Lookalike and Custom Audiences: A Practical Guide
Everyone debates which percentage to use. Almost nobody checks the list they built it from — which is the only part that actually decides the outcome.
Lookalike audiences generate more debate than almost any other Meta setting.
One percent or five. Website visitors or purchasers. Refresh weekly or monthly.
Most of that debate is downstream of a question nobody asks: is the list you built it from describing anyone in particular?
Custom audiences first — the raw material
Everything starts here. A custom audience is a list of people who have already done something.
| Source | Good as a retargeting audience? | Good as a lookalike seed? |
|---|---|---|
| Purchasers | Only for upsell or exclusion | Best available seed |
| High-value purchasers | Upsell | Better still — describes your ideal customer |
| Cart abandoners | Excellent | Decent — high intent, no purchase confirmation |
| All site visitors | Weak, too broad | Weak — describes anyone who ever landed |
| Video viewers | Good for warming | Poor — watching is not buying |
| Full email list | Depends who is on it | Poor if it includes non-buyers and old leads |
All of these depend on tracking being right in the first place. Thin or broken Pixel and Conversions API data produces thin audiences, which then get diagnosed as a targeting problem.
The seed decides the output
This is the part worth internalising.
A lookalike finds people who resemble your seed. If your seed is "everyone who visited the site", the resemblance it finds is "people who visit websites". That is not an audience, it is a demographic.
If your seed is "customers who bought twice in six months", the resemblance is something much more useful.
- Seed from money, not activity. Purchasers over visitors. Repeat purchasers over one-time buyers where you have enough of them.
- Seed from value where you can. Your top customers by spend describe your ideal customer far better than your average customer does.
- Prefer quality to size. A few thousand genuine repeat buyers beats a much longer list of everyone who ever gave you an email address.
- Keep it current. A seed built two years ago describes a customer base you may no longer have. Refresh periodically.
- Exclude the seed from the campaign. Otherwise you are paying to advertise to people who already bought.
- Give it a name that says what it is. "LAL 1% – Repeat Purchasers – Q3" is findable in six months. "Lookalike 3" is not.
Choosing the percentage
A smaller percentage means a closer resemblance and a smaller pool. A larger one means more reach and a looser match.
The right choice depends on two things, neither of which is a best-practice number:
- How good is the seed? A strong seed justifies going narrow, because the resemblance is worth preserving. A weak seed does not — you are just tightly matching noise.
- How much budget does the audience have to absorb? A narrow audience with a large budget will hit frequency problems quickly. If delivery is constrained or frequency is climbing, widen it.
Practical approach: start narrow with your best seed, and widen when the audience can no longer absorb the spend efficiently. Test rather than assume — the difference between percentages is often smaller than the debate about them.
The honest question: are lookalikes still necessary?
Less than they used to be.
Broad targeting now performs well on Meta because the delivery system has more signal and because creative has taken over most of the targeting work — the argument made in Facebook Ads structure and again in Meta Ads creative.
That does not make lookalikes useless. It means you should test them against broad rather than assuming they win.
Where they still earn their place:
- New accounts with a strong customer list but no on-platform conversion history to learn from.
- Genuinely niche products where broad delivery wastes a lot of impressions before finding anyone relevant.
- High-value segments you want to reach deliberately rather than hope the algorithm finds.
- Constrained budgets, where you cannot afford the learning period broad targeting needs.
Exclusions are half the job
Audiences are as much about who you remove as who you include.
Exclude existing customers from acquisition campaigns unless you are deliberately selling them something else. Exclude recent converters. Exclude your seed list from the lookalike campaign built on it. Exclude anyone already in a retargeting audience running in parallel, or you will pay twice to reach the same person with two different messages.
The same discipline applies across remarketing, and it is the single most common source of quiet waste in a Meta account.
Frequently asked questions
What is a lookalike audience?
An audience Meta builds by finding people who resemble a list you supply. You give it a seed — customers, high-value purchasers, engaged visitors — and it finds users with similar characteristics. The output is only ever as good as the seed, which is the part most accounts get wrong.
What percentage lookalike should I use?
A smaller percentage means a closer match to your seed and a smaller audience. Start narrow when your seed is high quality and your budget is modest, and widen when you need scale or delivery is constrained. There is no universally correct figure — it depends on seed quality and how much budget the audience has to absorb.
What is the difference between a custom audience and a lookalike audience?
A custom audience is people who already interacted with you — site visitors, customers, video viewers. A lookalike is new people who resemble them. Custom audiences are for retargeting and for excluding; lookalikes are for finding people who have never heard of you.
How big should a lookalike seed list be?
Meta requires a minimum, but the practical answer is different: quality matters more than size. A seed of a few thousand genuine repeat customers will outperform a much larger list of every email address you have ever collected, because the second list describes nobody in particular.
Are lookalike audiences still worth using?
They are useful, but less essential than they were. Broad targeting with strong creative now performs well because the delivery system has more signal to work with. Test lookalikes against broad rather than assuming they win — on many accounts the difference is smaller than expected.
We build seeds from customers, not clicks
Audience architecture built on clean event data, tested against broad, with the exclusions that stop you paying twice.