John Mueller’s August 20 comment, reported on August 25, reinforces Google’s formal guidance: AI Overviews and AI Mode still depend on the Search index, ranking systems and ordinary crawlability, not llms.txt or bespoke GEO markup. The strategic shift is real, but it sits in content selection, measurement and user behavior—not in a parallel technical protocol for Google Search.
Table of Contents
On August 20, Google Search advocate John Mueller wrote on Bluesky that, from Google’s point of view, there was “nothing really special” site owners needed to do for generative AI responses in Search. Search Engine Roundtable surfaced the comment on August 25. The timing matters because the remark was not a new policy announcement; it was a fresh restatement of guidance Google had already formalized in May and clarified again in June.
That guidance is unusually explicit for a fast-moving corner of search. Google says its generative features are rooted in core Search ranking and quality systems, retrieve material from the existing Search index, and do not require separate AI files, special schema, artificial “chunking” or AI-only rewrites. For Google Search today, there is no separate technical eligibility layer called GEO.
The important word, however, is “today.” AI Overviews and AI Mode materially change how queries are expanded, sources are selected, answers are assembled and traffic is distributed. Google is also developing a distinct technical agenda for agents that interact with websites rather than merely retrieve pages for Search. The defensible strategy is therefore neither “GEO is fake” nor “SEO is obsolete.” It is narrower: keep technical investment anchored in Search fundamentals, move more resources toward original evidence and measurable visibility, and reserve a separate engineering layer for agentic use cases only when the product actually requires one.
Google has turned a recurring answer into policy
Google’s position has moved beyond scattered comments from individual employees. On May 15, 2026, Google Search Central published a dedicated guide for generative Search and said conventional SEO remains relevant because AI Overviews and AI Mode are rooted in core ranking and quality systems. The accompanying Search Central blog described the document as a resource for understanding both what to optimize and what site owners can ignore. On June 15, Google’s documentation changelog recorded an additional clarification specifically about llms.txt.
The guide’s mythbusting section is direct. Google says Search does not use llms.txt or other new machine-readable AI files as a special mechanism for generative visibility. It says there is no requirement to break pages into tiny chunks, no need to rewrite copy in an AI-specific style, and no need for special Schema.org vocabulary beyond structured data that already serves Search features and accurately reflects visible content. Google also warns against pursuing inauthentic mentions merely to manufacture signals.
That creates a useful distinction between no separate technical eligibility layer and no optimization work at all. Google still requires a page to be indexable and eligible to appear with a snippet before it can serve as a supporting link in AI Overviews or AI Mode. Its minimum Search requirements remain familiar: Googlebot must not be blocked, the page must return a successful response, and it must contain indexable content. Eligibility is necessary, not a guarantee of crawling, indexing or serving.
For publishers and brands, the consequence is procurement discipline. A vendor can legitimately sell research, technical SEO, content development, analytics or experimentation for AI-era discovery. The weaker claim is that Google has created a hidden second protocol requiring a parallel stack of files and markup. Its documentation does not support that claim.
The retrieval layer changed more than the eligibility rules
The absence of new technical requirements does not mean Google’s generative results behave like ten blue links with a summary pasted on top. Google describes two mechanisms that make the retrieval process materially different. Retrieval-augmented generation uses Search systems to retrieve current web pages that can ground a generated response, while query fan-out lets the model issue multiple related searches to gather information across subtopics before composing the answer.
That architecture changes the competitive surface. A classic query can be treated as one visible expression of a larger information need. The model may seek definitions, comparisons, constraints, evidence, local details or product facts that the user never typed. Google says the fan-out process can identify additional supporting pages and produce a wider, more diverse set of links than a classic search. That is a retrieval difference, not proof of a new markup requirement.
Independent data points in the same direction, while remaining observational rather than a disclosure of Google’s ranking systems. Ahrefs analyzed 863,000 keyword result pages and roughly four million AI Overview URLs for a March 2026 study. It reported that about 38% of cited URLs also appeared in the first ten result blocks for the same original query, down from about 76% in its earlier 2025 analysis. Ahrefs interpreted the change as consistent with greater use of fan-out queries, though its dataset cannot establish Google’s internal cause.
