Lookalike Audiences

A Lookalike Audience starts with a source, which is any list Meta can read enough signal from - a Custom Audience built from customers, an Engagement Custom Audience of people who interacted with your Page or content, or a pixel-based Website Custom Audience of past visitors or purchasers - and a percentage, from roughly 1% up to 10% of a chosen country's population. Meta's modeling looks at the shared characteristics of the people in your source and finds a population of that percentage size who share the most similar signal profile, drawing from the full range of data Meta holds on users in that country, not just the demographic basics you can see in targeting.

The percentage setting is the main lever and it trades resemblance for reach in a straightforward, predictable direction: a 1% Lookalike Audience is modeled as closely as possible to your source and will be the smallest, while moving toward 10% widens the pool progressively at the cost of looser resemblance. There is no universally correct percentage - a 1% audience makes sense when your source is a strong, proven signal (recent high-value purchasers) and you need precision over volume, while a wider percentage makes sense when you need more delivery room than a 1% pool can support, particularly for a smaller advertiser whose source list and budget can't sustain a tightly modeled tiny audience.

Source quality matters more than source size in most cases, and this is where the setting most often gets misused. A value-based Lookalike Audience, built from a Custom Audience or Customer List Custom Audience that includes purchase value data, lets Meta weight the model toward your highest-value customers specifically rather than treating every person in the source list equally - this is generally the strongest source available when transaction value data exists and is worth using over a plain, unweighted purchaser list whenever the option is available. A large list of low-quality signal, such as everyone who ever visited a website regardless of what they did there, produces a lookalike model that resembles that undifferentiated mix rather than your actual best customer, which is the most common way Lookalike Audiences underperform expectations.

Lookalike Audiences have narrowed in relative importance as Broad Targeting and Advantage+ Audience have improved, since both of those now often find comparably good or better prospects using live conversion signal without requiring you to maintain, size, and refresh a static seed-derived list. Lookalikes still earn their place when you have a strong, specific source signal you want to lead with deliberately - a defined high-value customer segment, a specific product line's purchasers - and want a more controllable, inspectable starting audience than an algorithm searching freely, or in accounts where broad and Advantage+ approaches haven't had enough conversion volume to perform well on their own.

In reporting, a Lookalike Audience is judged the same way as any other targeted ad set, on cost-per-result and conversion rate relative to comparable campaigns in the account, but it's worth periodically checking source list freshness alongside performance - a lookalike modeled off a source list that hasn't updated in months is drifting further from your current customer base over time, even though the audience itself was built correctly at the time. Comparing a Lookalike Audience's performance against a parallel Broad Targeting or Advantage+ Audience test in the same account is the most reliable way to know whether maintaining the lookalike is still earning its keep versus simply letting the algorithm search freely from the same seed as a signal.