Lookalike audiences for musicians

Last updated: July 27, 2026

Everybody writing about music ads covers lookalikes. Almost nobody links to Meta. The numbers in circulation — the minimum, the recommended seed size, the percentage range — are quoted from each other rather than from a source. Here is the version with the documentation attached.

What a lookalike actually is

Not a list of people similar to your fans. A model of your seed, projected across a population.

You give Meta a seed audience. Meta learns what distinguishes those people from everyone else, applies that model to an entire country, scores every eligible person and hands you a slice off the top. A 1% lookalike is the top-scoring one per cent of that country.

Meta's Marketing API guide to lookalike audiences is the source for everything numeric here. It describes the similarity option as including “the top 1% of people in a selected country who are most similar to the seed Custom Audience”, with reach as the top 5% and “a less precise match”. The word lookalike sounds like matching; it is scoring and slicing, and that difference explains most of the mistakes people make with them.

What the percentage means

Not a similarity score and not a confidence level. It is the size of the slice you take off the top of a ranked country.

1% is the closest match. Smallest audience, tightest fit, highest cost per person reached.
Bigger percentages are looser, not different. A 5% lookalike contains the 1% plus four points of worse fits, so running both largely pays twice for the same top slice.
The documented range is 1% to 20%, in whole-point steps — Meta's API reference gives ratio as “0.01-0.20 incremented by 0.01”. If the Ads Manager control offers less, the interface is narrower than the API.
You can take a band rather than a slice. The API exposes a starting_ratio, so a 1%-to-2% lookalike is real: the second percentile only. That is the clean way to scale once your 1% is exhausted.

What we will not tell you is how many people a 1% lookalike contains in your country. That figure appears on competitor pages as though measured; it is arithmetic, and Meta's newer lookalike behaviour stops returning a size at all.

The minimum, and this time it is real

Elsewhere in this cluster we explain that Meta publishes no minimum custom audience size. Lookalikes are the opposite case, so it is worth being equally precise here.

Meta documents a lookalike minimum, and it is 100. The Marketing API guide states it twice: “If you have a Custom Audience with at least 100 people, you can build lookalike audiences based on it”, and “Origin audiences must have at least 100 members.” For conversion-based lookalikes it asks for “at least 100 unique conversions” and adds that “more converters result in a better predictive model and we suggest 200 or more” — Meta saying, in its own words, that 100 is a floor rather than a target.

Seed quality against seed size

You will read constantly that Meta advises prioritising seed quality over seed size. We could not verify that attribution — the advice sits on Meta's Business Help Center, and facebook.com/business/help blocks automated fetching. What Meta's developer documentation says under Best Practices is the opposite emphasis, verbatim: “Seed Custom Audience — Make it as large as possible so we have enough data to find similar people.”

Meta is optimising for training-data volume, and it is right to. But the two only conflict if you enlarge a seed by loosening its definition — which is exactly how artists enlarge seeds. Going from 400 buyers to 4,000 buyers is strictly better. Going from 400 buyers to 4,000 “people who visited the site” is a vaguer question, not a bigger seed: you have asked the model to describe a category containing buyers, bored scrollers, link previewers and whoever your last promotion sent you.

The rule that survives both sources: make the seed as large as you can without changing what it is a seed of.

The best seeds for a musician, ranked

Ranked by how hard the behaviour is to produce by accident, which is the same thing as how much information it carries.

Buyers. Anyone who gave you money: a record, a ticket, merch, a crowdfunding contribution. No bot has ever bought a t-shirt. With 100 buyers, that is your seed even if you have 40,000 site visitors.
Email subscribers. People who typed an address and confirmed it. Nearly as deliberate as a purchase, usually more numerous, and data you own. For most independent artists this is the practical answer.
Deep engagers. People who did something requiring intent: played a track through, reached a second page, started a checkout. This is where the events you fire decide which seeds you can build.
All site visitors. The weakest seed and the default, because it exists without effort. Not useless if it is all you have, but understand the request: model everyone who ever loaded a page.

One documented quirk before you get clever with segmented seeds: Meta warns that if your seed has attributes such as gender or geography, the lookalikes generated may not adhere to them. Apply those as ad set targeting instead.

A bad seed gives you a confident audience, not a vague one

Contaminate a retargeting audience and you get a slightly worse retargeting audience. Contaminate a lookalike seed and that contamination becomes the input to a model applied to millions of strangers.

Bought streams, engagement pods, purchased “playlist placement”, click farms: all of those produce a pattern, and a pattern is precisely what the model is built to find and extend. The failure is not a fuzzy audience. It is a large, sharply-defined audience of exactly the wrong people, delivered with total confidence, and no screen anywhere tells you it happened.

So if you have ever bought promotion, do not seed from a website audience whose retention window overlaps that period. Seed from buyers or your email list — the two sources bought traffic cannot enter. The clean-up is in bot traffic in retargeting audiences, and the subtraction side in who to exclude.

Countries and regions

A lookalike is built against a place, and this catches artists out, because music audiences are rarely national.

