26 Aug How to Build a High-Quality Meta Lookalike Audience
Lookalike quality isn’t luck. It comes down to three decisions, each one compounding on the others: which action you build the seed from, how long a window you pull it from, and how many people actually end up in that seed once matching is done. Get all three right and a lookalike can lift results meaningfully — up to 20% in accounts where this has been tested directly. Get them wrong and you end up with an audience that technically exists but doesn’t actually resemble your real customers.
Start with the deepest action you have
The action you seed a lookalike from determines how much it actually resembles your best customers. A lookalike built from people who merely watched a few seconds of a video is a low-quality source — that’s barely an action at all. A lookalike built from actual purchasers is a much stronger source, because it’s reflecting real buying behavior rather than passive attention.
As a general rule: the deeper the action, the higher the quality. Purchase beats add to cart. Add to cart beats a website visit. A website visit beats a passive video view. If you have the option to seed from something deeper, take it — even if the resulting seed audience is smaller.
Use the longest window available — this is the opposite of retargeting
Retargeting duration should be short, matched to how long someone actually takes to decide. Lookalike seeding works the other way entirely. You’re not trying to catch anyone in a live decision window — you’re trying to give Meta enough data to build an accurate profile of what your customers look like. For that, longer is better. Pull from the longest window available to you, up to 180 days for website-based custom audiences, rather than defaulting to a short window out of habit carried over from retargeting logic.
The 1,000-person floor — and why it’s a floor, not a target
Meta needs at least 1,000 valid, matched people in a seed audience to build a usable lookalike. Below that, the resulting lookalike won’t be reliable enough to trust. That’s worth knowing as a hard technical minimum — but treat it as the floor, not something to aim for. A seed with real depth well beyond 1,000 produces a noticeably stronger lookalike than one that’s just barely cleared the threshold.
Watch for match-rate loss shrinking your seed further
This catches people off guard: your seed list looking large enough doesn’t guarantee it stays that way. A customer list of even a few hundred people can shrink to under 100-200 matched people once Meta matches it against its own user base — well below the usable floor for a lookalike, even though the original list looked reasonable. What actually matters for lookalike quality is the size of the audience after matching, not the size of the file you uploaded. Check the matched number before building a lookalike from it, not the row count in your spreadsheet.
A quick self-check
- Am I seeding from the deepest action available, or defaulting to whatever’s easiest to pull?
- Is my seed window as long as possible, rather than carrying over a short retargeting-style window out of habit?
- Have I checked the matched audience size, not just the size of the list or event count I started with?
- Is my seed meaningfully above the 1,000-person floor, or just barely clearing it?
This breakdown is written by Jason Gan, a Meta Certified Professional and Badged Meta Business Partner who has personally audited over 1,000 advertiser accounts since 2010. You can see real account breakdowns on the Jason Gan YouTube channel.
Frequently Asked Questions
What makes a Meta lookalike audience high quality? Three decisions determine it: which action you build the seed from, how long a time window you pull that seed from, and how many people actually end up in the seed after matching. Get the action deep enough, the window long enough, and the seed large enough, and lookalike quality improves meaningfully — sometimes as much as a 20% lift in results compared to a poorly-seeded one.
Should a lookalike seed use a short or long time window? Long — this is the opposite of retargeting logic. Retargeting windows should match how long someone actually takes to decide, but a lookalike isn’t trying to catch anyone in a decision window. It just needs enough data to build an accurate profile, so pulling from the longest available window (up to 180 days for website-based audiences) is the right call here.
How many people do I need in my seed audience to build a lookalike? At least 1,000 valid, matched people is the floor — below that, Meta won’t be able to produce a reliable lookalike. That’s a technical minimum, not a target to aim for. A seed with real depth, well beyond that floor, produces a meaningfully stronger lookalike than one that’s just barely cleared 1,000.
Why did my customer list seed a poor lookalike even though my list was decent sized? Match-rate loss. A customer list of even a few hundred people can shrink to under 100-200 matched people once Meta matches it against its own user base, which is well below the usable floor for a lookalike seed. What matters is the size of the audience after matching, not the size of the list you uploaded.
If you’re not sure your lookalikes are pulling their weight
Book a Meta Ads Audit — $80 / 30 Minutes →
No login required. A real look at whether your seed audiences are actually strong enough to work from.