A lookalike audience, called a similar audience on some platforms, extends reach by modelling an advertiser's best existing users. It is the main way advertisers scale beyond people who already know them without falling back on broad demographic guesswork.

How it works

The advertiser designates a seed: purchasers, subscribers, high-value users. The platform analyses behavioural signals shared across that group and returns a much larger pool of users who resemble them. Most platforms let you set how closely the pool must match, trading precision against scale.

Why seed quality decides everything

The model can only learn from what it is shown. A seed of all buyers teaches the platform to find average buyers. A seed of the top revenue decile, or of users with high LTV, teaches it to find valuable ones, and the difference in outcome is large. A small precise seed generally beats a big indiscriminate one, though most platforms want at least a low four-figure count to model against.

Building better seeds

• Seed on value, not on volume. Top-decile customers produce a sharper model than everyone who ever converted.

• Use first-party data. As third-party signal has eroded, seeds built from first-party data have become the durable option.

• Refresh regularly. A seed from two years ago models a customer base the brand no longer has.

• Exclude existing customers from delivery. The lookalike is for prospecting; paying to reach people you already have is waste.

• Test narrow against broad. Run both similarity thresholds rather than assuming which will win.

Lookalikes and rewarded traffic

Rewarded placements reach users who are choosing to engage, which makes conversion data from them unusually clean as seed material. A conversion from an offerwall represents a deliberate completed action rather than an accidental click, so the behavioural signal underneath it is stronger than the equivalent from passive inventory.

Lookalikes vs. interest targeting

Interest targeting asks a marketer to hypothesize who the customer is and then buys against that guess. A lookalike infers it from people who already converted, which removes the hypothesis and usually outperforms it. The trade is transparency: interest segments are legible and can be reasoned about, while a lookalike is a model whose logic you cannot inspect. Most accounts run both, using lookalikes for scale and interest segments where a specific, defensible audience matters.

Common misconceptions

• A lookalike is not a demographic segment. It is a behavioural model, and the demographics it surfaces are often not the ones a marketer would have picked.

• Bigger is not better. A broader similarity threshold means more reach and a weaker match, not more of the same quality.

• It does not fix a weak offer. Better targeting sends more of the right people to the same page; the page still has to work.