The strategic implication is not that rankings no longer matter. Google says the system still relies on core Search ranking to retrieve pages. Yet a page can become useful because it answers one subproblem inside a wider task, even if it does not dominate the original query. The unit of competition is increasingly the information need behind the query, not only the literal keyword. That rewards topical competence without creating a reason to invent AI-only syntax.
Query fan-out rewards coverage without creating magic markup
Query fan-out is where much of the GEO industry’s legitimate work and dubious packaging can become confused. If an AI system decomposes a question into related searches, then mapping likely subquestions, filling evidence gaps and making pages unambiguous can improve the chance that useful material exists for retrieval. None of that requires pretending that the model needs paragraphs chopped into arbitrary lengths or headings written to a secret prompt template. Google explicitly rejects a required chunking formula.
The better interpretation is editorial. A useful page should resolve the main task and adjacent questions that determine a decision. A comparison may need specifications and trade-offs; a policy page may need dates and exceptions; a technical tutorial may need prerequisites and failure modes. The objective is not to simulate machine chunks. It is to publish distinct, reliable information that Search can retrieve when fan-out reaches a subtopic.
Google’s people-first content guidance has long asked whether a page contains original information, reporting, research or analysis and whether it adds substantial value instead of simply rewriting other sources. Its 2026 AI guide sharpens that logic by telling site owners to produce “non-commodity” content that a generative system could not easily reproduce from the common pool.
This is where ranking for the original query is not the whole retrieval problem becomes a useful operating principle. It does not imply that every page needs exhaustive coverage. It means content teams should understand the decision graph around important topics, then decide which questions deserve first-hand evidence, a specialist answer, proprietary data, a tool, a comparison or a clearer canonical page. Fan-out raises the value of information architecture and subject depth; it does not validate technical folklore.
For large organizations, that shift requires coordination between SEO teams, subject experts, data owners and editors. The work may look new on a budget sheet, but its durable components are research, publishing quality and technical accessibility rather than a parallel markup regime.
GEO’s weakest product is the technical shortcut
The appeal of llms.txt helps explain why the market keeps inventing technical rituals. It is concrete, inexpensive and easy to audit: create a file, put it at the site root, and tell clients they are “AI ready.” Google’s Search documentation undercuts that proposition for its own products. It says Google Search does not use special AI text files to determine generative visibility and that maintaining such a file for another service neither improves nor harms Google Search visibility. Google’s June documentation update was added specifically to clarify that point.
There is independent evidence that adoption has run ahead of observed use. Ahrefs examined server logs and traffic across 137,210 domains using its analytics products in May 2026. It found that 28% published a valid llms.txt file, yet 97% of those files received no requests at all during the month. Ahrefs cautioned that its customer base skews technical and SEO-aware, so the adoption rate should not be generalized to the whole web. The crawl finding is still a strong warning against claiming a demonstrated visibility benefit.
Chrome’s Lighthouse documentation creates a wrinkle: it includes an agentic-browsing audit for llms.txt and describes the file as an emerging convention for LLMs and AI agents. But it also says the file is optional; a missing file is not a failure. This guidance concerns agentic browsing, not Google Search eligibility.
That difference matters commercially. llms.txt is optional for Google Search, not universally meaningless. A company may choose to maintain it for a specific tool, agent workflow or experiment. What it should not do is confuse an optional convention in one technical context with a ranking requirement in another. The sound procurement question is simple: which named system consumes this file, for which task, and what evidence shows an outcome worth paying for?
The same test applies to every proposed GEO tag or AI-only rewrite: if the provider cannot identify the consuming system and observable benefit, the implementation is speculation.
The money moves toward originality and evidence
Google’s guidance does not make AI visibility cheap. It changes where the defensible spending sits. If special markup does not buy preferential treatment, then advantage has to come from assets that improve the underlying information Google can retrieve: first-hand expertise, original reporting, proprietary data, useful tools, accurate product information, distinctive images or video, and technically accessible pages. Google’s own AI guide says unique, non-commodity content is likely to influence long-term presence more than the other recommendations in the document.
That is a harder proposition for publishers than installing a file. Commodity pages are inexpensive to scale because the inputs already exist elsewhere. Original evidence requires access, expertise, testing, interviews, data collection or analysis. It also requires maintenance. In return, those assets give a retrieval system something that cannot be synthesized as easily from thousands of near-identical summaries.
