The brand uploads a seed — a customer list, a recent-purchaser segment, a high-LTV cohort, an email file, or a pixel event audience — and the platform’s modeling layer returns a much larger audience of users whose profiles resemble that seed. Meta’s Lookalike Audiences are the most familiar implementation; TikTok offers a direct parallel, and Google ran an equivalent feature (“Similar Audiences”) historically before winding it down.
Seed quality is the actual lever. The algorithm faithfully models whatever you give it: seed on “all customers” and you get an audience that resembles an average buyer; seed on the top LTV decile and you get an audience that resembles a high-value buyer. This is the most-violated rule in practice. Operators commonly seed on every purchaser or the full email file, then blame disappointing prospecting performance on the platform’s optimizer rather than on the seed they fed it. The seed is the brand’s first-party data activated against the platform’s user graph; its quality bounds everything downstream.
Sizing is a secondary control. Platforms expose a precision/reach setting — tighter values stay closer to the seed and target a narrower pool; looser values extend the model further out for more reach at the cost of similarity. Meta has historically used percentage rings to expose this tradeoff, though the specific UI and ranges have shifted across product generations. Platform minimums for the seed exist, but the platform floor sits well below the operator-practical floor: a few thousand high-quality records is the rough threshold where a lookalike behaves predictably, regardless of what the platform technically accepts. Two lookalikes built from similar seeds will also tend to overlap heavily on the same users, which shows up as audience overlap in account audits.
Lookalikes were the dominant prospecting tool through the late 2010s, when seed-list signal was rich and the platforms had deterministic third-party data to model against. Post-ATT the signal feeding the models thinned. In parallel, the platforms have shifted toward broad targeting plus creative-driven optimization (Meta Advantage+, Google Performance Max), and lookalikes have become less central in that mix. Two operator jobs they still do well: seeding prospecting at scale when a brand has a strong first-party list, and constraining a broad campaign when audience size needs a floor.
The lever the operator controls is the seed, not the ring size or the platform setting.