Lookalike audiences: when they work and what sort of source list to use

A lookalike audience is a list of people whom Meta considers similar to an existing list of yours. Its quality depends almost entirely on the source list: high-value buyers give a different result from all website visitors. The practical rule I work by is at least 500 people in the source, and preferably over 1,000, and the closer the event is to the real purchase, the cleaner the signal. The percentage is also a choice: a narrow percentage gives greater similarity and smaller reach, a wide one approaches broad targeting. With small accounts, broad targeting often performs no worse, so compare in separate ad sets with an equal budget. The source list is refreshed regularly.

My name is Hristiyan Panov — a performance marketing consultant with over 5 years’ experience and over €2 000 000 Managed advertising budgets for over 90 companies on Google and Meta. Below, you’ll find which list to use, what percentage to select, and when a lookalike audience isn’t the right approach at all.

Which input list gives the best result?

Listed in order of strength, from the best downwards:

• Highest-value buyers — a list of email addresses has been uploaded; the weight is the value of the orders.
• All buyers over the last 180 days.
• They started the checkout process but didn’t complete it.
• Added to the basket.
• Website visitors — the least favourable outcome, because it includes people who have no intention whatsoever.

The difference between the first and the last is not a matter of percentages, but of multiples. Most disappointments with lookalike audiences stem from the fact that the target audience was set to „all visitors“.

What percentage should you choose?

1% provides the closest match and the smallest audience; 5% and 10% broaden the scope but blur the similarity. On a small budget, 1% usually works better because the budget is sufficient to ensure adequate frequency. On a large budget, the broader percentage provides more scope for growth.

Do not test 1%, 2% and 3% at the same time — they overlap and compete in the same auction. If you do test them, test 1% against 5%.

When does broad targeting outperform lookalike targeting?

More often than you would expect. When the pixel has enough data, Meta’s algorithm finds the buyers even without a preset list — and a broad audience gives it more room to look for them. The practical indicator is the volume of events: an ad set that leaves the learning phase with a stable weekly number of purchases rarely gains from narrowing at the top.

How you can tell that broad beats lookalike in a specific account: the same creative gives a lower cost per result in the broad ad set; the lookalike ad set raises frequency without raising volume; the two ad sets start bidding against each other, because the lookalike segment is a subset of the broad audience.

Lookalike keeps its meaning when you have something the algorithm does not know: a list of your most profitable clients, a base from the CRM, or a segment defined by an offline criterion — orders by phone, customers from a physical location, repeat buyers whom the pixel never linked into one person.

How can we ensure fair testing?

In a separate ad set with a fixed budget, running in parallel with the current audience, using the same creatives. With a campaign-level budget, the new audience rarely receives enough impressions — see the difference between the CBO and the ABO.

Check the purchase price of new customers, not overall return on investment. A lookalike audience based on buyers easily reaches people who are already your customers.

How often is the source list updated?

Automatic lists from the pixel update themselves — the rule recalculates them as long as the event keeps happening. Uploaded files go stale — for an active shop, every 30 to 60 days. An old list means an audience similar to the customers you had half a year ago.

How you can tell that the file has expired: the match rate at upload drops compared with last time; the size of the source list stands still while the shop is selling; the lookalike audience stops bringing new buyers and starts covering the same people as retargeting.

The order for refreshing is replacement, not addition: you export anew for the same period, upload it over the existing list and let Meta recalculate the derived audiences. The export includes email and phone, and where you have the order value, that too, as a weight. Empty and duplicate rows are cleaned before uploading.

What are the restrictions?

The limits are checked before building the audience, not after the first weak week:

• A small country. In Bulgaria, 1% is a limited list; with a wide range of products it is quickly exhausted and frequency jumps. The sign is clear: reach stops growing, frequency goes up, and the cost per result starts rising with an unchanged creative.
• Overlap with retargeting. Exclude existing customers if the goal is new ones. The same list should not be both the source for a lookalike and an active audience for retargeting — otherwise two of your ad sets bid for the same people.
• Consent. Uploading lists with personal data requires a basis for processing — a question for a lawyer, not for the advertising specialist. This question is settled before the file is exported, not after the campaign starts.
• File quality. Unfilled phone numbers, old emails and records without a country lower the match rate, and with it the similarity of the finished audience.

FAQ — Frequently asked questions about lookalike audiences

How many people are required on the initial list?

Technically, the minimum is 100 from a single country. In practice, a result below 500 is unreliable.

Is it possible to run a lookalike campaign using an email list?

Yes, and that’s often the strongest option — especially if you factor in the customer’s value as a weighting.

Is the lookalike audience running out?

Yes, just like any other audience. The trend shows a rising frequency alongside a falling click-through rate.

Should I exclude existing customers?

When the aim is to attract new customers — yes. When the aim is to encourage repeat purchases — no.

Does ‘lookalike’ work for services?

Yes, if you have enough completed enquiries or customers. With a small volume, the list isn’t sufficient to produce a reliable result.

Conclusion

A lookalike audience isn’t magic — it’s a reflection of the list you provide. A good list of actual buyers delivers results; a list of all visitors yields roughly the same results as broad targeting would, only with less scope for the algorithm.

If you’re not sure which audience is bringing you new customers, start with an independent review.

Request a quote for a digital audit →

Related articles: CBO versus ABO · CAC and LTV · creative burnout