Meta Ads Lookalike Audience: What It Is and How to Build One That Actually Works

What Is a Lookalike Audience

A Lookalike audience is created by Meta's algorithm based on a source group you define. Meta analyses the demographic profile, behaviours and interests of people in your source and finds similar users on Facebook and Instagram — people who don't know you exist but share the profile of your best customers.

The logic: instead of manually selecting interests like "fitness enthusiasts aged 25–45", you give Meta a list of real people who already bought from you, and let the algorithm identify the patterns it can't articulate but can detect.

Source Quality Is Everything

The Lookalike is only as good as the source. Best sources in order of quality:

1. Repeat buyers (2+ purchases) — highest quality signal, these are your actual best customers
2. Top 25% customers by LTV — if you have CRM data, use it. A Lookalike based on high-LTV customers finds more high-LTV customers
3. All buyers (Purchase event) — good if you have minimum 1,000 buyers in the source. Below 1,000 the signal is weak
4. Add-to-cart visitors — high intent but unconverted. Useful if you don't have enough buyers
5. Facebook page fans — avoid. Fans are not customers; many liked the page from contests or paid boosts

Audience Size: 1%, 2% or More?

Lookalike size is expressed as a percentage of the selected country's population. In Romania:

• 1% ≈ 80,000–100,000 people — most similar to source, most expensive to reach
• 2% ≈ 160,000–200,000 — good balance between similarity and volume
• 5% ≈ 400,000 — less similar but cheaper per impression
• 10% ≈ 800,000 — almost a broad audience, similarity is diluted

Our recommendation for the Romanian market: start with 1–2% and test. If frequency exceeds 3–4 in 7 days, expand to 3–5%.

When Lookalike Works and When It Doesn't

Works well when: you have minimum 500–1,000 people in the source (ideally 1,000+), the source is homogeneous (all buyers, not a mix of visitors and buyers), and your product has broad demand.

Doesn't work well when: the source has fewer than 100 people, the source is polluted (newsletter subscribers who never purchased, page fans), or your product has an extremely specific ICP that Meta can't detect from online behaviours.

Why Lookalike Is Harder to Manage Than It Looks

Meta has introduced Advantage+ Audiences, which combines lookalikes with algorithm-controlled targeting. The dilemma is real: if you give the algorithm full control, you don't know who you're targeting and can't interpret results. If you maintain manual Lookalike, you risk limiting distribution at a moment when the platform sees signals you don't have.

The right answer depends on data from your specific account, the size of source audiences, and campaign history. There's no universally optimal configuration. And properly testing both approaches requires sufficient budget and several weeks of clean data, with no parallel creative changes that would distort the conclusion.