Lookalike Audiences Explained: How to Scale What Is Already Working

Your best customers already told you who to target next. Feed a platform's algorithm a list of your real customers or highest-intent visitors, and a lookalike audience finds new people who share the underlying patterns of that list, then puts your ad in front of them as fresh, cold traffic.

A lookalike audience is a targeting tool, built into Meta and TikTok ad platforms, that takes a source list (your customers, email subscribers, or site visitors) and finds new people who statistically resemble that list. You control the source and the match size; the platform's model does the matching. The better and more specific your source, the tighter the match.

The mechanism looks the same everywhere: pick a source, pick a size, launch. What decides whether the resulting audience performs is almost entirely the source you feed it. A lookalike built off your top spenders and one built off everyone who ever clicked an ad are technically the same feature, running on completely different signal, and the gap shows up directly in your cost per result.

What is a lookalike audience, exactly?

A lookalike audience is a cold-traffic audience the platform builds for you by matching the underlying pattern of a source list you provide, instead of one you build by hand from interests or demographics.

You give it a starting point instead of a description. Interest targeting has you guessing which traits describe your buyer: age, interests, pages they follow. A lookalike skips the guessing: you hand the platform actual people, your customers, email list, best site visitors, and it reverse-engineers the pattern connecting them, then finds more people who fit it.

It's a cold-audience tool, not retargeting and not your existing customers, just the platform's best attempt at finding strangers who resemble people you already know convert. If you haven't mapped out how cold fits against your warm and hot pools yet, our ad audience targeting guide lays out where lookalikes sit in that structure before you build one.

How does the algorithm decide who's a match?

The platform runs your seed list through its own model of that country's user base, isolates what your seed shares more than the average user, then ranks everyone else by how closely they fit.

Neither Meta nor TikTok publishes the exact feature set it weighs, but purchase behavior, page interactions, device patterns, and dozens of other signals the platform already collects all factor in, which is why a lookalike can beat an interest stack you'd never build by hand, chasing correlations you'd never think to target manually.

The output is a ranked pool, not a fixed list. Every person gets scored on similarity to your seed, and the size setting, covered next, decides how far down that ranked list your audience reaches. You aren't picking traits, you're picking how deep into an existing rank order you're willing to fish.

How do you choose the right seed audience?

Choose the seed that most tightly represents the exact behavior you want more of, and when you have a choice, feed it your highest-value customers instead of your broadest list.

This is the one decision that matters more than everything else here combined: the algorithm can only extrapolate from what you hand it, so a noisy or generic seed produces a noisy, generic lookalike no matter how well you tune the size afterward.

Rank your options roughly from strongest to weakest:

Your highest-value customers. Top spenders and repeat purchasers, not every one-time buyer. The tightest seed you can build, because it isolates the exact behavior you want repeated: real money, not just a click.

Your full customer list. Weaker, since it mixes in one-time, low-value, and refunded buyers, but still real purchase behavior, well ahead of anything based on browsing alone.

Email subscribers or leads who never bought. They raised a hand but didn't convert, so the pattern includes whatever's stopping people from buying, not just what makes them buy.

High-intent site visitors. People who hit your pricing or checkout page via the pixel without buying. Better than average traffic, weaker than a real buyer list.

All site visitors. The broadest, weakest option, useful mainly when you're too early to have anything better yet.

Here's the logic with real numbers. Say you've got 4,000 total customers, and 600 are repeat buyers or top-quartile spenders. Seed off all 4,000 and the algorithm partly learns from your cheapest, one-and-done buyers. Seed off just the 600 and it learns almost exclusively from people who spend the most and come back. Same feature, same size setting, a completely different pool of prospects on the other end, because you changed what the algorithm was told to look for.

If your seed is website behavior rather than a customer list, it only exists because your pixel or conversions API is capturing it, so get that plumbing right first: our guide to setting up the Meta pixel and Conversions API walks through it. Seed options feel thin right now? Your retargeting pools are often the fastest place to find one; our retargeting guide covers building the warm and hot audiences that double as seed material later.

What size should you pick, and why does it change the result?

Every lookalike tool gives you a size dial, shown as a percentage of a country's population: smaller means a tighter match and less reach, larger means more reach and a looser match.

On Meta, that dial runs from 1% up to 10%. A 1% lookalike is the closest possible match, the algorithm reaching only as far as it has to before it runs out of people who genuinely resemble your source. A 10% lookalike casts roughly ten times as wide a net, by definition of the percentage scale, which means it loosened its definition of "similar" considerably to fill that much larger pool. Same seed, same country, a very different audience depending on where you set that one number.

Start at 1% if your seed is strong, since a tight, specific seed built on real buying behavior usually gives you the best-performing pool available at that size, full stop. The tradeoff is reach: you'll burn through a 1% audience faster, especially on a modest budget, and be back rebuilding sooner than at a wider size.

