Lookalike Audience

A Lookalike Audience is a targeting option on Meta (and other platforms) that uses machine learning to find users who statistically resemble a seed audience - your existing customers, high-value purchasers, or email subscribers - but who have not yet interacted with your brand.

The idea is that people who look like your best customers are more likely to become customers themselves.

How they work

You provide Meta with a seed audience: a list of customer emails, a Custom Audience of purchasers, or a customer list with associated spend values. Meta analyzes the shared characteristics of people in that seed - demographics, interests, behaviors, engagement patterns - and builds a statistical model to find similar profiles across its user base in a given country.

You then set a percentage representing how broadly to match:

• 1% LAL: most similar to the seed, smallest audience, typically the best conversion rates
• 5% LAL: broader match, more reach, lower precision
• 10% LAL: broad matching, useful for scale when 1-5% audiences are exhausted

The percentage is relative to the target country’s population. A 1% US lookalike is roughly 2.2 million people.

How to create one

In Meta Ads Manager:

1. Go to Audiences in the Business Manager.
2. Click Create Audience > Lookalike Audience.
3. Choose your source: a Custom Audience (customer list, purchaser pixel event, or website visitors), or a Facebook Page engagement audience.
4. Select your target country or region.
5. Set your percentage (1-10%).
6. The audience builds in 1-6 hours and shows as “ready” when available.

For best results, your seed audience should contain at least 1,000 people in the same country as the lookalike. Smaller seeds produce less reliable models. Seeds of 10,000+ people generally produce the most consistent results.

Value-based lookalikes

Standard lookalikes treat all customers in the seed equally - a customer who spent $40 once is weighted the same as a customer who spent $800 across six orders.

Value-based lookalikes fix this. You upload a customer list with a value column (total lifetime spend, for example), and Meta builds a model weighted toward finding people who resemble your highest-value customers specifically.

The practical difference: a standard purchase lookalike finds people who look like anyone who ever bought. A value-based lookalike finds people who look like customers who bought repeatedly at high AOV. For brands with meaningful LTV spread across customers, value-based lookalikes are worth the extra setup.

Post-iOS 14: are lookalikes still effective?

This is the honest answer: it depends on your pixel signal quality, and the gap between lookalikes and broad targeting has closed significantly.

iOS 14 (2021) reduced Meta’s ability to track conversions from Apple devices, which degraded the purchase signal available to build lookalike models. Brands that had strong lookalike audiences built on years of purchase data saw performance drop. Brands building lookalikes on degraded pixel data saw even worse results.

Meta’s response was to improve its broad targeting algorithm. Advantage+ campaigns, which use no manual audience constraints, now perform comparably to or better than hand-crafted 1% lookalikes for many advertisers. Brands that moved to Advantage+ broad targeting saw an average 22% higher ROAS in Meta’s internal data, though the methodology behind that claim is not independently verified.

Practitioner consensus as of 2025: for accounts with strong purchase event volume (500+ events/week), broad targeting inside Advantage+ campaigns often matches 1% lookalike performance and sometimes beats it. For accounts with weak pixel history or small email lists, lookalikes still provide useful signal above broad.

The practical test: run a 1% lookalike ad set against a broad-targeting ad set with the same creative, same budget, for 14 days. Let the data tell you which one your account responds to. Do not assume either is better - this varies by brand, category, and pixel quality.

Seed audience quality matters more than size

A common mistake is building lookalikes from the largest available audience regardless of quality. A lookalike seeded from all website visitors will find people who look like browsers. A lookalike seeded from customers with 3+ purchases at $100+ AOV will find people who look like high-value repeat buyers.

The seed audience defines the problem you’re asking Meta to solve. Garbage in, garbage out.

Best seed audiences, roughly in order of quality:

• High-LTV customers (top 20% by lifetime spend)
• All purchasers from the last 180 days
• Add-to-cart events from the last 60 days
• Email subscribers who have opened in the last 90 days
• All website visitors in the last 180 days

Avoid building production lookalikes from audiences smaller than 500 people, or from audiences with mixed intent (like all website visitors, which includes everyone from buyers to five-second bounces).

Stacking lookalikes: usually not worth it

Some advertisers run multiple lookalike percentages simultaneously: 1%, 2-3%, and 4-5% in separate ad sets. The theory is that each tier catches a different audience segment.

In practice, the audiences overlap significantly at lower percentages and the incremental reach at 2-5% over the 1% audience is often not much better than broad targeting. The overhead of managing multiple lookalike ad sets usually isn’t justified unless you’re running at high scale and have already exhausted your 1% audience.

When lookalikes still have a clear advantage

Broad targeting works because Meta’s algorithm uses its platform data to find buyers. But there is one type of signal broad targeting cannot replicate: your CRM data.

Meta cannot see your email list, your purchase history, or your subscriber engagement unless you upload it. A lookalike seeded from your email list (people who opted in to your brand, not just visited your site) gives Meta information it cannot derive from pixel data alone. This remains a genuine edge for lookalikes in a broad-targeting world.

Cases where lookalikes still outperform broad:

• New ad accounts with weak pixel history
• International market expansion into countries where Meta has limited purchase data for your category
• Niche products where the buyer profile is distinct and your CRM data captures it clearly
• Brands with large, high-quality email lists of engaged buyers

Frequently asked questions

What’s the minimum audience size for a lookalike seed? Meta recommends at least 100 people in the same country, but practitioners consistently find that seeds under 1,000 produce unreliable models. For purchase-based lookalikes, aim for at least 1,000-5,000 purchasers. Value-based lookalikes work best with 10,000+ customers.

Should I use 1% or 5% lookalikes? Start with 1% for prospecting. The 1% is the most precise match to your seed. Expand to 2-5% only after you’ve exhausted 1% reach or want more volume at the cost of some precision. Running a 5% LAL alongside a 1% LAL usually just inflates frequency against the same users.

Should I exclude my existing customers from lookalike campaigns? Yes. Create a Custom Audience of all purchasers and exclude it from your lookalike targeting. Otherwise you pay prospecting CPMs to reach people who already bought from you - and your ROAS looks artificially good because it includes repurchases that would have happened without the ad.