Facebook Lookalike Audiences: Do They Still Work?
Facebook lookalike audiences still work, but they matter less than they did five years ago. Meta's delivery system has become good enough at finding buyers from creative signals alone that the source list now matters far more than the percentage you pick.
A 1% lookalike built from 300 newsletter signups will lose to broad targeting. A 3% lookalike built from 5,000 repeat purchasers will often beat it. Source first, size second.
What is a Facebook lookalike audience?
It is an audience Meta builds by finding people who resemble a list you supply.
You give Meta a source: a customer list you upload, people who took an action on your site via the pixel, people who engaged with your Page or Instagram account, or app users. Meta models the traits of that group and finds the closest matches in a country you choose.
The percentage is how wide you cast. A 1% lookalike in a country of 200 million is the two million people most similar to your source. A 10% lookalike is twenty million, and looks much more like the general population.
| Size | Approximate reach (US) | Behaves like | Typical use |
|---|---|---|---|
| 1% | ~2.5M | Tight, expensive, high intent | Small budgets, high AOV |
| 2–3% | ~5–7M | Balanced | Most accounts' default |
| 5% | ~12M | Loose | Scaling, larger budgets |
| 10% | ~25M | Nearly broad | Rarely worth the setup |
What makes a good lookalike source?
Value, recency and size, in that order.
Value. A list of your top 20% of customers by lifetime value outperforms a list of everyone who ever bought. You are asking Meta to find more people like this group, so put your best people in the group.
Recency. Behaviour shifts. A source list of 2024 purchasers describes a customer who found you under different conditions. Rebuild source lists quarterly.
Size. Meta's stated minimum is 100 people from a single country, but that is a floor, not a target. Below about 1,000 records, the model has too little to work with. One to fifty thousand is a comfortable band.
Ranked roughly by how well they tend to perform:
- High-LTV customers (purchase value-based lookalike)
- All purchasers, last 180 days
- Add-to-cart, last 90 days
- Video viewers at 75%, last 90 days
- Page or Instagram engagers, last 365 days
- All website visitors
The list descends from intent to attention. Engagement lookalikes are the fallback when pixel volume is too thin for anything better.
Do lookalikes still beat broad targeting?
Often not, and that is a genuine shift rather than a fashion.
Since the iOS tracking changes reduced pixel signal, Meta leaned harder on creative and on-platform behaviour to decide who sees what. In many accounts, a broad audience with no interest or lookalike constraint now performs comparably, because the creative itself does the targeting.
That does not make lookalikes useless. They still help when:
- Your budget is small and you need Meta to start somewhere sensible.
- Your product is genuinely niche and broad wastes impressions.
- You have a strong, recent, high-value source list.
- You are scaling and need new audiences horizontally.
They help less when your pixel is thin, your list is old, or you are running Advantage+ shopping, which largely takes audience selection out of your hands anyway.
How do you actually build one?
In Ads Manager, open Audiences, then Create audience, then Lookalike audience.
- Pick your source. Upload a customer list, or select a website custom audience, or choose a Page or Instagram engagement source.
- Pick the country. Lookalikes are country-specific — a US lookalike does nothing for a UK campaign.
- Pick the size, from 1% to 10%.
- Create several sizes at once if you plan to test them. Meta lets you generate 1%, 3% and 5% in one action.
Then use them at ad set level like any saved audience. One important habit: exclude your existing customers from prospecting ad sets so you are not paying to reach people who already bought. The mechanics are covered in excluding a saved audience.
Can you build a lookalike from a competitor's customers?
No, and any tool claiming otherwise is misrepresenting what it does.
Meta does not expose another advertiser's customer list, pixel data or audiences. You cannot target "people who bought from competitor X". The full answer on competitor targeting covers the workarounds people attempt and why most disappoint.
What you can do is study the creative that is winning in your category and build ads good enough that broad targeting finds the right people. The Meta Ad Library shows every active ad from any advertiser, free, with a start date on each card. Ads still running after 90 days are ads someone keeps funding.
Ads vanish from the library when they are paused, so save what you want to brief from. A free Meta Ad Library Downloader handles that — HD video, full carousels, bulk ZIP with a CSV and a searchable swipe board (disclosure: Klipio makes one).
How should you test lookalikes?
Use an ABO campaign with equal budgets so every audience gets a fair read. Identical creative across ad sets, so the audience is the only variable.
Give each ad set enough budget to clear the learning phase — roughly three times your target cost per purchase per day. Judge after it has spent one to two times your target CPA, not before.
Then keep the winners, exclude the overlapping tiers, and move them into your scaling campaign.
If the harder problem is finding angles worth putting in front of any audience, paid Klipio watches named competitors and turns their winning angles into on-brand creative, from $79/mo. Disclosure: ours.
How often should you rebuild lookalike audiences?
Meta refreshes lookalike membership automatically, roughly every few days, as the source audience updates. That handles drift inside a source that is itself live, like a website custom audience.
An uploaded customer list is different. It is a snapshot, and it ages. A list uploaded in January describes people who bought under January's conditions, prices and creative. Re-upload quarterly.
Rebuild sooner if something structural changed: a price repositioning, a new hero product, a shift into a different country, or a rebrand. In those cases the old source describes a customer you no longer sell to.
FAQ
What is the best lookalike audience percentage?
Start at 1–3% for most accounts. Go to 5% or higher only when you are scaling and need reach. The percentage matters less than the quality and recency of the source list.
How many people do you need for a lookalike audience?
Meta's minimum is 100 from a single country, but that is too thin to model well. Aim for at least 1,000 records, ideally your highest-value customers.
Do lookalike audiences still work in 2026?
Yes, though less decisively than before. Reduced pixel signal means Meta now leans on creative to find buyers, so broad targeting often matches lookalikes. Lookalikes still help on small budgets and niche products.
What is a value-based lookalike audience?
A lookalike built from a source that includes how much each customer spent. Meta weights the model toward high spenders rather than treating every customer equally, which usually improves results for brands with wide order-value ranges.
Should you exclude customers from a lookalike audience?
Yes, in prospecting campaigns. Exclude your purchaser custom audience so you are not paying to reach people who already bought. Keep them available for retargeting campaigns instead.
Can you create a lookalike from a competitor's audience?
No. Meta does not share another advertiser's customer data, pixel events or audiences. Tools that claim to do this are describing something else, usually interest targeting or ad-library research.
The Klipio extension adds a download button to every ad in the Meta Ad Library: one click per ad, or bulk-save a whole search as a ZIP with a searchable swipe file inside. Free, no sign-up.
Get the free extensionKlipio reads a competitor's live Meta ads, ranks them by how long they have been running — the honest signal that an ad is profitable — and turns the winning angle into on-brand creative for your own brand. 3-day free trial.
Start free — 3-day trial →
