← All articles
Audience Intelligence6 min read

Are lookalike audiences still worth it?

When a Meta lookalike audience still earns its place under Advantage+, why the seed decides the result, and how to test it against broad.

You inherit a client account and find eleven lookalike audiences: 1% of purchasers, 3% of purchasers, 1% of page engagers, 5% of 2023 newsletter signups, and seven nobody can explain. Half sit in ad sets with Advantage+ audience switched on. On the first call the client asks: are these doing anything, or did somebody build them because that is what you did in 2019?

I have been that somebody. For years, lookalikes were the first thing I built on every new account, and it worked.

Short answer: Meta lookalike audiences are still worth building when the seed is strong: real purchasers, ideally filtered by value, on a young account with little history. Under Advantage+ audience a lookalike is a hint Meta can deliver past, so the seed matters more than the percentage. On mature accounts, broad plus strong creative often matches it.

The takeaways

  • The seed is the whole decision. A lookalike of 300 top spenders is a different instrument from a lookalike of 40,000 page engagers, and Meta will build either from as few as 100 people in one country.
  • Under Advantage+ audience a lookalike steers delivery without fencing it. Delivery can go past it, so tuning the 1% to 10% slider buys you far less than it did.
  • Settle it per account with one split test. Same creative, same budget, same optimization event, judged on new customers in your own shop data.

What does a lookalike audience do under Advantage+ now?

A Meta lookalike audience still does what it always did: it takes a source audience and finds people who resemble it, sized from 1% to 10% of a country. What changed is what happens next. With Advantage+ audience on, a lookalike goes in as an audience suggestion, and Meta can deliver beyond it when it predicts better results. For sales, leads and app campaigns, lookalikes can expand past their percentage as well.

So the lookalike stopped being a fence. It is a starting point the delivery system reads, weighs against everything else it knows about your account, and leaves when it wants to. That is why the 1% versus 3% debate feels stale: the edge you drew matters less than what you told Meta to look for.

Check the toggles in your own ad set before trusting any blog, mine included. Meta renames them often.

Why does the seed decide whether a lookalike works?

A lookalike is a model of its seed, and nothing more. Meta's documentation sets the floor at 100 people from a single country, which tells you how little it needs to build one and nothing about whether the result is any good.

Feed it page engagers and you get people who like and comment on giveaways. Feed it newsletter signups and you get people who sign up for things. Feed it your top spenders of the last 12 months and you get a model of the customer you want more of. Meta recommends value-based sources for exactly this reason.

Customer lists add a catch: the seed is only the share Meta could match, and matching is not random, so a value-filtered list can shrink to a few hundred findable people. The custom audience match rate post walks through that failure. Check what the seed contains before you judge the lookalike built on it.

When is a lookalike still worth building?

A lookalike still earns its place in three situations.

The first is a young account. A pixel with a few weeks of history gives Meta's delivery little to learn from, and a lookalike of your real customers hands it a shortcut. Meta's own learning-phase guidance puts the bar for stable delivery at about 50 optimization events in a week, and a new account is usually far below it.

The second is a value seed. If your top 20% of customers by revenue look different from the rest, a lookalike of them tells Meta which buyer you mean.

The third is a reach check. Run a lookalike ad set next to broad and compare CPM, frequency and new-customer share to see whether broad finds a different crowd or the same people at a different price.

When does broad beat a lookalike?

Broad tends to win on mature accounts: plenty of conversion history, a pixel that has seen thousands of purchases, and creative that does the targeting work itself. Meta already knows what your buyers look like, often better than a seed list does, so the lookalike adds a constraint without adding information.

People underrate the creative here. An ad that names the buyer's problem in the buyer's words filters the audience on its own: the wrong people scroll past, the right ones stop. That targeting holds up when Meta widens delivery, because the message travels with the ad.

The broad versus detailed targeting post covers the delivery setting itself. For lookalikes the rule is simple: if your account is mature and your best ad sets already run broad, a lookalike has to prove it adds something.

How do you test a lookalike against broad fairly?

Use Meta's A/B test tool so the two audiences don't bid against each other for the same people. Keep everything else identical: the same ads, the same budget, the same optimization event, the same placements. Only the audience differs, broad with no suggestion on one side and your best lookalike on the other.

Judge on the right number. Platform CPA alone flatters whichever side reached more existing customers, so pull new-customer purchases from your shop data for the test window and compare cost per new customer.

Decide the stopping point before you start, and give each side enough conversions to mean something: a 12% gap after 30 purchases per side is noise. If the result is close, keep the simpler setup, usually broad. If the lookalike wins clearly on a value seed, keep it and retest in a quarter.

What should the seed tell you about your buyer?

A strong seed is a list of your best buyers, and that list is also the brief for your creative. Who are they, what were they stuck on, which words did they use when they finally bought?

Adscalr does not build, sync or push audiences to Meta. Its audience intelligence side works on that second question: it searches 5 named public sources (Reddit, Amazon reviews, the App Store, Google Play and forums you seed with your own and competitors' URLs), keeps the exact phrases people use, maps them to awareness stages and builds 4 to 6 personas, each backed by real quotes. That tells you what the seed and the ads should say. The audience setting stays your call.

This is the thinking behind Adscalr.

See the product →