Layer sizes once you need to scale, rather than jumping straight to a loose 5% or 10% audience. Stack narrow bands instead: a 1-2% lookalike as one ad set, a 2-3% as another, each excluding the tighter bands below it, for more reach without diluting any single audience. Widen when a 1% audience you're burning through in days, with frequency climbing fast, signals it's time, not on a fixed schedule regardless of what's happening in the account.

Should you use a value-based lookalike or a standard one?

Use a value-based lookalike whenever you have real order-value or lifetime-value data attached to your list, because it tells the algorithm to chase your best spenders instead of treating every buyer as equal.

A standard lookalike treats every person in your seed the same, whether they bought once for $17 or spent thousands with you over two years. That's fine without value data to work with, but it leaves real signal on the table when you have it.

A value-based lookalike, which Meta supports natively when you upload a customer list with a value column attached, weights your seed by that value before matching. Someone who resembles your biggest spenders gets prioritized over someone who merely resembles an average buyer. You're not asking for "more people like my customers," you're asking for "more people like the customers who made me the most money."

The catch is data quality. A value-based lookalike is only as good as the values attached to it, so missing, stale, or inconsistent order values confuse the model more than they help. If you're not confident your data is clean, a standard lookalike off a tightly filtered high-spender list will usually beat a value-based one built on messy numbers.

Where do you actually build one, platform by platform?

Meta and TikTok both still give you a direct lookalike-style tool with a real seed and a real size control, while Google has moved this same logic into automated targeting signals instead of a standalone feature you configure by hand.

Meta. The core mechanic lives on, though Meta has increasingly folded it into Advantage+ Audience, its broader push toward automated targeting. In a lot of current setups, you're feeding a source audience into Advantage+ rather than picking "Lookalike" from a menu, and the system blends lookalike-style expansion with its own broad-targeting signal. For the fuller picture of how much control Advantage+ takes over, our Advantage+ vs manual campaigns guide breaks that decision down.

TikTok. TikTok Ads Manager keeps a more traditional, explicit Lookalike Audience builder: pick a source (a Custom Audience from your customer file, app activity, or engagement), set how tight or broad the match should be, and TikTok builds the pool. Close enough to Meta's mechanics that everything here about seed quality applies directly.

Google. This is where operators used to Meta or TikTok get tripped up. Google retired its standalone Similar Audiences feature years back, and the seed-and-match logic got absorbed into automated targeting inside Performance Max and Demand Gen, where the system treats your conversion data as an ongoing signal instead of a lookalike you build once and manage by hand. Don't go looking for a like-for-like button inside Google Ads. You're feeding the same kind of signal through a different, more automated door.

Everyone else. Most platforms with enough scale to model similarity offer some version of this, under names like "actalike" or "similar audience," and the same seed-quality questions apply no matter what the button's called.

How does a lookalike fit into the rest of your account?

A lookalike is cold traffic, playing the same role in your funnel as any other prospecting audience, just built on real buying behavior instead of a guess. Your strongest top-of-funnel creative goes here, not a hard pitch: these people have never heard of you, no matter how closely they resemble your customers, so the ad still has to earn attention from zero.

The exclusion that gets skipped constantly: pull your existing customers and your warm and hot retargeting pools out of every lookalike campaign. Skip that and you'll pay cold-traffic prices to keep showing prospecting ads to people who already bought, since a customer often resembles other customers closely enough to land inside their own lookalike.

Run it alongside your other cold audiences, not instead of them. A lookalike and a broad or interest campaign pull from different logic, resemblance to known buyers versus stated traits or the platform's own signal, so an even test tells you which one this offer responds to. Don't assume a lookalike wins automatically.

When does a lookalike audience need to be rebuilt?

Rebuild it when your seed has meaningfully changed, when performance decays even though nothing else in the account moved, or on a routine cadence if neither of those has happened yet, whichever comes first.

A lookalike is a snapshot, not a living connection to your customer list. Most platforms refresh the underlying match periodically on their own, but the seed you fed it can go stale in ways the platform won't fix: your product mix shifts, your price point moves, a new channel brings in a different kind of buyer, and the lookalike is still quietly built on last year's version of your customer.

Watch two signals. Plain performance decay in an audience that isn't shrinking or getting hammered by frequency means the match has drifted from what actually converts now. An audience that's visibly shrinking, climbing frequency and cost with no creative changes, means you're running out of new people inside that match. That second case overlaps heavily with plain ad fatigue, so rule that out before you blame the seed.

A reasonable habit: rebuild your core lookalikes any time your seed source has grown meaningfully, say your customer count has roughly doubled, or every few months on a fixed schedule, whichever hits first. It costs a few minutes and a short relearning window, and a fresh seed almost always outperforms a stale one.

What mistakes wreck lookalike audiences before they get a fair shot?

Most lookalike failures trace back to a bad seed, not a bad platform, and the same handful of setup mistakes show up in account after account.