Google’s broader people-first guidance provides the same economic signal. It asks whether content offers original information, substantial analysis and value beyond other results, and it warns against mass-producing pages across many topics simply to capture search traffic. Original information is an asset, not a formatting trick.
For brands, this shifts budget away from interchangeable pages and toward assets that are harder to substitute. A manufacturer can publish verified testing data; a software company can expose versioned documentation and failure cases; a consultancy can publish methods and datasets. The asset differs by sector, but the principle matches Google’s preference for distinctive content.
Technical SEO still protects the investment. If Googlebot cannot access the page, if canonicalization fragments signals, if key facts are trapped in inaccessible interfaces, or if structured data contradicts visible content, originality may never reach the retrieval layer cleanly. Google therefore has not demoted technical SEO; it has narrowed the case for inventing technical work that sits outside the documented Search system.
AI visibility is measurable even when clicks become scarcer
A major change since the first wave of AI Overviews is that publishers can increasingly measure exposure directly. In June 2026, Google announced dedicated Generative AI performance reports in Search Console for Search and Discover. The Search report covers impressions in AI Overviews and AI Mode and can break them down by page, country, device and date. Google says the rollout remains limited to a subset of properties while it tests the feature. AI visibility is now measurable inside Google’s own tooling, even if the reporting is not yet universal.
Google has also added a Search generative AI control for a subset of site owners. The default option allows a site’s links and content to appear in supported generative features and to help ground responses; the exclusion option prevents that participation and the associated impressions or traffic. Separately, Google documents nosnippet, data-nosnippet, max-snippet and noindex as controls that affect how content can be used or displayed in AI Overviews and AI Mode. Those are real technical controls, but they govern access and presentation rather than confer an optimization boost.
The commercial stakes are not limited to citation counts because generated answers can alter click behavior. Pew Research Center analyzed the March 2025 browsing activity of 900 U.S. adults and 68,879 Google searches. When an AI summary appeared, users clicked a traditional result in 8% of visits, compared with 15% when no AI summary appeared; a cited link inside the AI summary was clicked in 1% of visits. The study reflects one U.S. panel and one month, so it should not be treated as a universal 2026 click-through rate. It does show why exposure and traffic need to be measured separately.
The measurement layer is also immature. Google recorded a logging error that depressed generative Search impressions from August 13 through August 17, 2026. That was a reporting defect, not lost visibility, and shows why one AI-impression chart is not a complete business outcome.
Google’s position has limits beyond Google Search
The strongest counterargument to “GEO is just SEO” is that Google does not define the entire AI discovery market. Microsoft uses the GEO label openly. In February 2026, Bing introduced an AI Performance dashboard in Webmaster Tools that reports citations, cited pages and sampled grounding queries across Microsoft Copilot, Bing AI summaries and selected partner experiences. Microsoft called the release an early step toward GEO tooling.
That does not prove Google needs a parallel layer. It shows that Google’s position is platform-specific. Different systems can expose different controls, retrieval paths, feeds, agent protocols and measurement surfaces. A company seeking visibility across Google Search, Copilot, ChatGPT, Perplexity and browser agents may therefore need a broader discipline than traditional Google SEO. The mistake is to collapse that cross-platform work into a claim that Google itself requires a new file or markup.
Google’s own developer organization supplies the second limitation. Chrome is building tools for agentic browsing, where software does not merely retrieve information but interacts with forms, buttons, shopping flows or booking interfaces. Its developer guidance says agents may interpret screenshots, HTML and the accessibility tree, and it recommends stable layouts, semantic interactive elements and machine-readable states. Agentic browsing is a different problem from being cited in an AI Overview.
This resolves an apparent contradiction. Search Central can say llms.txt is unnecessary for Google Search while Chrome experiments with an optional audit for agents. One concerns retrieval inside Search; the other concerns machine interaction after an agent reaches a site.
The uncertainty is real because these layers may converge. Search agents could eventually transact, compare or complete tasks on sites in ways that require explicit protocols. Google’s current AI guide already points site owners toward agent-friendly practices and emerging protocols. That possibility justifies monitoring and experimentation. It does not justify retroactively calling every speculative GEO mechanism a current Google Search requirement.
