How to Build a Lookalike Audience and When to Use One

Broad targeting is the best choice early on, since Facebook has no data yet to learn from. But once your pixel has collected 50+ purchase events, a lookalike audience usually starts outperforming broad — we see this shift consistently across our own campaigns, and getting the source list and refresh cadence right afterward matters as much as the initial decision to switch.

What a Lookalike Audience Actually Is

A lookalike is an audience Facebook builds by finding people whose behavior patterns resemble your existing customers (or any source list you give it). The better the quality of that source list, the better the lookalike performs.

You Need a Source List First

Before building a lookalike, your pixel’s Purchase event needs to be working — something we covered in the pixel setup section of an earlier article, and our pixel + Conversions API guide goes further into making sure those events aren’t quietly being lost. Best source lists, in order of preference:

• Purchase events (last 180 days) — the strongest signal, once you have 100+ purchases
• Add to Cart / Initiate Checkout — if purchases are still low
• Top 25% of website visitors by time spent — if the pixel is brand new

Never build a lookalike from a source list under 100 people — Facebook needs enough data to find a real pattern, or the lookalike quality suffers badly.

Choosing the Lookalike Percentage

Facebook offers lookalike sizes from 1% to 10%:

• 1% — the closest match, but a smaller audience size
• 2–5% — where we usually start, until the audience saturates
• 6–10% — more reach, but weaker match quality, and CPR usually climbs too

How This Fits With CBO

Test lookalike audiences as separate ad sets inside a CBO campaign — put a 1% lookalike, a 2-5% lookalike, and broad targeting all in the same CBO, and let Facebook decide which one deserves the most spend.

A Common Mistake

A lot of people get stuck on one lookalike source (just “Purchase”) and never try new ones. Refresh or expand the source list every 3-4 weeks — for example, build a separate lookalike from high-value purchases only (above your average order value) — this keeps performance from going stale.

Value-Based Lookalikes Go a Step Further

A standard lookalike treats every person in the source list equally, but Meta also supports value-based lookalikes, which weight the source list by purchase value rather than treating a $20 customer the same as a $200 one. This tends to find an audience that resembles your highest-value customers specifically, rather than just anyone who bought once — worth testing once there’s enough purchase history to build a meaningful value-weighted list, typically after the same 100+ purchase threshold that applies to a standard lookalike.

Always Exclude Existing Customers

A lookalike audience should almost always exclude people who are already customers — without this exclusion, some ad spend goes toward showing ads to people who’d likely convert (or already have) regardless of the ad, inflating apparent performance without actually acquiring new customers. Building a Custom Audience of existing purchasers and excluding it from every prospecting lookalike keeps the audience genuinely focused on new customer acquisition.

Lookalikes Need Periodic Rebuilding, Not Just Refreshing

Meta automatically updates a lookalike’s specific member list over time as the source audience grows, but the underlying source list itself still benefits from a deliberate rebuild every few months, not just the incremental refresh mentioned above — a full rebuild using the most recent purchase data captures shifts in who’s actually buying now versus who was buying when the lookalike was first created, which matters more for a business whose customer base evolves over time (a new product line, a shift in price point) than the routine 3-4 week refresh cycle alone accounts for.

Building Lookalikes for a New Market

A lookalike built from purchase data in one country can sometimes be applied directly to a new market being tested, but this works best when the two markets have genuinely similar buyer demographics — for a market that’s meaningfully different in income level, platform usage, or purchasing behavior, a same-market lookalike often underperforms broad targeting in that new market, since it’s optimizing for a buyer profile that may not actually translate. Testing broad first in a genuinely new market, then building a market-specific lookalike once local purchase data accumulates, is usually more reliable than assuming an existing lookalike transfers cleanly.

When Broad Targeting Is Still Better

If your pixel has fewer than 30-40 purchases, or you’re launching in a completely new product or market, broad targeting is still the better call — wait on lookalikes until you have a solid source list to build from.

Our broad vs. detailed targeting comparison covers this same broad-vs-narrow decision in more depth, including where manually layered interest targeting still earns its place even after a pixel matures enough for lookalikes to work well.