A lookalike audience is a platform-generated targeting group of people who statistically resemble a source ("seed") audience you provide — typically your customers or converters. Meta, TikTok, and other platforms model shared traits and behaviors to find new prospects most likely to act like your existing best ones.
Lookalikes industrialize the oldest targeting instinct — "find me more people like these" — using the platform's behavioral graph. You upload or pixel a seed (customer list, purchasers, high-LTV segment), the platform profiles the thousands of signals those users share, and builds an audience of the closest matches who aren't already in the seed.
Seed quality is the entire game. The model can only amplify the pattern you hand it: seed with all site visitors and you get people who resemble window-shoppers; seed with your top-quartile customers by LTV and you get prospects resembling buyers worth keeping. Practical guidance: seeds of roughly 1,000–50,000 of your *best* converters outperform bigger-but-diluted lists, value-based seeds (where you pass purchase values so the platform weights toward high spenders) outperform flat ones, and recency matters — refresh seeds so the model tracks who your good customer is now.
Size is a precision dial: a 1% lookalike (the closest ~1% of the country's users) is tightest and usually converts best; 5–10% trades similarity for reach and lower CPMs. The classic structure tested 1% / 1–3% / 3–5% in separate ad sets, but the Advantage+ era has changed the calculus: Meta's automated targeting now treats lookalikes and custom audiences as *signals* it may expand beyond, and broad targeting with strong creative plus conversion optimization frequently matches manual lookalike stacks. Lookalikes remain most valuable when you have genuinely distinctive seeds (high-LTV niches), in newer accounts without much conversion history, and on platforms where you want explicit control. Always exclude the seed and current customers from prospecting sets — paying to acquire people you already have is the quietest budget leak in social accounts.
A meal-kit brand seeds a 1% lookalike from its 8,000 customers with ≥3 orders (value-weighted), excluding all current subscribers. Result vs interest targeting: 22% lower CPA. A 5% lookalike runs alongside for reach with a 15% higher CPA — kept for scale.
Meta's technical minimum is 100 people from one country, but 1,000–50,000 quality converters is the practical range. A smaller list of verified buyers beats a larger list of mixed traffic — the model copies whatever pattern you provide.
Often as signals rather than strict targeting: Meta's automation expands beyond them anyway. They still earn their keep with distinctive high-value seeds, in low-data accounts, and where you want explicit audience control or clean testing.