Lookalike Audiences — How Meta and TikTok Find Similar Users

Key takeaways

01Lookalike = new audience modeled to resemble a seed audience (your high-LTV customers, paying users, retainers).
02Quality of seed dominates outcome — small seed of high-LTV users outperforms large seed of mixed-quality users.
03Major networks (Meta, TikTok, Google) all offer lookalike audiences; mechanics differ post-ATT (smaller seeds, less precise matching on iOS).

Lookalike audiences are an ad-platform feature where you upload a seed audience (typically a list of high-value users from your CRM — paying subscribers, top-LTV cohort, frequent users) and the platform's ML model finds new users who resemble that seed. The lookalike audience is what you actually target with ads. Meta's Lookalike Audiences, TikTok's Similar Audiences, Google's Customer Match all work on this principle.

Why lookalike outperforms broad targeting

The ad network's audience graph is rich — it knows a lot about each user (behavior, interests, purchases, app usage).
Your seed audience tells the network "these are the users worth finding more of".
The network bids harder for users who match the seed pattern.
Result: typically 20-50% lower CPI than equivalent broad targeting, with higher downstream LTV.

Seed quality matters more than seed size

Small seed (1,000 high-LTV users) often outperforms large seed (100,000 mixed-quality users).
The network is modeling the pattern in your seed; if the seed mixes high and low LTV, the model gets confused.
Best practice: seed with your top 10-25% LTV cohort, not your entire user base.
Meta and TikTok both require minimum seed sizes (typically 100-1,000 users) — once you clear the minimum, quality dominates.

Post-ATT changes on iOS

Smaller effective audiences on iOS — users who opted out of ATT can't be matched against the seed.
Probabilistic matching where deterministic matching fails — lower precision.
Networks (Meta, TikTok) compensate by leaning more heavily on their own first-party signal (in-network behavior) for matching.
Net effect: lookalike still works on iOS but is meaningfully weaker than pre-ATT. The Android version remains close to original effectiveness.

Quick answers

What is a lookalike audience?

A lookalike audience is an ad-platform feature where you upload a seed audience (typically high-value users from your CRM) and the platform's ML model finds new users who resemble that seed. The lookalike audience is what you target with ads. Meta's Lookalike Audiences, TikTok's Similar Audiences, Google's Customer Match all work on this principle. Typical CPI improvement: 20-50% lower than broad targeting.

Should I seed lookalikes with my whole user base or just top customers?

Just top customers — typically your top 10-25% by LTV. Seed quality matters more than seed size. A small seed of high-LTV users (1,000 users) often outperforms a large seed of mixed-quality users (100,000) because the model is learning a clear pattern. Mixing high and low LTV in the seed confuses the model and dilutes the targeting.

Does lookalike audience targeting still work post-ATT?

Yes, but materially weaker on iOS. Users who opted out of ATT can't be matched against your seed, so the iOS lookalike audience is smaller and less precise. Networks (Meta, TikTok) compensate by leaning more on in-network first-party signal for matching. Android lookalike remains close to original effectiveness because GAID-based matching still works for the bulk of users.

Back to glossary