As of June 1, 2026, your Meta Ads targeting settings are suggestions — not instructions.
Andromeda — Meta's AI retrieval engine, hardware-deployed across NVIDIA Grace Hopper superchips — is now the default for all new campaigns globally. It does not honor manual audience boundaries. It treats them as soft inputs and overrides them when its own conversion prediction is more confident than your targeting specification.
This is a COMPLETE paradigm shift for how Meta delivery works.
What Actually Changed
The core shift is in the Retrieval Stage — the initial filtering step that selects candidate ads from a pool of tens of millions before the auction runs. Andromeda converts every ad asset (visuals, audio, copy) into high-dimensional vector embeddings and matches them to users based on real-time semantic and computer vision signals.
Key findings driving the new mechanics:
- Soft suggestion override: Manual interest targeting, demographic filters, and lookalike audiences are advisory inputs. Andromeda ignores them when its own prediction confidence is higher. There is no opt-out for new campaigns.
- Entity ID clustering: Andromeda groups visually and semantically similar ads under a single Entity ID. Testing button colors or minor headline variants is not testing — it is giving Andromeda the same asset twice. You need conceptually distinct creative to register as distinct delivery signals.
- Targeting-as-creative: The audience you reach is now determined by what your ad looks like and says. Your creative IS your targeting layer.
Timeline
| Date | Event |
|---|---|
| October 2025 | Andromeda hardware deployment completed globally |
| January 15, 2026 | Thousands of detailed interest categories and strict behavioral exclusions removed |
| June 1, 2026 | Creative-led targeting becomes the default for all new campaigns |
The 2026 Creative Playbook
This is not a marginal shift in targeting levers — the entire optimization surface has moved.
| Old lever | New lever |
|---|---|
| Audience stacking (interest + demo + lookalike) | 15–20 conceptually distinct creatives per ad set |
| A/B headline or color testing | Different persona or emotional hook per creative concept |
| Broad vs. narrow audience A/B | One campaign, one ad set — consolidate for signal quality |
| Manual bid adjustments by segment | Event Match Quality (EMQ) score maintenance at or above 7.0 |
The 40% similarity rule: If your running ads are more than ~40% visually similar, Andromeda consolidates their data under one Entity ID and suppresses newer variants. Monitor creative similarity actively — not just creative count.
Root cause: Andromeda exhausts the most responsive audience segments quickly — winning creatives now fatigue in 2–3 weeks, down from 6+ weeks previously. High-velocity creative production is not a competitive advantage anymore. It is the baseline operating requirement.
What You Must Do Now
-
Audit your creative library for conceptual diversity. Count distinct personas, hooks, and formats — not variant counts. If 10 of your 12 ads share the same visual concept, you effectively have 1 ad running.
-
Implement or repair CAPI with EMQ at or above 7.0. With targeting automated, the pixel/CAPI feedback loop is your only control over what Andromeda learns from your conversions. EMQ below 7.0 means the engine is flying blind on your conversion signal.
-
Consolidate campaign structure. Fragmented budgets starve the AI of conversion data. Fewer, larger ad sets win under Andromeda — the engine needs volume to validate its creative-to-user matching.
-
Plan for faster creative refresh cycles. 2–3 week creative lifespans mean production cadence is an operational requirement, not a creative preference.
| Action | Impact | Effort | Owner |
|---|---|---|---|
| Audit creative library for conceptual diversity | Critical | Low | Creative Strategy |
| Repair CAPI to EMQ at or above 7.0 | Critical | Medium | Engineering |
| Consolidate to fewer, larger ad sets | High | Low | Media Buyer |
| Establish bi-weekly creative refresh cadence | High | High | Creative |
| Remove low-similarity variant testing | Medium | Low | Media Buyer |
The "media buyer" role at the targeting layer has been automated. The value-add is entirely in creative strategy and signal engineering now.
Big difference.
Founder CEO/CTO