You need 100 seed members from the target country. If not, Meta looks elsewhere only if allow_international_seeds is on, and it defaults to false. A worldwide following of 300 spread thinly may not qualify anywhere.
One lookalike per market. Releasing in the UK, Germany and the US means meeting the threshold three times, not one audience with three countries bolted on. The API does support region and free-trade-area groupings.
Where the location is set has moved. Meta documents a change in which lookalike creation stops taking a country and the location comes from the ad set's targeting instead. That note carries a 2021 date and says the change was delayed, so check your own account rather than a five-year-old notice.

When a lookalike beats interest targeting

Both describe strangers; they differ in what the description is made of. Interest targeting is a hypothesis you wrote — you decide your listeners like a particular band, and bet on your own theory. A lookalike is a hypothesis the platform wrote from evidence you supplied: illegible, but not limited to attributes you thought of.

Lookalikes win when you have a real seed of unambiguous behaviour and enough of it, and when your audience does not map onto nameable interests. Interest targeting wins when the seed is small, old or of uncertain provenance; when you are entering a market where you have no seed at all, since a lookalike cannot be built from nothing; and when you know something the platform cannot infer — scene, city, label, a tour you supported. On why a lookalike is not a fix for a small warm pool, see how small is too small.

Refreshing the seed

The lookalike is maintained for you: Meta states it “refreshes members in a lookalike every 3 days” while the audience is attached to a live ad. What is not refreshed is the model's idea of who you sell to, because that came from the seed at the moment it was built. So rebuild when the seed becomes a materially different list:

After a release cycle, when the buyer list has roughly doubled.
After a tour, when the geography of your buyers has changed.
After you clean the source — a narrowed retention window or an excluded bought period does not reach a lookalike that already exists.
After a change of direction. If the new record is not for the people who bought the last one, the seed describes the wrong audience perfectly.

Two housekeeping rules from the same documentation. A lookalike unused in active ads for 90 days is marked inactive and stops returning a size estimate. One unused for two years is flagged as expiring and deleted 90 days later. Neither is an error, and both look like one.

A note on numbers

Every figure here — 100 members, 100 per country, 1% to 20%, top 1% against top 5%, the three-day refresh, the 90-day and two-year rules — comes from Meta's Marketing API lookalike documentation, linked above, accessed 27 July 2026.

What you will not find: a headcount for a 1% lookalike, a recommended seed size, or a performance multiple over interest targeting. The first is arithmetic dressed as measurement. The second lives on a Meta help page we could not fetch and will not pretend to have read. The third is traceable to nothing at all.

Common questions

What is a lookalike audience, actually?

It is a model, not a list of similar people. You hand Meta a seed audience, Meta learns what distinguishes those people from everyone else, then applies that model across a whole country and ranks the population by how well each person fits. A 1% lookalike is the top-scoring 1% of that country. Nobody in it has heard of you. The word "lookalike" makes it sound like matching; it is closer to scoring and slicing.

What is the minimum size for a lookalike source audience?

Meta's Marketing API documentation is explicit where its help pages are not: origin audiences must have at least 100 members, and you need at least 100 of them from a single country. If the seed does not contain 100 people from the target country, Meta looks elsewhere only if you have allowed international seeds — a setting that defaults to off. This one is a real, documented, checkable minimum, unlike the custom-audience minimum the rest of the vertical quotes.

Should I use a 1% or a 10% lookalike?

Start at 1%. Meta describes the percentage as the top x% of a selected country most similar to your seed, so 1% is the tightest match and every step up trades precision for reach. The Marketing API documents a range of 1% to 20% in whole-point steps, and also supports a band — a 1% to 2% slice, for instance, which lets you scale into fresh people without paying again for the ones the 1% already reached.

What is the best lookalike seed for a musician?

In order: people who paid you, then people who gave you an email address, then people who did something deliberate on your site, then everyone who ever loaded a page. Buyers win almost every time because paying is the hardest thing on that list to do by accident, and the value of a lookalike comes entirely from how unambiguous the seed behaviour was.

Can I build a lookalike from a contaminated audience?

You can, and it is the most expensive mistake in this area. Contaminate a retargeting audience and you get a slightly worse retargeting audience. Contaminate a lookalike seed and the contamination is amplified, because Meta models whatever pattern is in the seed and projects it across millions of people. Bought promotion, engagement pods and bot traffic all produce a pattern, and a pattern is exactly what the model looks for. You get a large, confident, precisely wrong audience.

How often should I refresh a lookalike audience?

Refresh the seed, not the lookalike. Meta's documentation says it refreshes members of a lookalike every three days while it is attached to a live ad, so the audience is not stale in the way people assume — but the model behind it is only as current as the seed. Rebuild when the seed becomes a materially different list: after a release cycle, after a tour, or after you clean the source.

Where this fits

This is part of our guide to retargeting for musicians. A lookalike is only as good as its seed — if your conversions are browser-only, server-side tracking is what makes “seed from buyers” mean the actual buyers, and the Conversions API setup takes a few minutes. Not connected an ad account yet? Start with connecting Meta.