Lookalike Audiences on Meta: Size, Source and 1% vs 10%

A lookalike audience is a group of people Meta finds because they resemble a source you provide, such as buyers, leads or a customer list. The source needs at least 100 people from one country, but lookalikes built from 1,000 to 50,000 high-quality people (ideally recent buyers) tend to perform best. Start with 1% for precision and move to 3-10% when you need more reach.

What a lookalike audience is

A lookalike audience starts from a source audience: a custom audience of your website visitors, purchasers, leads, video viewers, page engagers or an uploaded customer list. Meta analyses what those people have in common and then finds other people in the countries you choose who look statistically similar.

You never see who is inside the lookalike. You only control three things: the source, the location and the percentage. That is why almost all the performance difference between a lookalike that prints money and one that burns budget comes from the source.

How big does the source need to be?

Meta requires at least 100 people from a single country in the source audience to create a lookalike. That is the technical minimum, not a recommendation. With 100 people the model has very little signal and the result is close to broad targeting.

A smaller source of real buyers beats a huge source of casual visitors. If you have 2,000 buyers and 200,000 page visitors, the lookalike of the 2,000 buyers will almost always convert better.

Which source should you use?

Pick the source that is closest to the result you want the campaign to produce. The further the source is from a purchase, the more the lookalike will bring people who behave like browsers, not buyers.

Recency matters too. People who bought in the last 90 to 180 days represent your current offer and current audience better than a list collected three years ago. If lead quality dropped, rebuild the lookalike from recent qualified leads, not from everyone who ever filled a form.

Customer lists and privacy

You can upload a CSV with emails, phone numbers and other identifiers. Meta Ads Manager hashes the data in your browser (SHA-256) before it is uploaded, uses the hashes only to match accounts and then discards the unmatched ones. You still need a legal basis to use that data under GDPR or LGPD, and your privacy policy should say you share hashed data with advertising partners for this purpose.

1%, 3%, 5% or 10%: which percentage?

The percentage is the share of the chosen country's population included in the lookalike. 1% contains the people most similar to your source; each additional percent adds people who are progressively less similar.

In large countries even 1% is millions of people, so there is rarely a need to start above 1% or 2%. You can also create non-overlapping ranges, such as 0-1%, 1-2% and 2-5%, and test them as separate ad sets to see where performance falls off.

If you select a lookalike built for a whole country and then target only one state or city in the ad set, delivery is restricted to people inside that location. The lookalike does not get rebuilt for the state; it just gets smaller.

How to create a lookalike audience, step by step

1. Open Ads Manager, go to Audiences and click Create audience, then Lookalike audience.
2. Select the source: a custom audience (for example, Purchase in the last 180 days) or an uploaded customer list.
3. Choose the location. Pick the countries you actually sell to.
4. Choose the audience size (1% to 10%) and, if you want, several non-overlapping ranges.
5. Create it. Meta usually takes from a few hours to a day to populate it, and keeps it updated while it is used in active ads.
6. Use it in an ad set and compare it against broad targeting under the same budget and creative.

Lookalikes and Advantage+ audience in 2026

Meta has been moving targeting toward Advantage+ audience, where your audiences work as suggestions rather than hard limits. In many campaign setups a lookalike is now a starting signal that Meta can expand beyond when it predicts better results. Lookalikes still matter: a strong source gives the algorithm a better starting point, which usually shortens the learning phase.

In practice, run both: one ad set with the lookalike as a suggestion or strict audience, and one broad or Advantage+ ad set, then keep the one with lower cost per result over at least 7 days.

How to test whether a lookalike works

• Compare it against a broad or Advantage+ ad set with the same creative and budget, running at the same time.
• Judge by cost per purchase or cost per qualified lead, not by CTR or CPM.
• Give it enough data: wait for the ad set to exit the learning phase or at least 7 days.
• Check audience overlap between lookalikes and exclude recent buyers if you only want new customers.
• If results decay, refresh the source (recent buyers) rather than increasing the percentage.

Why tracking quality decides lookalike quality

A purchaser lookalike can only be as good as the purchase data behind it. When the pixel misses sales because of ad blockers, iOS restrictions or checkout on another domain, the source audience is incomplete and biased toward the buyers the browser happened to report. Sending purchases server-side with the Conversions API fills that gap.

IzeAds sends sales from checkout platforms like Hotmart, Kiwify and Stripe to Meta through the Conversions API and ties each sale to the ad that generated it, so purchaser audiences and lookalikes are built from the sales that really happened, not just the ones the browser caught.

Common mistakes

• Using all website visitors as the source for a sales campaign.
• Starting at 10% with a small budget, which spreads spend too thin.
• Mixing several countries without checking where the source actually lives.
• Never refreshing the source, so the lookalike reflects old customers and an old offer.
• Letting several lookalike ad sets compete for the same people instead of consolidating.