What Is a Lookalike Audience?

A lookalike audience is built by taking a list of people who have already done something you want, and asking a platform to find others who resemble them. The resemblance is statistical: it is not a statement about any individual's character, it is a prediction about a probability.

Where the seed comes from matters more than the size

A lookalike is only as good as the list it was built from, and the quality of that list depends entirely on what the seed people did. A seed of purchasers produces an audience of people who resemble purchasers. A seed of everyone who visited the site produces an audience that resembles visitors, which is a much larger and much weaker group. The failure mode is nearly always a weak seed described as a small one.

Why percentage is a range and not a setting

Platforms describe lookalikes by a percentage, and the number is routinely misread. A one percent lookalike is not a smaller audience with worse targeting; it is a narrower slice of the model, which in most implementations means the highest-confidence matches and the smallest reach. Percentages trade size for expected similarity, and there is no universally correct setting, only a reach-versus-relevance choice that has to be made against a budget.

When lookalikes make sense

They work when there is a genuinely informative seed and enough of it. Hundreds of high-quality records is a workable seed; a few dozen is usually too few for the model to find structure, which is why small accounts are told the feature is unavailable. They also work well as a prospecting mechanism against a cold audience, because the alternative is interest-based targeting, which describes what people say they like rather than what they did.

When they do not

When the seed is contaminated. A list containing customers, refunds, testers, employees and anyone who arrived through a mispriced offer will produce a lookalike that resembles the contamination, and the resulting campaign will look efficient in every metric except the one that matters. This is the most common reason a lookalike campaign fails, and it is a data hygiene problem rather than an advertising one.

How to know whether it is working

Do not judge a lookalike on its own reported performance, because the platform is optimising toward the conversions you are counting. Judge it against what would otherwise have happened: hold out a portion of budget for untargeted or interest-based delivery and compare against that, not against a historical average from a different period.