Lookalike Audiences on Meta — The Operator's Guide

Meta Lookalike Audiences let you find new customers who resemble your existing high-value customers. The mechanics changed meaningfully post-ATT; here's what works in 2026.

Meta Lookalike Audiences let advertisers find new audiences modeled on a source audience — typically existing customers, high-LTV customers, or recent purchasers. The mechanics: Meta analyzes the behavioral and demographic patterns of your source audience and finds users with similar patterns across its 3.5B+ user base.

Post-ATT (April 2021) the lookalike mechanic changed meaningfully. The source signal quality dropped (Customer Match audiences had lower match rates), the Meta-side modeling shifted to value-based-bidding signals, and Advantage+ Shopping campaigns moved toward broad targeting + creative-led optimization rather than narrow lookalike targeting. Lookalikes still work — they just sit in a different place in the modern Meta playbook.

Building strong source audiences

  • Customer Match upload — hashed customer emails / phones / addresses uploaded weekly. Match rate matters; target 70%+ match rate by ensuring email lowercase, standard formatting, and inclusion of phone and address where available.
  • High-LTV customer cohort — top 10-20% of customers by total spend, not all customers. Quality over quantity.
  • Recent purchaser cohort — last 30-180 days. More predictive than all-time purchaser cohort.
  • Value-based custom audience — upload customer list with LTV values. Meta uses values to weight the modeling.
  • Source audience size minimum — Meta recommends 1,000-5,000 source users for a usable lookalike. Below 1,000, the modeling becomes unreliable.

Lookalike percentage and how to think about it

Meta lets you build lookalikes at 1%-10% of population. 1% = closest 1% of users to source = highest similarity, smallest audience. 10% = top 10% = broader similarity, larger audience.

In 2026, the practical advice has shifted from 'narrow lookalikes only' to 'broad lookalikes plus creative diversity.' Advantage+ Shopping campaigns work better with 5-10% lookalikes (or no lookalike at all, just broad targeting) because the Andromeda model does its own audience optimization within the campaign.

Narrow 1% lookalikes are useful for non-Advantage+ campaigns and for very specific high-LTV mining, but they're no longer the default for scaled prospecting.

RGM Experts Say

The post-ATT shift in Meta's modeling means the right Customer Match audience for a lookalike is meaningfully different from pre-ATT. We refresh Customer Match weekly with LTV values, not just email lists. The value signal lets Meta's Andromeda model weight the lookalike toward users likely to be high-LTV, not just users likely to convert. This single change — weekly upload with LTV values — typically lifts blended ROAS 15-30% over basic Customer Match within 60 days.

Related guides

Sources

  1. [1]RGM internal benchmarks and operator data.