Lookalike audiences explained

A lookalike audience is an audience built by an ad platform to resemble your existing customers. You provide a list (a seed) and the platform finds people in its user base who share similar characteristics, behaviours, or signals. The goal is to find new prospects who look like your best customers, rather than targeting by interest or demographic alone.

The logic is sound. The problem is the seed.

Most lookalike audiences are built off first-party seeds that are thinner, noisier, or less representative than the advertiser realises. When the seed is poor, the resulting audience inherits its flaws. You end up finding more people who look like your average customer rather than more people who look like your best customer. The distinction matters more than most media plans acknowledge.

What a lookalike audience is

A lookalike audience is a targeting method used across paid social and paid search platforms. You upload a list of existing customers or website visitors (the seed) and the platform uses its own data to find users who share similar patterns.

What those patterns are depends on the platform. Meta uses behavioural signals from Facebook and Instagram: content engagement, purchase history, app activity. Google uses browsing behaviour, search patterns, and intent signals across its properties. LinkedIn uses job title, company size, seniority, and professional activity.

The platform does not tell you exactly how the match is made. What you can control is the quality and composition of the seed you provide.

Lookalike audiences by platform

Meta lookalike audiences (Facebook and Instagram). Meta builds lookalikes from a Custom Audience: a customer list, a website pixel audience, or an engagement audience. You can set the audience size from 1% (closest match) to 10% (broader reach) of a target country's population. Smaller percentages are more precise; larger ones trade precision for scale. Meta's matching works best when the seed contains enough signal: name, email, phone number, and ideally purchase value or purchase frequency so the model can weight toward high-value behaviour.

Google lookalike audiences. Google's equivalent is Customer Match, which allows you to upload a customer list and target or exclude those users, then expand to similar audiences. Google's signals draw from Search, YouTube, Gmail, and Display behaviour. Similar Audiences have been partially sunset in favour of Google's AI-driven targeting tools, but the principle remains: a high-quality first-party seed improves the model's output.

LinkedIn lookalike audiences. LinkedIn's Matched Audiences allow you to upload a contact list and build a lookalike from it. Given LinkedIn's professional data, these audiences are most useful for B2B targeting: finding companies or roles that resemble your existing customer base. The volume is smaller than Meta or Google but the match precision on professional attributes is higher.

The seed quality problem

A lookalike audience is only as good as the seed it is built from. This is where most campaigns underperform.

The typical seed is an email list or a pixel audience: all customers, or all purchasers, with no weighting by value. The platform then finds users who resemble the average of that list. If your customer base is mixed, containing both high-value repeat buyers and one-off discount purchasers, your lookalike audience will reflect that mix, and a significant proportion of the audience will have the characteristics of the lower-value segment.

Building a seed from your top 20% of customers by lifetime value produces a materially different audience. The platform is now matching against the characteristics of people who buy repeatedly and spend more, rather than the characteristics of everyone who ever converted. The audience is smaller to start with, but the match quality is higher and the resulting media spend is more efficient.

The same principle applies to B2B seeds. A seed built from logos that churned within six months produces a different lookalike than a seed built from accounts that expanded. If your CRM contains both, it matters which one you upload.

What makes a good seed

A good lookalike seed is not your largest audience. It is your most representative audience for the outcome you want.

For B2C: segment your customer list before uploading. Build separate seeds for high-lifetime-value customers, repeat buyers in a specific category, and recently reactivated lapsed customers. Each produces a different lookalike for a different campaign objective.

For B2B: filter to accounts with the characteristics of your best customers: company size, sector, seniority of buyer, time to conversion, contract value. A seed of fifty well-matched accounts will outperform a seed of five hundred mixed ones.

For both: include as many matching fields as the platform accepts. Email and name are the baseline. Adding phone number, postal code, and purchase value improves the match rate and gives the algorithm more signal to work with.

The platforms do the matching. Your job is to define, with precision, who you want them to match against. That definition starts with your own customer data, and how well it is structured.