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
Quick Answer
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
| Type | Built from | Best for |
|---|---|---|
| Core | Location, age, interests, behaviors | First campaigns, no existing data |
| Custom | Your own data (visitors, customers, engagers) | Retargeting, once you have traffic |
| Lookalike | A custom audience Meta expands from | Scaling once you have 100+ conversions |
Lookalike Source Quality Matters More Than Size
| Lookalike source quality | Result |
|---|---|
| Built from purchasers/leads | Higher quality, more relevant expansion |
| Built from website visitors only | Lower quality — visitors include many non-buyers, diluting the pattern |
| Built from a very small (under 100) source audience | Unpredictable, 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
- Start with core targeting — location, age, interests roughly matching your ideal customer.
- Install Pixel and Conversions API immediately, so custom audience data starts accumulating from day one.
- Build a custom audience from website visitors or engagers once meaningful traffic exists.
- Build a lookalike once you have 100+ conversions, ideally sourced from purchasers or qualified leads rather than general visitors.
- Test lookalike vs Advantage+ once conversion volume is high enough for both to have a fair chance.
Variations by Business Type
| Business type | Targeting approach |
|---|---|
| Ecommerce | Lookalike from purchasers, once volume allows; retargeting from cart/checkout events |
| Lead generation | Lookalike from qualified leads specifically, not just form-fills |
| B2B | Custom 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
| Situation | Recommendation |
|---|---|
| Brand new account, no data yet | Core audience targeting |
| Traffic exists but under 100 conversions | Custom audience retargeting; hold off on lookalikes |
| 100+ conversions available | Build a lookalike from purchasers/qualified leads specifically |
Common Mistakes
- Jumping to lookalikes before there's enough source data.
- Building lookalikes from website visitors instead of actual purchasers/leads.
- Targeting too narrow a core audience, starving the algorithm of data.
- 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.
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
- Meta Business Help Center — About the Learning Phase — Meta's own explanation of what the ad set learning phase is and why it exists.
- Meta Business Help Center — Significant Edits and Learning Phase — Meta's documentation on which account changes restart the learning phase.
- Meta Business Help Center — About Learning Limited — Meta's documentation on the "Learning Limited" delivery status and what causes it.
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.