Meta lookalike audiences work when the source behind them is made of real buyers or real leads, and they point Meta at the wrong people when the source is people who liked a Page. Check what every lookalike in your account is built on, rebuild the strongest seed from a fresh customer list, and build the exclusion lists before you scale.
In the video I go through one account I manage: three of its four lookalikes were seeded on Page likes, the only buyer seed was 11 months old, and the labels that should have held customers and rejected leads were empty. Below is the same check written out, Claude's rebuild order for that account, and the exact steps to create the replacement in Ads Manager.
Why the seed decides where Advantage+ looks
With Advantage+ audience switched on, a lookalike you add is an audience suggestion, not a fence. Meta's Advantage+ audience page says it shows your ads to people who match your suggestions first, then searches more widely when that is likely to improve performance.
So the lookalike tells the system where to start looking. Meta builds it from the demographics, interests and behaviours of the people in your source, and the percentage sets how close the match is: a smaller range is closer, a bigger one is broader (Meta's lookalike help page).
A weak seed does not stay small and harmless. It points a very capable delivery system at the wrong kind of person, and the system then looks for more of them.
Search Engine Journal made the same point about AI and audience data in general: automation acts on the data at a scale no team can, so wrong data spreads faster (Greg Jarboe's column). If your Meta Ads performance dropped after you turned on Advantage+, the seeds are one of the first places to look.
Check 4: read what every lookalike is built on
The name of an audience tells you what someone hoped it would be. The source tells you what it is. Open Audiences in Ads Manager and read the type and source of every lookalike, one by one.
Open Ads Manager and go to Audiences.
Group the list by label if you use labels, so you can see what each group really holds.
For every lookalike, read the source underneath it, not the audience name.
Note the range, the date it was created and whether Meta can show its size.
- Is it built on people who bought, people who became leads, or people who only engaged?
- Is it built on the same source as another lookalike, just at a wider range?
- When was the source last refreshed?
- Does Meta still show a size for it?
People who like a Facebook Page are not leads, and they are not buyers either. A lookalike built on them asks Meta to find more people who like Pages.
What the lookalikes on one real account were built on
This is the account from Part 1 of the series, the one whose pixel sent page views, view content and search but no lead event. Its audiences, read on 12 October 2026 with names removed, held 8 custom audiences, 4 lookalikes and 9 saved audiences. I manage it as part of my Meta Ads work.
| Lookalike | Range | Source | Size | Created |
|---|---|---|---|---|
| L1 | 1% to 10% | People who like the brand's Facebook Page | Varies by ad set | Jul 2026 |
| L2 | 1% | People who like the same Page | Varies by ad set | Jul 2026 |
| L3 | 1%, value-based | Offline Purchase conversions | Not available | Nov 2025 |
| L4 | UAE 4% | People who like a second Page | Not available | Aug 2025 |
Three of the four are built on Page likes. Two of them, L1 and L2, even share the same seed, one at 1% and one at 1% to 10%. That is the same weak signal twice, people who liked a Page rather than people who paid.
The best seed on the account had gone stale
The only lookalike built on buyers is L3: value-based, made from offline purchases, created in November 2025. On paper it is the strongest seed this account has, yet its size reads not available.
The account's Customers label holds no audiences at all. So there is no fresh buyer list to rebuild L3 from, and the best seed is now 11 months old.
A strong seed that nobody refreshes slowly stops being strong. If your buyer lookalike shows no size and no new customer list sits behind it, treat it as a rebuild, not an asset.
A value-based buyer seed needs purchase or revenue data flowing back to Meta. If that data stopped, fixing it is a conversion tracking job before it is an audience job.
What Claude said when I pasted the audiences in
I pasted the whole audiences table into a fresh Claude chat on 12 October 2026, with every name removed and one line of context: last month the pixel sent no lead or purchase event. No customer list or personal data goes into a chat like this, only the audience types, sizes, sources and dates.
Here is the Audiences list from a Meta ad account I manage (names anonymised). Which lookalikes come from seeds that are small, stale, overlapping or built on low-quality signals, and which should I rebuild first?
