What is a lookalike audience?

A lookalike can only extend a definition of the customer that someone already wrote down in the seed. The platform has no view of your best customers beyond the people you hand it, so it looks for more people who share traits with that group. Seed it with closed-won customers and the audience leans toward buyers. Seed it with every site visitor and it reproduces whatever mix of buyers and bystanders happened to browse, because the model scales the seed faithfully, flaws included. Skipping the ICP(opens in a new tab) work and expecting the platform to supply it only moves that decision into the seed, and the cost per qualified lead eventually shows where it landed.

The seed decides what the platform finds

Meta describes the process in plain terms: it takes the source audience, looks at demographics, interests, and behaviors, and reaches people who share them. Its help pages ask for a source of at least 100 people, recommend 1,000 to 5,000, and note that a list of best customers usually beats a list of all customers. People in the source are excluded from the resulting lookalike, so it reaches new people only.

That makes seed selection a marketing-ops decision made in CRM terms. A seed built from closed-won deals or renewed customers carries a definition of value, and refreshing it as deals close keeps the model pointed at the customers you have now instead of the ones you had at launch.

What the size setting trades away

Meta lets you pick a percentage when you create a lookalike. A smaller percentage matches the seed more closely and reaches fewer people, and a larger one reaches more people with a looser resemblance. Neither end is correct by default. A narrow audience suits a tight, high-value segment with a small addressable market, and a broad one suits a team that needs volume and can tolerate waste. Meta no longer asks for a location when the lookalike is created. The ad set's own location settings decide where it is delivered, and the estimated size shows up there.

Which platforms still offer lookalikes

Fewer than most guides imply. Google Ads stopped generating similar audiences in May 2023 and replaced them with optimized targeting, a form of AI-assisted targeting(opens in a new tab) that starts from landing page keywords and creative assets and treats your segments as hints. LinkedIn discontinued lookalike audiences in February 2024. Existing ones stopped refreshing, and LinkedIn pointed advertisers to predictive audiences, built from sources like contact lists, company lists, conversions, and lead gen forms, and to Audience Expansion, which leans on professional demographics. Meta remains the main platform with a lookalike control you set yourself, so any guide that treats the three networks as interchangeable is out of date.

Platform lookalikes versus lookalike accounts

Platform lookalikes model people. They look at the individuals in your seed, find similar individuals, and never return companies. A lookalike account is a different idea: a company that resembles your closed-won customers on industry, size, and tech stack. That modeling comes from a data vendor or an internal model, and it grows a target account list(opens in a new tab) before it grows any ad audience. A team that expects a platform lookalike to hand back a list of accounts will not get one.

Account-level versus contact-level lookalikes

A model learns from the roles the seed contains. A seed made of the champions who signed up for trials gives it champions to imitate and says nothing about the finance approver or security reviewer who also had to agree. A seed of named contacts from every role on won deals gives it more of the buying committee to learn from. It also gives a team something to measure against, since the same contacts can be uploaded as a custom audience(opens in a new tab) of named buyers and the two compared on cost and response.

Where lookalike audiences fit in an ABM program

A lookalike sits outside the target account list by design, since its job is to reach people the list doesn't already name. That makes it a prospecting layer, and the accounts a team has chosen to pursue get their coverage from named-account campaigns. Teams that run one usually exclude current customers and the accounts those campaigns cover. The audience is a pool of people with no company attached, so whether they belong to accounts that fit the profile has to be checked after delivery.