Lookalike Audiences: finding new people similar to existing customers

Custom Audiences (part 6) reach people with a known, direct relationship to the business already. A Lookalike Audience does something different: it finds new people, who've never interacted with the business at all, but who share real characteristics with an existing audience — Meta's own machine learning identifying the underlying pattern, not just the surface-level interests a Core Audience relies on.

How it actually works

Source: Bright Leaf Coffee's "Purchased in Last 180 Days" Custom Audience
Lookalike: 1% similarity, United States

Meta analyzes the source audience — real customers, in this case — for shared patterns across potentially hundreds of signals (not just declared interests, but broader behavioral and demographic patterns Meta's own data reveals), then finds new people across the platform who match that pattern closely. This is a genuinely more sophisticated targeting mechanism than manually guessing which interests might correlate with being a good customer.

The similarity percentage: a real precision-vs-reach tradeoff

1% Lookalike:  smallest, most closely matched to the source audience
5% Lookalike:  larger reach, less tightly matched
10% Lookalike:  largest reach, loosest match

A 1% Lookalike represents roughly the top 1% of people, by similarity, in the target country — the closest match, but the smallest available pool. Moving to 5% or 10% trades some of that precision for meaningfully more reach. For Bright Leaf Coffee's first Lookalike test, starting at 1% is the right instinct — the tightest, most likely-to-perform match — with room to test broader percentages once real performance data exists to justify expanding.

Source audience quality matters more than any other setting

Weak source:  "All Website Visitors" — includes plenty of people who never bought anything
Strong source: "Purchased 2+ Times" — Bright Leaf Coffee's actual best, repeat customers

A Lookalike is only as good as what it's modeled from — building one from a source audience that includes a lot of low-value or non-converting people means the "similar" people it finds share whatever pattern that source actually represents, not necessarily "good customer." A source audience of Bright Leaf Coffee's repeat subscribers specifically — its highest-value customers, not just anyone who ever visited the site — produces a meaningfully more valuable Lookalike than a broad, low-quality source would.

Minimum source audience size

Meta requires a minimum source audience size (typically at least 100 people, though a source of several hundred to a few thousand tends to produce a more reliable pattern) before it can build a Lookalike at all — a new business with very few existing customers may need to build from a broader source (all website visitors, rather than repeat purchasers specifically) until enough real purchase data accumulates to build a higher-quality one later.

A real structure combining Lookalike with the rest of the account

Ad Set: Lookalike 1% — Purchasers (180 days)     → cold acquisition, high-quality match
Ad Set: Core Audience — Coffee Enthusiasts        → cold acquisition, broader interest-based
Ad Set: Retargeting — Website Visitors            → warm, from part 6

A mature account typically runs Lookalike and Core Audience cold-acquisition ad sets side by side, letting real performance data over time reveal which cold-audience approach is actually delivering better results for this specific business — they're not mutually exclusive, and testing both against each other is itself a legitimate, common strategy.

The three core audience types are covered — Core, Custom, and Lookalike. Next: the Meta Pixel and Conversions API, the tracking foundation every one of these has quietly depended on.