Lookalike audiences let Meta find people who resemble your best existing customers. You provide a seed audience — a list of your customers, pixel events, page engagers — and Meta's algorithm identifies the characteristics that distinguish your seed from the general population, then builds an audience of people with similar characteristics. The quality of the lookalike is entirely determined by the quality of the seed.
Seed audience quality is the primary variable. The best seed audiences are lists of your highest-value customers, not your broadest audience. A seed of 200 customers who spent over €500 produces a better lookalike than a seed of 10,000 website visitors who spent any amount or nothing. The algorithm needs to identify specific signals that distinguish high-value customers — those signals are obscured when the seed contains many people who share few characteristics with your ideal customer.
Minimum viable seed size is 100 people. Meta requires at least 100 seed members to build a lookalike, though performance improves significantly between 100 and 1,000, and begins to plateau around 5,000. For most advertisers, a customer email list of 1,000–5,000 high-value customers produces the most effective seed.
Seed audience quality hierarchy: customer purchase list (highest — actual buyers), Pixel Purchase event audience (actual tracked buyers), Pixel Add-to-Cart audience (strong purchase intent), Pixel ViewContent audience (lower intent, larger), video view audiences (depends on content quality), page engagers (broad, mixed intent), all website visitors (lowest signal quality for purchase lookalikes).
Size percentage selection. Meta lookalike size ranges from 1% (smallest, most similar to seed) to 10% (largest, least similar). The 1% lookalike is tighter and typically converts better for direct-response campaigns. Larger percentages are appropriate for awareness campaigns or when the 1% audience is too small for your budget.
Post-iOS 14.5 match rate reality. iOS 14.5 App Tracking Transparency significantly reduced Meta's signal from iOS devices. Customer email list matching has become more important relative to pixel-based lookalikes, because list matching relies on deterministic identity matching (email hashing) rather than probabilistic cross-device tracking that iOS restrictions disrupted.
Lookalike refreshing. Customer purchase lists should be refreshed every 30–90 days as new customers accumulate. A lookalike seeded from purchases two years ago reflects who your customers were then — which may differ if your product positioning has shifted. Pixel-based lookalikes update automatically.
When lookalikes thin out. In smaller markets — Belgium, Netherlands, specific European countries — lookalike audiences at 1% can be very small (under 100K people). At this size, the learning phase requires longer, frequency rises faster, and lookalike performance degrades sooner. In small markets, a 2–3% lookalike or broad targeting may be more practical.
The case for broad audience testing. Meta's algorithm in 2026 is strong enough that some advertisers with mature pixels and sufficient conversion history are finding that removing audience targeting restrictions outperforms lookalike targeting. For accounts with thin conversion history, lookalikes remain the more efficient starting point.
Stacking vs. separating lookalike percentages. Running multiple percentages as separate ad sets gives granular control. Combining multiple percentages and using Meta's Advantage+ audience increasingly outperforms manually layered lookalike structures in accounts with sufficient conversion data. Digital Face monitors your Meta audience performance, tracks lookalike audience size trends, and surfaces when seed audience refresh would likely improve lookalike quality. Free plan at digital-face.nl, no credit card required.