What it means
A lookalike audience is created by Meta when you provide a seed: a custom audience, a list of high-value customers, or a recent-converter segment. Meta's algorithm analyses what those people have in common (behaviour, interests, demographics, friend graph) and finds new users in your target geography who share those traits.
Lookalike audiences range from 1 percent (the closest 1 percent of users to your seed) to 10 percent (a much broader population). Smaller percentages are more accurate; larger percentages give you more reach. Most ad accounts run 1 to 3 percent lookalikes for highest precision.
Why it matters
Cold targeting (interests, demographics) is increasingly noisy. Lookalikes use Meta's full machine-learning stack to find people who actually behave like your buyers, not people who fit a stereotype of your buyer.
The seed quality matters more than the audience size. A lookalike of 200 high-LTV customers will dramatically outperform a lookalike of 5,000 random leads.
Example
A fitness studio uploads their top 300 lifetime members as a seed. Meta builds a 1 percent lookalike for Singapore (around 25,000 users). The studio runs a CTWA campaign to that audience. Cost-per-qualified-lead is roughly half what they paid on interest-based targeting, and show-up rate at trial classes is noticeably higher.