Most of what has been written about Andromeda since late 2024 arrives at the same conclusion: you need more creative diversity. Produce more. Vary your formats. Feed the system.
That conclusion is correct. But it stops short of the more important insight. And the brands and teams that go one level deeper will find a meaningful strategic advantage waiting for them.
The question worth asking is not just how much creative diversity you need. It is what kind of creative diversity the system actually rewards, and why.
Content
What the System Is Actually Reading
Andromeda is Meta’s ad retrieval engine. Its job is to decide which ads are eligible to enter the auction before bidding begins. It does this by reading creative content and matching it to users based on predicted relevance.
But Andromeda does not operate alone. It works in conjunction with GEM, Meta’s Generative Ads Recommendation Model, which Meta describes as the central brain of its ads recommendation system. And the way GEM works reveals something that has not yet made its way into most strategic conversations about creative.
GEM is built to understand sequences of user behaviour, not just individual signals. According to Meta’s engineering team, it processes user behaviour sequences spanning thousands of events, learning from a much longer history of organic and ad interactions to build a deeper and more accurate understanding of the user’s purchase journey. Meta’s stated ambition for GEM’s future development is explicitly to power intent-centric user journeys.
That phrase is significant. Intent-centric. Not format-centric, not volume-centric. Intent-centric.
The system is not just reading your creative for visual diversity. It is building a model of where each person is in their purchase journey, what their intent is at this moment, and which creative is most likely to match that intent. Meta’s earlier sequence learning research, which underpins GEM, confirmed that this approach to understanding behaviour before and after conversion drove 2 to 4% more conversions on select segments.
Why This Changes the Creative Diversity Conversation
When the industry talks about creative diversity in the context of Andromeda, the conversation is almost always about format diversity. Different video lengths. Different aspect ratios. Different hooks. Different product angles.
These things matter. But they are not what the system is primarily optimising for. The system is optimising for signal diversity. Specifically, it is trying to understand whether your creative library contains meaningfully different signals about different human states, different motivations, and different moments in the consideration journey.
A brand that produces ten creative variations around the same emotional register, the same motivational hook, and the same moment in the customer journey is giving the system one signal, dressed ten different ways. The retrieval system is sophisticated enough to read that sameness. Meta’s own documentation makes clear that the system dynamically reconstructs latent user-ad interaction signals, meaning it is reading what your creative is actually communicating at a deep level, not just cataloguing its surface characteristics.
A brand that produces creative built around genuinely distinct moments, a different emotional state, a different point in the consideration cycle, a different motivation, is giving the system something it can act on with precision. Each piece of creative represents a different signal. The system can match those signals to the users whose behavioural sequences suggest they are in that state, at that moment, with that intent.
This is not a creative production insight. It is a strategic one.
The Purchase Journey Dimension
Meta’s Adaptive Ranking Model, which launched on Instagram in Q4 2025, makes this even more explicit. It is designed to capture user intent with unprecedented fidelity through longer behavioural sequences, and since launch has delivered a 3% increase in ad conversions and a 5% increase in click-through rate.
What this means for D2C brands is that Meta’s infrastructure is increasingly capable of identifying where a specific person is in their consideration journey, not just who they are. The system can distinguish between someone at the beginning of their research and someone close to a decision. It can understand the difference between a person in a high-motivation state and one who is browsing passively.
But it can only act on that understanding if your creative library contains something appropriate to serve at each of those moments. A brand with creative that speaks to different moments across the consideration cycle, each one briefed around a specific emotional state and motivation, is giving the system exactly what it needs to do its job well. The more precisely the creative reflects the human reality of those moments, the more effectively the system can deploy it.
What This Means for Creative Strategy
The teams that will get the most from Andromeda and GEM are the ones whose creative production is anchored in a genuine understanding of their customer’s moments.
When you know which specific windows drive your highest-propensity customers to buy, and what emotional state they are in at those moments, you have the foundation for creative that is meaningfully diverse in the ways the system is built to detect and reward. Each piece of creative is not just a different format. It is a different signal about a different human moment. And the system, which is now reading user behaviour at a level of sophistication that most creative briefs have not yet caught up to, knows how to deploy each one.
This is where the understanding of behavioural patterns across the market becomes particularly valuable. Knowing not just who your customer is but when they are most receptive, and what is happening in their life at that moment, produces creative briefs that are genuinely distinct at the level of human intent. That distinction is what the system is built to act on.
Meta has engineered a system that rewards creative aligned with human intent and moment. The brands and teams that understand it at that level will find the performance uplift available from Andromeda is considerably greater than those treating it purely as a production challenge.
The strategic opportunity is real. And most of it is still untaken.
The Graygency is a performance marketing agency for D2C brands. We practise True Performance Marketing to identify micro-moments, building targeted creative for those moments, and constructing growth systems that compound over time.
Sources: Meta Engineering Blog, December 2024 (Andromeda); Meta Engineering Blog, November 2024 (Sequence Learning); Meta Engineering Blog, November 2025 (GEM); Meta Engineering Blog, March 2026 (Adaptive Ranking Model).
