How does Meta decide who looks alike?
From what it knows about the source group that it also knows about everyone else: age, location, the pages they follow, what they buy through ads, how they behave in the apps. Meta scores every person in the country you choose against that pattern and takes the top slice. The slice is the percentage you set. A 1 percent lookalike is the one percent of people in that country most like your source; a 10 percent lookalike is a far bigger and looser crowd. Meta's own guidance (Meta Business Help Centre, "About lookalike audiences") is that smaller percentages match more closely and larger ones reach further.
The source matters more than the percentage. A lookalike of your buyers finds people who look like buyers. A lookalike of everyone who ever visited the website finds people who look like visitors, which includes the ones who bounced in four seconds. The better the source, the better the strangers.
What does the source need to be? (example figures)
A Newcastle physio wants new patients from Meta ads and has three possible sources.
| Source | Size | Good for a lookalike? |
|---|---|---|
| Patients who booked in the last two years, uploaded as a customer list | 640 people | Yes. Above Meta's minimum of 100, and every one of them is a real patient. |
| Website visitors in the last 180 days, from the Pixel | 3,900 people | Yes, and larger, but it includes job seekers and people checking the address. |
| People who liked the Facebook page | 210 people | Barely. Many are friends and family, and a lookalike of them finds more friends and family. |
The physio builds a 1 percent lookalike from the 640 patients, which in Australia is roughly 200,000 people, and narrows it to a 25 km radius so the ads only reach people who could come in. The campaign runs at $30 a day. The first month brings 19 new patient bookings at about $47 each, against $63 each from the broad campaign it replaced, and the difference is the source: Meta was looking for people like the ones who already pay, not people like the ones who already scroll.
Why has Meta made lookalikes harder to find?
Since 2023 Meta's Advantage+ audience treats your lookalike as a suggestion rather than a wall. You give it the source, and Meta reaches beyond the slice when its system expects a better result elsewhere. For most owner-led businesses this is fine, and the manual, fenced lookalike still exists as an option in the ad set. The thing to keep doing is the part Meta cannot do for you: keep the source audience clean and current, and exclude the people who are already customers so the budget finds new ones.
What does Cora do with a lookalike?
CoraVision reads your Meta campaigns and ad sets every night once Meta is connected, so an ad set built on a lookalike of your customers and one built on a broad audience sit side by side, each with its spend, results and cost per result for the last 7 days on your dashboard. When a strategy in your plan calls for new customers from Meta, building the source audience and the lookalike is a step in the action plan, and Mira, the Meta Ads specialist, points at each setting on your own screen while you do it. CoraVision never builds or changes an audience itself.
Questions
How many customers do I need to make a lookalike?
Meta accepts a source audience of 100 people in one country, and recommends 1,000 to 50,000 for a useful one. A list of 80 customers is too small; a Pixel audience of the last 180 days of website visitors usually is not.
Does Google Ads have lookalike audiences?
It had similar segments and retired them in 2023. Google now does the same job through audience expansion and optimised targeting inside a campaign, which you switch on rather than build. The idea is the same; the control is less.
Should I pick 1 percent or 5 percent?
One percent for a campaign that has to pay for itself from the first week, five or ten percent when you need reach and can accept a lower rate of sales. In Australia, 1 percent is about 200,000 people, which is plenty for most owner-led businesses.