Publishers need a portfolio decision, not a new checklist
For an SEO leader, publisher or brand, the practical response is to separate work by evidence level. Protect crawlability first. Verify that important pages are accessible to Googlebot, return successful responses, are indexable, have coherent internal links and expose key information in text. Use structured data where it serves documented Search features and ensure it matches visible content. These are ordinary technical fundamentals, but they remain prerequisites for generative Search participation.
Fund distinctive evidence second. Audit the pages that answer strategically important questions and ask whether they contain anything a model could not cheaply reconstruct from the rest of the web. Replace recycled summaries with first-hand observations, proprietary datasets, expert analysis, original visuals, useful tools or verified operational detail where those assets make sense. Google’s guidance repeatedly points toward unique, people-first, non-commodity content rather than AI-specific prose patterns.
Measure AI exposure third. Where Search Console’s Generative AI report is available, establish a baseline by page, device and country, then connect that exposure to analytics and business outcomes. Do not confuse impressions with citations of equal prominence, clicks with conversions, or short-term reporting noise with a ranking shift. Google itself warns that third-party tools do not have access to its internal ranking or AI systems, so proprietary visibility scores should be treated as modeled indicators rather than internal Google truth.
Experiments can then sit in a bounded budget. If another platform documents a feed, protocol or file it actually consumes, test it. If an agent workflow depends on semantic HTML or accessibility-state quality, engineer for that flow. If a vendor proposes a Google-specific GEO tactic, demand a falsifiable claim: the exact mechanism, the target surface, the expected metric and a control condition.
This portfolio approach avoids freezing an old SEO playbook while retrieval changes, without spending heavily on rituals whose main attraction is that they sound machine-readable.
The next break will come from agents, not citation markup
As of August 26, 2026, the evidence supports a firm operating judgment: no parallel Google Search optimization stack is justified today. Mueller’s August 20 comment matches the formal Search Central guide, the June llms.txt clarification and Google’s longstanding technical requirements. AI Overviews and AI Mode change retrieval, synthesis and presentation, but Google still describes their web grounding as an extension of Search rather than a separate crawl-and-rank system.
That conclusion can change. The clearest trigger would be Google documenting a Search-specific machine protocol, file, markup type or eligibility requirement that affects generative retrieval independently of ordinary indexing and Search ranking. Another would be first-party evidence that a new control measurably changes source selection rather than merely access or presentation. Until then, invented technical layers should be treated as hypotheses, not requirements.
The more plausible source of genuinely new web engineering is already visible elsewhere. Chrome’s agent-ready work treats an AI agent as a visitor that must understand and operate interfaces, and Google’s AI Search guide points readers toward agentic experiences and emerging protocols. That can require additional engineering because completing a purchase, booking or workflow is not the same task as retrieving a paragraph for an answer.
So the “yet” belongs to agents and transactions more than to citation markup. Search teams should watch that boundary carefully. If Google Search begins depending on agent protocols for particular commercial experiences, the optimization surface could expand. If it does not, the durable advantage remains less glamorous: be crawlable, publish information worth retrieving, make important facts easy to verify, and measure where the new interfaces actually create value.
That leaves room for GEO as a business discipline—cross-platform visibility, monitoring, content strategy and experimentation—without granting it an unsupported technical mythology. Google’s message is not that generative Search changes nothing; the changes that matter are currently above the old technical foundation, not beside it.
Questions publishers are asking about Google’s generative Search optimization
No. Google says there are no additional technical requirements or special optimizations needed to appear in AI Overviews or AI Mode. A page must be indexed and eligible to appear in Google Search with a snippet, and ordinary SEO fundamentals still apply.
Google’s Search Central guide says Google Search does not use llms.txt as a special mechanism for appearing in Search, including its generative AI capabilities. Google later added a documentation clarification specifically addressing the file.
No. Chrome’s Lighthouse documentation treats llms.txt as an optional emerging convention in the context of agentic browsing. That is separate from Google Search ranking and retrieval, so usefulness depends on the specific system that consumes it.
No required chunk size exists in Google’s guidance. Google says site owners do not need to break content into tiny pieces for generative Search and should choose page length and organization for the audience and subject.
