A lookalike audience is an advertising audience of new people that an ad platform builds by finding users who resemble a seed audience, such as existing customers, high-value buyers or engaged website visitors. In performance marketing, lookalikes are a standard prospecting tool: they extend the reach of campaigns beyond known contacts while keeping a higher expected conversion rate than broad targeting would usually deliver.
The platform, not the advertiser, decides who is similar. The advertiser supplies the seed (source) audience and chooses a size or similarity level, and the platform’s models use behavioral, demographic and contextual signals to select matching users.
How lookalike modeling works
Lookalike modeling is a form of supervised similarity scoring. The platform matches the seed list to its own users, learns which characteristics distinguish those users from the general population and then scores every other user in the chosen market by similarity.
- Seed creation: the advertiser uploads a customer list or selects a custom audience built from pixel, app or engagement events.
- Matching: hashed identifiers such as email addresses or phone numbers are matched to platform accounts.
- Feature learning: the model identifies patterns in interests, behavior, device use and other signals.
- Scoring and ranking: users outside the seed are ranked by how closely they resemble it.
- Audience cut-off: the top-ranked share of the market becomes the lookalike audience.
The resulting audience is refreshed periodically by the platform as user behavior and the seed change. The exact features and refresh cycles are not disclosed, which makes lookalikes a black-box targeting method that must be judged by results rather than by inspection.
Seed audiences and their quality
The seed audience is the single most important input. A seed of loyal, profitable customers produces a lookalike oriented toward similar value, while a seed of all website visitors produces a much broader and weaker signal.
| Seed type | Signal strength | Typical use |
|---|---|---|
| High-value customers | Strong, value-oriented | Acquiring buyers with high expected CLV |
| All purchasers | Strong | General customer acquisition |
| Leads or sign-ups | Medium | Lead generation and trial campaigns |
| Engaged visitors | Medium to weak | Seeds for new accounts with few conversions |
| All website visitors | Weak | Upper-funnel reach when other seeds are too small |
Platforms set minimum seed sizes, often in the range of a hundred to a few hundred matched users, but such small seeds give unstable results. Practitioners generally prefer seeds of at least one to several thousand matched people. Value-based seeds, which include a customer value column, let some platforms weight the model toward more profitable customers.
Audience size and similarity
Most platforms let the advertiser choose between tighter similarity and larger reach. On Meta, the size is expressed as a percentage of the target country’s users, typically from 1% to 10%; other platforms offer labels such as narrow, balanced and broad.
| Setting | Similarity to seed | Reach | Typical outcome |
|---|---|---|---|
| 1% or narrow | Highest | Smallest | Higher conversion rate, faster saturation |
| 2% to 5% or balanced | Medium | Medium | A common starting point for scaling |
| 6% to 10% or broad | Lowest | Largest | Lower cost per impression, weaker intent |
Narrow audiences tend to show better early results but saturate quickly, which raises frequency and cost per acquisition. Broader audiences give the delivery algorithm more room to find converters and often perform comparably once the campaign has accumulated enough conversion data.
Lookalikes compared with other targeting methods
Lookalikes sit between precise first-party targeting and fully automated broad targeting. Their role has changed as platforms have shifted toward algorithmic audience expansion.
| Approach | Who is reached | Main strength | Main weakness |
|---|---|---|---|
| Custom audience | Known contacts and visitors | Highest intent and relevance | Limited reach |
| Lookalike audience | New people similar to a seed | Efficient prospecting | Opaque modeling, saturation |
| Interest targeting | Users with declared or inferred interests | Transparent selection | Often loose fit with buyers |
| Broad or automated targeting | Anyone the algorithm selects | Maximum reach for the model | Needs strong conversion signals |
Several platforms now treat lookalikes as audience signals rather than strict boundaries. Meta’s Advantage+ audience and Google’s Performance Max can use a seed list as guidance and still deliver beyond it, while Google retired its earlier similar audiences feature. Advertisers should check whether a setting is a hard limit or a suggestion.
Best practices and common pitfalls
Lookalike performance depends heavily on data quality and campaign structure. The most frequent problems come from weak seeds, overlapping audiences and missing exclusions.
- Exclude existing customers so that prospecting budget is not spent on people who already buy.
- Refresh customer-list seeds regularly, because stale lists describe past customers rather than current ones.
- Avoid heavy overlap between several lookalike ad sets, which makes them compete in the same auctions.
- Feed accurate conversions through the pixel and Conversions API, since the delivery model learns from them.
- Measure incrementality with holdout or lift tests, because lookalikes can capture people who would have bought anyway.
- Respect consent rules when uploading customer data, using hashed identifiers and a lawful basis for processing.
Lookalikes also behave differently across business models. Products with long consideration cycles, such as B2B software, often lack enough purchases for a strong seed, so teams build seeds from qualified leads or sales opportunities imported from the CRM. Mobile apps typically seed from users who completed a meaningful in-app event rather than from all installs, because install-based seeds tend to attract low-retention users and inflate cost per install without improving revenue.
Finally, a lookalike audience should be evaluated against a clear benchmark, usually a broad or interest-based ad set with the same creative and budget. If the lookalike does not beat that benchmark on cost per acquisition or ROAS over a statistically meaningful period, the seed or the size setting needs revision.
Put simply
A lookalike audience is the ad platform offering to find more people like your best customers. The business gives a list, and the platform searches its users for similar habits and interests.
It works like a matchmaking service. You describe the people who already love your product, and the service introduces you to strangers who seem to share those traits. The better your example list, the better the matches.
In practice
A subscription box company selling specialty coffee wants new subscribers in one country. It builds a seed from about 6,000 customers who stayed subscribed for at least six months, uploading hashed email addresses together with each customer’s total revenue to create a value-based lookalike.
The team launches three ad sets at 1%, 3% and 6% sizes, excludes all current subscribers and uses the same creative in each. After three weeks, the 1% audience has the lowest cost per subscription but its frequency is rising, while the 3% audience is close in cost and has much more room to scale.
The company shifts most of the budget to the 3% audience, refreshes the seed monthly and runs a conversion lift test. The test confirms that the lookalike campaign drives a clear share of incremental subscriptions rather than only reaching people who would have joined anyway.