A lookalike audience is a Meta targeting option that takes a “seed” list of your existing users (for example, your subscribers or purchasers) and asks Meta to find new people who resemble them. You pick the seed and a target country, and Meta builds a prospecting audience of similar users. Lookalikes were a core scaling lever for years, but after Apple’s App Tracking Transparency (ATT) and the rise of Advantage+ automation, Meta increasingly does this similarity-matching on its own. Manual lookalikes still have specific uses, but they matter less than they used to, and the only reliable way to know if one helps your app is to test it in your own account.
Page Contents
What exactly is a lookalike audience?
A lookalike audience starts with a source, called a seed. The seed is a custom audience you already have: app event audiences (people who installed or purchased), customer list uploads, website visitors, or people who engaged with your content.
Meta looks at the patterns shared by the people in that seed and then finds other users who look similar. The output is a net-new audience of people who do not yet know your app but resemble the ones who do.
The simplest way to hold the two ideas apart:
- Custom audience: a list of people you already know (installers, purchasers, your CRM). Used mostly for retargeting.
- Lookalike audience: new people Meta thinks are similar to that list. Used for prospecting and acquisition.
The seed is the input; the lookalike is the output. A practical consequence: you generally want to exclude your existing users from a lookalike prospecting campaign, so you are spending on new people rather than ones who already use the app.
How does Meta build a lookalike from a seed?
Meta does not publish exactly how the model works, and the details have changed over time. At a high level, it compares the people in your seed against the broader user base in your chosen country and surfaces those who share similar signals.
Two things follow from that, and they are worth keeping in mind:
- The quality of the seed shapes the quality of the lookalike. A seed of your highest-value subscribers points Meta at a different kind of person than a seed of everyone who ever installed.
- You can usually choose how tightly the audience matches. A tighter match means fewer, more similar people; a looser match means more reach but a weaker resemblance to your seed. There is a trade-off between precision and scale, and the right point on it depends on your goals and budget.
None of this requires you to memorize a fixed audience size or a “best” percentage. Treat the match-tightness setting as a dial to test, not a number to copy from a blog post.
Why do lookalikes matter less after ATT?
Two shifts have changed the role lookalikes play.
The first is ATT. On iOS, many users do not consent to tracking, which means Meta sees fewer confirmed conversion events than it used to. Your seed of iOS converters can be smaller and less complete than your true user base, and the people Meta can confidently match may not be representative of everyone who converted. That weakens the signal a lookalike is built on, especially for deeper events like subscriptions.
The second is automation. Meta has moved toward broad, creative-led delivery through Advantage+ campaigns, where the system finds likely converters itself rather than relying on you to hand it a tightly defined audience. When the algorithm has enough conversion history, an explicit lookalike often adds little, and a hard audience constraint can even hold delivery back by excluding people who would have converted anyway.
Put together: Meta is now doing more of the “find people like my customers” work automatically, and the inputs that powered manual lookalikes are noisier than they were pre-ATT. That is why the manual lookalike is a smaller part of the playbook than it once was. Increasingly, the lever that moves results is the creative itself, because broad, automated delivery leans heavily on creative to decide who sees your ad.
When can a seed or lookalike still help?
Lookalikes are not obsolete. There are situations where giving Meta a starting signal can still be useful, particularly when the algorithm has little to work with:
- Brand-new campaigns or apps with little conversion history, where any directional signal about your ideal user can help early delivery.
- Deeper-funnel events with low volume (such as subscriptions or renewals), where confirmed conversions are sparse and a higher-quality seed can sharpen what Meta optimizes toward.
- Entering a new country, where Meta has no local conversion history for your app and a seed from an existing market gives it a place to start.
- Value-based seeds, where you point Meta at your highest-value customers rather than treating every user equally, so it optimizes toward people who tend to stick around rather than just install.
Even in these cases, a seed is a head start, not a guarantee. As a campaign accumulates its own conversion data, the value of an explicit audience usually fades and broad or automated delivery tends to catch up.
How should you actually decide whether to use one?
The honest answer is to test it against the alternative in your own account, because results vary by app, country, event, and creative.
A reasonable, conservative approach:
- Build a seed from meaningful events (purchasers or subscribers), not just installers, and refresh it periodically so it reflects your current customers rather than who they were months ago.
- Run the lookalike alongside a broad or Advantage+ campaign with the same creative and the same optimization event, so the comparison is clean.
- Judge both on the deepest reliable business metric you have (cost per trial, subscriber, or purchase), not on impression-level costs.
- Keep the structure simple. Splitting budget across many overlapping audiences starves each one of the conversion volume it needs to optimize.
- Let the result, not a rule of thumb, decide where budget goes.
If the lookalike does not clearly beat broad in your account, that is a finding, not a failure. It is consistent with where Meta has moved.
Frequently asked questions
What is the difference between a lookalike audience and a custom audience?
A custom audience is a list of people you already know, such as your installers, purchasers, or email subscribers, and it is mainly used to re-engage them. A lookalike audience is new people Meta finds because they resemble that custom audience, and it is used to reach prospects who do not yet know your app. The custom audience is the seed; the lookalike is the result.
Are lookalike audiences still worth using in Meta ads?
Sometimes. They are most useful when Meta has little conversion data to work with, such as a brand-new campaign, a low-volume deep-funnel event, or a new-country launch. Once a campaign has enough of its own conversion history, broad and Advantage+ delivery often perform as well or better, so the right move is to test the lookalike against broad rather than assume it wins.
How did ATT change lookalike audiences?
ATT reduced the number of confirmed conversions Meta sees on iOS, which makes seeds smaller and less complete and the resulting similarity matching less reliable, especially for deeper events. It did not remove lookalikes, but it weakened the inputs they depend on and pushed advertisers toward server-side measurement and first-party data to recover some of that lost signal.
What makes a good seed audience?
A seed that reflects the customers you actually want more of. Seeds built from purchasers or subscribers usually point Meta in a more useful direction than seeds built from all installers, because they represent real value rather than the average user. Keeping the seed reasonably fresh matters too, since an old seed teaches Meta to look for the customer you had, not the one you have now.
A note on method
This guide is intentionally qualitative. It describes how lookalike audiences work conceptually and how their role has shifted with ATT and Meta’s automation, without quoting specific cost figures, audience sizes, match rates, or performance multiples. Meta’s mechanics and reporting change frequently, and the only numbers you should trust for your decisions are the ones from a controlled test in your own account. Treat any benchmark you see elsewhere as a hypothesis to verify, not a fact to act on.
