The essentials
- Definition: a lookalike audience is a set of users the platform selects because they resemble, in their characteristics and behaviour, a source audience you provide (customers, converters, visitors).
- The rule: the quality of the source makes the quality of the lookalike; 1,000 to 5,000 recent, high-value people are worth more than 50,000 random visitors.
- Sizes: 1 to 3% of the country's population for precision, 5 to 10% for volume.
- 2026 : Meta pushes Advantage+ (automatic targeting); lookalikes remain useful as a signal and on LinkedIn, TikTok and Google (lookalike audiences reintroduced in 2024-2025).
The lookalike audiences, or lookalike audiences, make it possible to reach strangers who resemble your best customers. It is one of the mechanisms behind the success of social advertising: rather than guessing at interests, you give the platform a sample of people who converted and let it find their counterparts. Their place has changed with automatic targeting (Advantage+ at Meta), but the principle remains central. This article defines the mechanism, describes the sources that work, the sizes and the use by platform. It complements optimising your campaigns with lookalike audiences.
How a lookalike audience is built
- You provide a source: a customer list (hashed emails), pixel converters, visitors to a page, people who interacted with your posts or your videos.
- The platform analyses the source: demographic characteristics, interests, behaviours, devices, connections, across thousands of signals that only it knows.
- It ranks its users by similarity with the source and keeps the closest ones, in the proportion you choose (1%, 5%, 10% of a country's population).
- It updates the audience as the source evolves (every 3 to 7 days).
The platform does not tell you which criteria it kept; you judge on the result. The mechanism is close to that of Google Ads custom audiences, with different signals.
The sources that produce the best results
| Source | Quality | Minimum size | Note |
|---|---|---|---|
| High-value customers (CRM list of the top 20% of customers) | Excellent | 1,000 matched people | The best source; email match rates of 40 to 70% |
| Buyers or converters from the last 90 days (pixel) | Very good | 500 to 1,000 events | Freshness; value-based if the event carries a value |
| Qualified leads (form + CRM qualification) | Very good | 500 | Better than all leads |
| Add to cart, checkout starts | Good | 1 000 | Intent without purchase |
| Engaged visitors (time spent, key pages) | Medium | 1,000 to 5,000 | Filter out bounces |
| All site visitors | Poor | 5,000 and above | Too heterogeneous; to be avoided as a sole source |
| Interactions with the page or the videos (75% viewed) | Medium | 1 000 | Useful when the site has little traffic |
| Page followers | Poor | A mix of customers, curious visitors and competitors |
Rule: a small, homogeneous source (your real customers) beats a large, vague one. Value-based sources (Meta's “value-based lookalike audience”) weight by amount spent and improve the result further.
Sizes
| Percentage | Size in France (Meta) | Use |
|---|---|---|
| 1 % | Around 400,000 to 500,000 people | Maximum precision; lowest cost per result; runs out quickly on a small budget |
| 2 to 3% | 1 to 1.5 million | The usual compromise for an SME |
| 5 to 10% | 2 to 5 million | Volume, awareness, large budgets; diluted precision |
Create several bands (1%, 1 to 3%, 3 to 5%) excluding the previous ones to compare their cost per result, and always exclude the source itself and existing customers from the acquisition campaign.
By platform in 2026
| Platform | Name | Specifics |
|---|---|---|
| Meta (Facebook, Instagram) | Lookalike audiences; Advantage+ audiences | Meta recommends Advantage+ (automatic targeting) where lookalikes serve as a “suggestion”; manual lookalikes remain available and useful for keeping control; see Meta Ads Manager |
| Predictive audiences (replacing lookalikes since 2024) | Built from a list or a conversion, combined with professional attributes | |
| TikTok | Lookalike audiences | Effective on pixel events and video interactions; narrow, balanced and broad sizes |
| Google Ads | Similar segments (removed in 2023, reintroduced as “lookalike audiences” in Demand Gen and as signals in Performance Max) | Work mainly as signals for the automation |
| Pinterest, Snapchat | Actalike, Lookalike | Same logic |
Lookalikes and automatic targeting
Since 2023-2024, Meta has pushed Advantage+ Audience and Advantage+ Shopping, which choose the audience from conversions; lookalikes are suggestions there that the algorithm can go beyond. On accounts with many conversions (more than 50 a week), the automatic option often does as well or better. On low-volume SME accounts, manual lookalikes built on a good source keep the advantage: they give the algorithm a starting point it does not have. Common practice: test in parallel an Advantage+ ad set with the lookalike as a suggestion and a manual 1 to 3% lookalike ad set, over two weeks. Ad set segmentation is covered in audience segmentation in SMA.
Measuring and maintaining
- Compare the cost per result of each band and each source; cut those above target.
- Refresh the source: a customer list from 2022 no longer describes your 2026 customers; dynamic sources (pixel, connected CRM) are preferable.
- Watch frequency: a 1% lookalike wears out in a few weeks on a high daily budget.
- Exclude converters so you do not pay to reach customers.
- Read the results in GA4 and the CRM, not only in the platform; see the SMA KPIs.
How GreenRed helps
Rather than juggling several tools, GreenRed's social media tracking brings these metrics together in a single dashboard, compares them over time and tells you which actions come first. You can try it free, with no card, from the Pricing.
Frequently asked questions
What is the best source for a lookalike audience?
A list of your best customers (the top 20% by value, 1,000 matched people at minimum) or buyers from the last 90 days via the pixel. A small, homogeneous source produces a more precise lookalike than a broad, heterogeneous one such as all site visitors or page followers.
What percentage should you choose for a lookalike?
1% for precision and the lowest cost per result, 2 to 3% as a compromise for an SME, 5 to 10% for volume at the price of diluted precision. Create several exclusive bands and compare their cost per result; cut those that exceed your target.
Are lookalikes still useful with Advantage+?
Yes, differently: on Meta, Advantage+ Audience uses the lookalike as a suggestion and can broaden it; on accounts with a high volume of conversions, the automatic option often does just as well. On low-volume SME accounts, a manual lookalike on a good source keeps the advantage. Test both in parallel.
How many people do you need in the source?
Meta requires a minimum of 100 people in a single country, but quality starts at around 1,000 matched people; beyond 10,000, the marginal gain is small. For an email list, expect a match rate of 40 to 70%, so import 2,000 to 3,000 contacts to obtain 1,000 matches.