No. Google says generative Search still relies on core Search ranking systems, but fan-out can issue related searches and retrieve pages from a wider set of subtopics. That means the original query’s ranking is not the only path by which a page may become useful to an AI response.
Yes, for properties that have access to the phased rollout. Google’s Generative AI performance report in Search Console shows impressions for AI Overviews and AI Mode and can be segmented by page, country, device and date.
Google is rolling out a Search generative AI control that can exclude a property’s links and content from supported generative features. Existing preview controls such as nosnippet, data-nosnippet, max-snippet and noindex also affect how content can appear or be used.
Yes, if GEO means managing visibility across multiple AI-driven discovery systems, measuring citations and grounding, improving distinctive content, and testing documented platform features. Microsoft, for example, explicitly uses the GEO label in its Bing Webmaster tooling. The unsupported leap is treating every GEO tactic as a Google Search requirement.
A material change would be Google documenting Search-specific technical requirements or protocols for generative retrieval that operate independently of ordinary Search indexing and ranking. Current Google documentation does not establish such a layer; its separate agentic guidance instead concerns agents interacting with websites.
Author:
Jan Bielik
CEO & Founder of Webiano Digital & Marketing Agency
This article is an original analysis supported by the sources cited below
Primary source for Mueller’s August 20, 2026 statement that Google sees nothing special site owners need to do for generative AI responses in Search.
Google says there is nothing special to do for generative AI responses in Search
Search Engine Roundtable’s August 25 report that surfaced Mueller’s comment and connected it with Google’s earlier public guidance.
Optimizing your website for generative AI features on Google Search
Google’s central 2026 guidance on RAG, query fan-out, non-commodity content, technical structure and the AEO/GEO practices it says Search does not require.
A new resource for optimizing for generative AI in Google Search
Google Search Central’s May 15 announcement establishing the guide’s purpose and its emphasis on SEO foundations, mythbusting and distinctive content.
Google documentation on eligibility, Search indexing, supporting links, fan-out and the conventional technical requirements that apply to AI Overviews and AI Mode.
Google Search technical requirements
Primary documentation for Googlebot access, successful HTTP responses and indexable content as minimum conditions for Search eligibility.
Creating helpful, reliable, people-first content
Google’s quality guidance on original information, reporting, research, analysis, first-hand expertise and avoiding search-engine-first mass production.
Robots meta tags specifications
Google documentation explaining how nosnippet, data-nosnippet, max-snippet and noindex affect AI Overviews and AI Mode.
Generative AI performance report in Search Console
Google Help documentation defining the phased report, included generative Search features, impression metrics and available dimensions.
Introducing Search Generative AI performance reports in Search Console
Google’s June 3 launch announcement for dedicated generative AI reporting in Search Console and its initial measurement scope.
Google Help documentation for including or excluding a property from supported generative Search features and grounding.
Latest Google Search documentation updates
Google’s changelog recording the May generative optimization guide and the June 15 clarification about llms.txt.
A developer toolkit to make your website agent-ready
Chrome’s explanation of the distinct agentic-browsing problem and the engineering signals agents need when interacting with websites.
Chrome developer documentation showing llms.txt as an optional emerging convention for agentic browsing rather than a Google Search requirement.
Introducing AI Performance in Bing Webmaster Tools public preview
Microsoft’s first-party description of citation, cited-page and grounding-query measurement and its explicit framing as early GEO tooling.
We analyzed 137K sites and 97% of llms.txt files never get read
Ahrefs’ May 2026 log study on llms.txt adoption and request activity, including its methodology and sample limitations.
Update on AI Overview citations from Google’s top ten results
Ahrefs’ March 2026 analysis of 863,000 SERPs and roughly four million cited URLs, used to assess overlap between original-query rankings and AI Overview citations.
Google users are less likely to click on links when an AI summary appears
Pew Research Center’s browsing-data study of March 2025 Google use, providing evidence on click behavior when AI summaries are present.
| Citing this article? Brief excerpts are welcome. Please credit Webiano.digital, name the author where stated, and include a link to https://webiano.digital and to this original article. Full or substantial republication requires prior written permission. Read our Copyright and Content Use Policy. |
This article was prepared with the assistance of artificial intelligence tools. The content underwent expert human review, and Webiano Digital & Marketing Agency assumes editorial responsibility for its final version and publication.
