed APIs. That does not mean open-weight models dominate the whole market, but it shows that many developers already use them when cost, customization or control matters.

Open-weight models are especially relevant in enterprise and developer markets. A company may prefer a closed frontier model for the hardest tasks, but use an open model for internal summarization, classification, routing, extraction or privacy-sensitive workloads. This reduces the total addressable share available to any one closed provider.

Open models also change pricing expectations. If a cheaper model performs “good enough” for 80% of tasks, premium models must justify their cost on the hardest 20%. That pushes OpenAI, Anthropic, Google and xAI to offer faster, cheaper tiers alongside their flagship models.

The consumer market is different. Ordinary users rarely choose a model because it is open-weight. They choose an app. Open-weight adoption reaches them indirectly through products built on top of open models.

Open-weight competition will be more visible in API and enterprise economics than in consumer assistant rankings. It may not knock ChatGPT out of first place, but it can reduce margins and push customers toward multi-model architectures.

Market share will be measured by tasks, not only users

The next stage of AI competition will be task-based. Which model writes the best first draft? Which handles legal research with sources? Which edits code safely? Which can use a browser? Which can plan travel? Which can analyze spreadsheets? Which can summarize meetings? Which can remember preferences? Which can work inside Gmail, Slack, GitHub, Notion, Salesforce or Excel?

A user-level chart cannot answer these questions. A person may use ChatGPT for general questions, Claude for writing and code, Gemini for Gmail and Android, Perplexity for citations and Grok for X. That is one user spread across five products. The market share of that user depends on task, time and payment.

This shift favors companies with clear task ownership. Claude wants coding and deep work. Perplexity wants cited research. Gemini wants Google-connected tasks. ChatGPT wants the broad default and more platform actions. Grok wants social and media-rich AI.

For enterprise buyers, task mapping becomes procurement strategy. They will not ask only “Which model is best?” They will ask which model is best for each workflow and what the fallback model should be if cost, latency or quality changes.

For analysts, this means market share should be reported in layers: consumer reach, professional depth, API tokens, enterprise spend, developer adoption, AI search referrals, task category and revenue. A single pie chart is no longer enough.

The most honest answer to “Who leads AI model market share?” is layered: ChatGPT leads mass consumer usage, Gemini is the fastest large-platform challenger, Claude is strongest in high-intent professional growth signals, Perplexity is strongest as a cited answer specialist, and Grok is strongest where AI meets social attention.

The 2026 market is plural but unequal

The AI assistant market is not winner-take-all, but it is not evenly distributed. ChatGPT is far ahead in consumer scale. Gemini is large enough to be the main broad challenger. Claude is smaller but rising in the workflows that influence professional spending. Perplexity is smaller still but owns a distinct answer-engine identity. Grok has attention, social distribution and fast iteration, but trust questions remain.

This creates a plural but unequal market. Many users will try multiple assistants. Many businesses will buy several providers. Many developers will route tasks across models. But most consumers will keep one default, and ChatGPT remains that default today.

The most likely near-term pattern is not a sudden dethroning of ChatGPT. It is share leakage at the edges. Gemini takes more Google-native tasks. Claude takes more coding and serious writing. Perplexity takes more cited research. Grok takes more X-native and media-rich sessions. Open-weight models take more cost-sensitive API workloads. Copilot keeps enterprise Microsoft distribution.

For OpenAI, the danger is not losing the lead overnight. It is having the highest-volume product while rivals capture the highest-intent tasks. For Google, the challenge is converting distribution into active trust. For Anthropic, the challenge is scaling without losing the qualities that made Claude attractive. For Perplexity, the challenge is turning answer trust into a large business. For xAI, the challenge is turning Grok’s attention into reliable utility.

Market share in AI is becoming less like search share and more like cloud share: one market, many workloads, several strong providers, and constant switching pressure. The largest player can stay largest while customers still demand alternatives.

The practical reading for companies and publishers

Companies should stop asking only which AI model has the largest share. The better question is where their customers, employees and buyers are asking questions. A consumer brand may care most about ChatGPT, Gemini and Google AI Overviews. A B2B software company may care about Perplexity, ChatGPT, Gemini and Claude. A developer-tool company may care heavily about Claude, ChatGPT, OpenRouter, GitHub, Stack Overflow and technical docs. A media company may care about Perplexity, Google AI Mode, ChatGPT Search and citation patterns.

The practical response has four parts. First, strengthen factual source pages. Second, make product and company entities clear. Third, publish comparison and use-case content that answers real buyer questions. Fourth, monitor AI answers directly across ChatGPT, Gemini, Claude, Perplexity and Grok.

AI systems reward clarity. They need to know what a company is, what it sells, who it serves, how it differs, what evidence supports its claims and which sources confirm it. Vague marketing pages are weak source material. Current documentation, transparent pricing, original research, expert authorship and credible third-party mentions are stronger.

Publishers face a harder task. They need visibility inside AI answers while defending the value of their work. That means better metadata, clear authorship, licensing strategy, content freshness, source reputation and direct audience relationships.

