The way Meta's ad algorithm decides who to show your ad to has fundamentally changed.
On June 3, 2026, Meta announced the global rollout of Advantage+ AI Chat Signals — what it built internally under the name Project Andromeda. The system incorporates anonymized conversational data from over 1 billion monthly Meta AI users into the Advantage+ delivery algorithm.
The shift is from behavioral history to predictive intent. Meta's algorithm is no longer primarily asking "who did something like this before?" It's asking "who is actively thinking about buying this right now?"
What Project Andromeda Actually Changes
The old model: Advantage+ delivery was built on behavioral signals — page likes, ad interactions, purchase history, Lookalike seeds. Who you are, what you've done.
The new model: The algorithm processes anonymized natural language queries from Meta AI conversations. When a user asks Meta AI to "compare running shoes under $150," that intent signal — processed as a high-level intent cluster, not as an individual message log — enters the delivery pool. Advantage+ identifies that someone is in an active research-and-purchase window and bids accordingly.
Meta's Andromeda engine generates real-time conversion probability scores for every impression with a claimed 87% predictive accuracy as of Q2 2026.
The result is "Creative as Targeting" — your ad's visual and semantic elements are read by the system to find their best match within the chat-derived intent clusters. Audience selection has become a function of creative quality, not targeting precision.
Performance Benchmarks
Early data across Advantage+ Shopping Campaigns:
- 18–22% ROAS lift vs. pre-Andromeda baselines
- 17% CPA reduction from improved mid-funnel alignment
- CPMs for high-intent chat clusters are significantly higher than broad browsing audiences — efficiency gains come from better conversion rates, not lower media costs
Availability and Restrictions
Global rollout as of June 3, 2026. EU, UK, and South Korea are currently restricted due to GDPR/DMA — conversational signals are limited to first-party business interactions only. Standard behavioral targeting remains the primary delivery input in those markets.
What This Means for How You Run Meta Ads
Manual interest targeting now actively hurts you. Adding interest restrictions to your Advantage+ campaigns constrains the algorithm's access to conversational intent pools. A campaign targeted at "Running → Fitness → Sports" can't reach someone who asked Meta AI about treadmills last night but has never engaged with fitness content. Remove interest targeting from Advantage+ campaigns and let the creative signal do the matching.
Creative volume is the primary performance lever. Meta recommends 15–50 active creative assets per campaign. The Andromeda system needs creative variation to test which visual and semantic elements resonate with different intent clusters. A campaign with 3 creatives gives the algorithm almost nothing to work with.
CAPI is mandatory, not optional. Conversions API provides the feedback loop the algorithm needs to verify whether its intent predictions resulted in actual sales. Without CAPI: the model can't learn which signals converted, performance caps, attribution degrades.
Prioritize Event Match Quality (EMQ) above 8.0. Your CAPI events need to match to Meta users with high confidence. Below that threshold, the predictive models run on incomplete feedback.
The Paradigm Shift
Andromeda completes a direction Meta has been moving for three years: removing advertiser control over audience selection and replacing it with AI delivery. The trade-off is real — better performance in aggregate, in exchange for visibility into who is seeing your ads.
For most Advantage+ Shopping advertisers, the data says the trade is worth making.
But the implication is important: if your competitive advantage in Meta has historically been sophisticated audience targeting, that advantage is eroding. The new competitive advantage is creative production quality, CAPI signal richness, and catalog data completeness. Those are the inputs the Andromeda system can actually differentiate on. Advertisers who adapt their workflow around those three inputs will outperform those still trying to optimize audience selection in a world where the algorithm has taken that function away.
Sources: Meta for Business News · Meta for Developers Blog: Project Andromeda
