How lookalike audiences work
You provide a seed audience: a list of users you want more of. The platform compares the seed against its own user base, identifies common patterns, and finds other users who match those patterns.
The seed can be:
- A customer list uploaded as hashed emails (Meta Custom Audiences, Google Customer Match)
- A website audience built from pixel events (purchasers, repeat visitors, high-LTV customers)
- An engagement audience built from past ad interactions
The platform returns an audience of similar users you can then target with ads. Effectively a way to scale your first-party advantage into the platform’s audience graph.
Seed quality dominates outcome
The single biggest determinant of lookalike performance is the quality of the seed:
- All purchasers, decent seed, decent lookalike
- Repeat purchasers, better seed, better lookalike (filters for engagement)
- Top 10% LTV customers, best seed, sharpest lookalike (filters for value)
- Recent purchasers (last 30-60 days), best seed for adapting to current trends
- Cart abandoners, usually bad seed (over-represents browsers, not buyers)
Refresh the seed regularly. A lookalike built on 2024 customers is finding 2024 patterns, not 2026 patterns.
Lookalike size and tradeoffs
Lookalikes come in different size percentiles (1% lookalike = closest match. 10% = broader):
- 1-2% lookalike, tightest match, smallest audience, highest CVR, fastest fatigue
- 3-5% lookalike, balanced. Common starting point
- 5-10% lookalike, broad reach, weaker match, lower CVR, longer creative runway
Start tight, expand if the tight audience can’t sustain spend.
How privacy changes hit lookalikes
The seed comes from first-party data, which is durable. The MATCHING happens inside the platform’s user graph, which has shrunk as third-party signals have decayed. Effects:
- Lookalikes built on hashed emails (deterministic match) are more durable than lookalikes built on website pixel data
- iOS 14.5+ App Tracking Transparency reduced the signal richness in Meta’s audience graph
- Apple’s SKAdNetwork and Google’s Privacy Sandbox push lookalikes toward cohort-level rather than user-level matching
The pattern survives, but it’s less precise than it was in 2018.
Common mistakes
- Letting the platform pick the seed. Curate the seed deliberately based on your highest-value customers, not “anyone who visited.”
- Never refreshing the seed. Patterns drift. Refresh quarterly.
- Treating lookalikes as set-and-forget. Audience fatigue hits within weeks. Rotate.
- Building lookalikes from low-quality first-party data. Garbage seed → garbage lookalike.
FAQ about Lookalike Audience
What is a lookalike audience?
A lookalike audience is a modeled audience of users who share characteristics with your existing customers. You upload a seed audience (e.g. recent purchasers). The platform finds similar users in its network.
What seed audience gives the best lookalike?
Tight, high-value seeds outperform broad ones. Top-decile LTV customers or repeat purchasers from the last 30-60 days typically produce sharper lookalikes than “all customers” or “anyone who visited.”
How big should the lookalike audience be?
1-2% lookalikes are tightest match, smallest audience, highest CVR. 3-5% is the balanced default. 5-10% gives broader reach but weaker match. Start tight and expand when the tight audience cannot sustain spend.