The AI market-share race is also a visibility race. The products with the largest or most trusted answer surfaces will shape what users believe, compare and buy. Brands that wait for a stable winner will miss the period when AI systems are learning which sources deserve trust.

A realistic outlook for ChatGPT, Gemini, Claude, Grok and Perplexity

ChatGPT is likely to remain the largest mass-market AI assistant through the near term. Its user base, paid subscribers, brand habit and product expansion are too strong for a quick reversal. The real question is whether it can keep high-value sessions inside ChatGPT as rivals specialize.

Gemini is likely to keep gaining, especially through Google Search, Android, Workspace and Chrome. If Gemini becomes a true assistant layer across Google products rather than a feature users tolerate, it can narrow the gap with ChatGPT in daily utility.

Claude is likely to remain the strongest high-intent challenger. It may not become a mass-market default at ChatGPT scale, but it can win paid professionals, developers and enterprise teams. That may produce economic share beyond its consumer reach.

Perplexity is likely to stay smaller but influential. Its future depends on whether cited AI search becomes a durable user habit and whether it can manage publisher relationships while scaling distribution.

Grok is likely to grow where X, media generation, real-time social context and personality matter. Its ceiling depends on trust. If xAI can make Grok reliable enough for broader work while preserving its distinct identity, it can become more than a social AI feature.

The broad market will not settle into a single winner. The next AI market-share battle will be decided by defaults, task ownership, paid conversion, trust and infrastructure cost. ChatGPT leads the scoreboard most people see. Gemini and Claude are changing the scoreboard itself.

Questions readers ask about AI model market share

Author:
Jan Bielik
CEO & Founder of Webiano Digital & Marketing Agency

This article is an original analysis supported by the sources cited below

OpenAI Scaling AI for everyone
Official OpenAI page stating ChatGPT’s weekly active users and consumer subscriber count.

Reuters ChatGPT app hits 1 billion monthly active users in record time
Reuters report citing Sensor Tower estimates for ChatGPT and Claude app usage in 2026.

Google The Gemini app becomes more agentic
Official Google post stating that more than 900 million people use Gemini monthly.

Alphabet investor presentation June 2026
Alphabet investor presentation discussing Gemini monthly users, Personal Intelligence and Google’s agent strategy.

Google I/O 2025 from research to reality
Official Google I/O 2025 keynote recap with Gemini app monthly active user figures and token-processing context.

Alphabet 2025 Q2 earnings call
Alphabet investor-relations transcript with Gemini monthly active users and AI token-processing updates.

Google Q4 2025 earnings remarks
Google CEO remarks covering Gemini app monthly active users, subscriptions and Gemini Enterprise seats.

Similarweb top AI Chatbots and Tools websites
Similarweb ranking of the most visited AI chatbot and AI tool websites worldwide.

Statcounter AI chatbot market share worldwide
Statcounter’s public AI chatbot percentage market-share chart for worldwide usage.

Momentic top generative AI chatbots and LLMs by market share
Similarweb-based analysis of web-visit share across major AI chatbot domains.

Anthropic Economic Index March 2026 report
Anthropic research report analyzing Claude usage patterns and task distribution in February 2026.

Anthropic Claude Code product page
Official Anthropic page describing Claude Code as an agentic coding system.

Anthropic higher usage limits for Claude
Anthropic announcement on higher Claude Code and API usage limits in May 2026.

Ramp AI Index March 2026 update
Ramp Economics Lab analysis of business AI adoption, including Anthropic, OpenAI, Google and xAI usage among Ramp customers.

Perplexity AI for the Curious
Official Perplexity page describing its AI answer engine, cited answers and multi-model routing.

Reuters Perplexity planning 2028 IPO
Reuters report on Perplexity’s stated 2028 IPO plan and its position in the AI industry.

xAI official site
Official xAI site describing Grok-related frontier AI models and API access.

Andreessen Horowitz Top 100 Gen AI Consumer Apps 6th edition
a16z analysis of leading generative AI consumer applications and geographic usage patterns.

Andreessen Horowitz State of Consumer AI 2025
a16z consumer AI analysis discussing ChatGPT, Gemini, Claude, Perplexity, Grok and subscription overlap.

OpenRouter State of AI 2025
OpenRouter and a16z report based on a 100-trillion-token study of LLM usage through the OpenRouter platform.

Artificial Analysis LLM API Providers Leaderboard
Provider comparison resource tracking latency, speed, pricing and other API endpoint performance metrics.

Chatbot Arena paper
Academic paper describing the crowdsourced pairwise model-evaluation method behind Chatbot Arena.

OpenAI How people are using ChatGPT
OpenAI and Harvard-linked research paper analyzing ChatGPT usage growth, message categories and consumer behavior.

Grok in the Wild
Academic study examining Grok interactions on X and the social roles played by AI in public conversations.

Evaluating commercial AI chatbots as news intermediaries
Academic paper evaluating commercial AI chatbots on same-day news questions across languages and regions.