Seeding off a list that's too small or too mixed. A genuinely tiny seed often won't generate a usable lookalike, and one that's big enough but mixes buyers, refunds, and freebie claimers together just confuses the signal.

Never excluding existing customers. The single most common leak: a lookalike with no exclusions quietly re-serves cold-campaign budget to people who already bought.

Defaulting straight to a broad size. Jumping to 10% because it reaches more people skips the stage where you'd learn whether this seed produces a strong match at all. Start tight, prove it, then widen on purpose.

Ignoring value data you already have. Plenty of accounts sit on a customer list with real order values attached and still build a standard lookalike out of habit, for no real reason.

Running too many layers on too small a budget. Five stacked size bands sound sophisticated, but if your budget can't feed each one enough spend for a readable result, you've built five noisy tests instead of one good audience.

How do you build your seed and size plan with AI?

Run your actual customer data and budget through a direct prompt instead of guessing at a seed and size combination that just sounds right.

You are a paid-media strategist. Build me a lookalike (or "similar
audience") targeting plan for the campaign below, specific to my
numbers, not generic advice.

My business: [WHAT YOU SELL + PRICE POINT]
Sources I could use as a seed: [CUSTOMER LIST WITH VALUES / EMAIL LIST /
WEBSITE PURCHASERS VIA PIXEL / CART ABANDONERS / VIDEO VIEWERS / PAGE
ENGAGERS - list what you actually have and roughly how many people are
in each]
Do I have order value or LTV data attached to my customer list: [YES / NO]
Platform: [META / TIKTOK / OTHER]
Current monthly ad budget: [BUDGET]

Do this:
1. Rank my available seed sources from strongest to weakest for a
   lookalike, and say why each one ranks where it does.
2. Tell me whether my best source has enough people in it to build a
   reliable match, and what to use instead if it doesn't.
3. If I have value data, tell me whether to build a value-based or
   standard lookalike, and why.
4. Recommend a starting size (or range of sizes) and whether to layer
   multiple sizes, based on my budget.
5. List the exact audiences I need to exclude from this campaign so
   I'm not paying to reach people who already know me.
Be direct. No hedging.

Rerun it any time your customer list grows meaningfully or you add a new value data source. The right seed changes as your business does, even when the prompt doesn't.

Honest limits

Everything above assumes you already have real conversion or customer data to seed with. If you're pre-launch or your pixel has barely fired, you don't have a lookalike problem yet, you have a data problem, and no seed selection or size tuning fixes that. Build cold traffic through interest or broad targeting first, bank real conversions, then come back once you have something worth seeding.

Platform minimums are real, and so are the tools themselves, and both move. Every platform enforces some floor on how small a seed can be before it refuses to build a match, or builds one so loose it's barely different from broad targeting, and the specific button names and size controls described here belong to Meta and TikTok as they exist right now. Both have reshuffled this feature's name and location before, Meta's Advantage+ push is a live example of it happening again, so treat any specific number or menu path in this piece as approximate and confirm what's current inside your own account.

None of this fixes a weak offer or creative that doesn't earn the click. A lookalike only improves who sees your ad, not what happens once they do. If your broad and interest campaigns already convert badly, expect a lookalike to convert somewhat better on the same weak creative, not to rescue it outright.

Frequently Asked Questions

What's the minimum audience size I need to build a lookalike?

You need a real seed list of actual people, not a rough guess. Platforms won't build a usable match off a source that's too small or too thin on genuine behavior, and even when they technically will, the result is often too weak to trust. If your list is still small, build it up through retargeting and conversion tracking first, then come back once you have a real base.

Is a 1% lookalike always better than a 5% or 10%?

Not always, but it's the right place to start when your seed is strong. A 1% lookalike gives you the tightest match to your source, usually meaning better performance per dollar, at the cost of a smaller pool you'll exhaust faster. Once it proves itself and you need more volume than it delivers, widening the size or layering additional bands is the next move, not a default starting point.

How is a lookalike audience different from retargeting?

They pull from different pools. Retargeting shows ads to people who already interacted with you directly, so it's warm or hot traffic. A lookalike uses a list of people you know, often the same list retargeting draws from, as a pattern to find brand-new strangers who resemble them, so it's cold traffic. Same data source, opposite jobs in your funnel.

Do lookalike audiences still matter now that broad and Advantage+ targeting exist?

Yes, because the seed still matters even when the platform handles more of the targeting automatically. Advantage+ and similar systems increasingly blend lookalike-style expansion into a broader automated pool, but you're still the one feeding it a source audience, and a sharp, high-value seed still beats a generic one inside that system.

Where to take this next

A lookalike audience is only as sharp as the seed feeding it and the account structure reading its results. Once you've got one built and spending well, how to optimize and scale ads covers the pace to grow it without tripping the same learning phase you just worked your way out of. Build the seed right, watch it for drift, and let the numbers tell you when it's time to widen or rebuild.