- L3 first
the strongest signal, but stale. Rebuild it from a fresh customer list with purchase values.
- L1 and L2
both seeded on Page likes, and L2's 1% sits inside L1. Keep one and reseed it from real qualified leads.
Until those leads exist, it suggested C3, about 30k recent Instagram engagers, as the stand-in.
- L4
likes of a second Page, UAE only, 14 months old. Retire it.
It also guessed that L3's size reads not available because the offline uploads stopped. That is a guess, and it fits: the account has no customer list behind that seed any more.
Its order is right. Rebuild the buyer seed, cut the duplicate Page seed, retire the old one, and fix tracking before any of it.
A label called qualified leads was website visitors
The most useful line in Claude's reply was the last one: do not seed from C1. On this account, the audience under the Qualified leads label is really all website visitors, because no lead event is firing.
Without a lead event, that label is a name, not a list of leads. Anything you build on it inherits the same problem, however well it is named.
✅ Do
- Seed from customers with purchase values, or from leads your sales team confirmed.
- Check the source of every audience behind a label before you use it.
- Rebuild a seed when its source stops refreshing.
❌ Avoid
- Seed from an audience because of its label.
- Keep two lookalikes on the same seed at different ranges.
- Seed from audiences under 1,000 people; Claude flagged C4, C7 and C8 as too small.
Check 5: exclusions only work if the list exists
Advantage+ audience stays inside four settings. Locations, minimum age, languages and excluded custom audiences are audience controls, and Meta will not go beyond them unless you let it reach past your locations (Meta's help page on audience controls).
Excluded custom audiences are the one that matters here. This account has no customer list and no rejected leads to put there, and its Disqualified leads label holds no audiences. Without those two lists, your existing customers and the leads you already turned down stay in the pool.
- Your customers, uploaded as a customer list and kept fresh.
- The leads your sales team already rejected, exported from your CRM.
- Both added as excluded custom audiences on the campaigns where they matter.
If your rejected leads only live in someone's inbox, that is the real gap. A custom CRM or even a disciplined sheet makes the list possible.
When an account has no audiences at all
The other account I manage in the video had no audiences at all: Ads Manager showed Meta's empty panel. With Advantage+ audience on, that means it has no custom audience or lookalike from you to start from.
On that account the empty Audiences page matched an empty data setup: it had no dataset connected in Part 1. Build the events and the customer list first, then the audiences. My guide to Meta Ads lead generation structure covers where those audiences sit in a campaign.
Nothing is broken on an account like that, there is simply nothing to steer with. Once the first real list exists, the rest of this article applies.
Fix it in this order
Order matters, because each step feeds the next. A new lookalike built before the events work is built on the same weak signal again.
Step one is often the hardest. If events are the gap, start with Google Tag Manager and a clean data layer before you touch any audience.
How to build the new lookalike in Ads Manager
Open Audiences in Ads Manager.
Select Create audience, then choose Lookalike audience.
Pick the fresh customer list as your source, not a Page.
- Set the size with the slider
I would start at 1% for the closest match.
Select Create audience.
Meta needs at least 100 people in the source and recommends 1,000 to 5,000 (Meta's create-a-lookalike page).
There is no location step any more, because the lookalike uses your ad set's location. The new audience can say populating for up to 3 days, and you can use it straight away.
For a lead business, I'd use confirmed leads as the source rather than raw form fills. My lead generation case study shows how that looks in a live account.
Before you scale the new seed
- A lead or purchase event fires, and it carries a value.
- The customer list behind the lookalike is fresh and has values.
- Duplicate seeds are gone, and only one Page lookalike remains, if any.
- Customers and rejected leads are excluded where they should be.
- No seed comes from an audience whose label does not match its source.
The full set of checks, including lookalike source checks, is in my 47-point Meta Ads audit checklist. Work through it on the account you care about most, then decide what to rebuild.
If you would rather have someone else read the account, book a call and we will go through your audiences together.
