GUIDES

Meta Targeting Guide: Core, Custom, and Lookalike Audiences

Core, custom, and lookalike audiences explained — and how to progress through them naturally as your data grows, rather than jumping straight to the most advanced option.

Vineeth N.A · 8 min read · Updated July 2026

Meta Targeting Guide: Core, Custom, and Lookalike Audiences — key topics (Meta Ads Academy guide by Zephra)
Where this sits: Stage 2 (Targeting) of the Meta Ads Operating System — confirm Campaign Structure is solid first, since fragmented ad sets can make even correct targeting look broken.

Quick Answer

Start with core audiences (no data required), move to custom audiences once you have website or customer data, then lookalikes once you have 100+ conversions to build from — a natural progression tied to the tracking-coverage check in the Zephra Campaign Health Score™. Jumping straight to lookalikes without enough source data produces unpredictable, lower-quality results. Read the Andromeda update below first — Meta's 2025-2026 algorithm change has meaningfully shifted how much manual audience selection actually matters.

The Andromeda Update: Why Manual Targeting Matters Less Than It Did

Meta began rolling out a new ad-retrieval system, internally called Andromeda, in late 2024, reaching full deployment across most objectives and placements by around October 2025 and becoming the default behavior for most ecommerce and lead-gen accounts by early 2026. The practical change: rather than Meta showing an ad primarily to the audience an advertiser defines (interests, demographics, a lookalike percentage), the system now leans much more heavily on signals read directly from the ad creative itself — engagement patterns, hook performance, and completion behavior — to decide who's likely to respond, and expands delivery accordingly. Meta has described Andromeda as infrastructure built specifically to strengthen Advantage+ automation.

The reported practical effects: manually-built lookalike audiences, once the default scaling tool covered later in this guide, now frequently underperform broad targeting paired with a genuinely diverse creative library, since narrow audience definitions can restrict the volume of signal the algorithm needs to work with. Meta's own documentation now treats detailed targeting inputs as a starting suggestion the system can expand beyond, not a hard boundary.

This doesn't make the audience types below irrelevant — a custom audience of actual purchasers is still valuable as a signal source and for exclusions, and core targeting is still a reasonable way to give a brand-new account without any data a starting point. What's changed is the emphasis: creative diversity (see Creative Testing Framework) and clean tracking (see Facebook Pixel Guide) now do more of the targeting work than manual audience-building does, and accounts still running the older playbook — many small, narrowly-targeted ad sets — are working against the algorithm rather than with it. See Campaign Structure Guide for what a consolidated structure looks like in practice.

The Three Audience Types

TypeBuilt fromBest for
CoreLocation, age, interests, behaviorsFirst campaigns, no existing data
CustomYour own data (visitors, customers, engagers)Retargeting, once you have traffic
LookalikeA custom audience Meta expands fromScaling once you have 100+ conversions

Lookalike Source Quality Matters More Than Size

Lookalike source qualityResult
Built from purchasers/leadsHigher quality, more relevant expansion
Built from website visitors onlyLower quality — visitors include many non-buyers, diluting the pattern
Built from a very small (under 100) source audienceUnpredictable, often low-quality expansion regardless of source type

A lookalike is only as good as the pattern Meta can learn from its source — a lookalike built from actual purchasers will typically outperform one built from generic website visitors of the same size, since the underlying behavioral pattern is more specific and relevant.

A Natural Progression as Data Grows

  1. Start with core targeting — location, age, interests roughly matching your ideal customer.
  2. Install Pixel and Conversions API immediately, so custom audience data starts accumulating from day one.
  3. Build a custom audience from website visitors or engagers once meaningful traffic exists.
  4. Build a lookalike once you have 100+ conversions, ideally sourced from purchasers or qualified leads rather than general visitors.
  5. Test lookalike vs Advantage+ once conversion volume is high enough for both to have a fair chance.

Variations by Business Type

Business typeTargeting approach
EcommerceLookalike from purchasers, once volume allows; retargeting from cart/checkout events
Lead generationLookalike from qualified leads specifically, not just form-fills
B2BCustom audiences from CRM-uploaded customer lists often outperform generic core targeting

Case Study

An ecommerce brand built its first lookalike audience from a custom audience of all website visitors, assuming any conversion data was better than none, at a stage where they had barely 100 total site visitors logged. The resulting lookalike performed inconsistently, with cost per result swinging widely week to week. Waiting to accumulate over 300 actual purchase conversions, then rebuilding the lookalike from that purchaser-specific custom audience instead, produced a meaningfully more stable and better-performing lookalike within the first month of the new approach.

Decision Matrix

SituationRecommendation
Brand new account, no data yetCore audience targeting
Traffic exists but under 100 conversionsCustom audience retargeting; hold off on lookalikes
100+ conversions availableBuild a lookalike from purchasers/qualified leads specifically

Common Mistakes

  1. Jumping to lookalikes before there's enough source data.
  2. Building lookalikes from website visitors instead of actual purchasers/leads.
  3. Targeting too narrow a core audience, starving the algorithm of data.
  4. Never revisiting or refreshing a lookalike as more conversion data accumulates.

Troubleshooting

Lookalike performance is inconsistent: check the source audience size and quality — under 100 conversions or a visitor-only source are the most common causes.

Core audience targeting feels too broad or too narrow: broad, well-defined interest categories usually outperform overly narrow, hyper-specific combinations that limit reach unnecessarily.

Checklist

☐ Pixel and Conversions API installed from day one
☐ Custom audiences built from actual site/engagement data
☐ Lookalikes built only once 100+ conversions exist
☐ Lookalike source is purchasers/qualified leads, not generic visitors
☐ Lookalikes refreshed periodically as conversion data grows

FAQ

What is the difference between custom and lookalike audiences?

Custom is built from your own data; lookalike is Meta finding new people who resemble a custom audience you provide.

When should I use a lookalike audience?

Once you have at least 100 conversions to build the source audience from.

Should I use core targeting or Advantage+?

Core targeting for new accounts with no data; Advantage+ often performs comparably once enough conversion history exists.

HOW ZEPHRA HELPS

Zephra builds and tests audiences as your data grows.

Progresses from core to custom to lookalike targeting automatically as conversion volume allows, sourcing lookalikes from purchasers rather than generic visitors.

Start Free Audit →

Sources & Further Reading

Figures and platform mechanics referenced in this guide are cross-checked against the above as of publication; ad platform thresholds and benchmarks change over time, so confirm current figures directly with the source before making budget decisions.