If you run paid social campaigns, the ground under your feet has shifted. We recently attended the Performance Marketing Summit in San Jose, where Meta's engineering and partnership teams pulled back the curtain on the AI infrastructure now driving every ad on Facebook, Instagram, and Messenger. Here's what stood out, and what it means for your strategy.

‍

Meta's Ad Engine Has Been Rebuilt with the help of AI

Meta's team walked us through five interconnected AI systems that now handle everything from retrieval to ranking to dynamic copy adaptation — and they're doing all of it in under a second. That speed is a direct result of how deeply Meta has invested in AI to accelerate their own internal engineering processes, not just advertiser outcomes (source).

Here's how the stack works:

Meta Andromeda is the retrieval engine. Instead of starting with your audience settings, it starts with your creative, analyzing visuals, themes, and messaging to find users most likely to engage. Andromeda represents a 10,000x increase in model complexity for ad retrieval, narrowing tens of millions of ads down to a few thousand candidates per user in real time (source). 

Meta GEM (Generative Ads Recommendation Model) is the intelligence layer,  a foundation model trained at LLM scale. GEM catalogs, analyzes, and connects trillions of pieces of information to make the ad system more intelligent and effective. 

Meta Lattice is the unified ranking architecture. It replaced Meta's previous collection of specialized ranking models with one giant system that learns from all campaign objectives and surfaces simultaneously,  so what works on Instagram Stories informs Facebook Reels automatically. 

Sequence Learning is the journey intelligence within Lattice — helping the system understand where someone is in their purchase path before deciding which ad to show. 

The Adaptive Ranking Model ties it all together, enabling the entire pipeline to process, rank, and deliver a personalized ad decision in under a second. Meta pointed to this as a clear example of AI improving its own engineering velocity,  and the result for advertisers is the right creative reaching the right person at the right time, faster than ever before (source).

‍

Catalogs & Product Ads

One of the most actionable takeaways from the Summit was around product catalog strategy. Meta is pushing hard for richer catalog data — not more catalogs, but better ones. That means consolidating SKUs and loading up on detailed descriptions, attributes, and tags so the AI has more signals to work with (source). 

Dynamic video is the next frontier here. Using catalog feeds to generate video from static product images is now table stakes, with multiple partners offering scalable templates. And a new ad format is on the horizon that blends static and video product ads in a single ad set, letting the system automatically flip between formats based on performance. 

If you haven't run catalog product reporting yet, start now. It's one of the clearest signals for identifying your top performers and prioritizing them for video creative.

‍

Incrementality & Measurement

The Summit was refreshingly direct about incrementality: conversion lift and brand lift studies (source #1) | (source #2) should be running consistently, not just when something feels off. If your platform reporting looks dramatically different from your real-world results, the answer isn't to trust one over the other, it's to run more tests until you understand the gap. Short-term performance dips during testing phases are a worthwhile trade for long-term calibration.

At NetConversion, this philosophy aligns closely with the work we've been doubling down on for our clients. We've been deepening our Marketing Mix Modeling practice,  and one of the most significant upgrades has been incorporating actual Google Search volume data directly into our MMM inputs. By pulling in the raw search feed rather than relying on modeled proxies, we're able to capture real consumer demand signals and understand how media activity is influencing intent at the top of the funnel, not just at the point of conversion.

The combination is powerful: MMM gives us the macro view of how each channel contributes to business outcomes over time, while tools like Meta's conversion lift and brand lift studies let us fine-tune and validate the model's insights at a more granular level. When they agree, we have a strong conviction. When they diverge, we have a clear signal to investigate.

As privacy constraints continue to limit signal at the individual level, MMM will only become more critical, not as a replacement for platform measurement, but as the foundation that demonstrates the real value of media to the business. The agencies and brands that have invested in building rigorous measurement infrastructure now will be the ones with a durable competitive advantage as the landscape continues to evolve.

‍

Creators & Partnership Ads

Meta's partnership ads team shared what they called the "Dinner Party" philosophy: when you work with a creator, you're a guest in their community. Provide a framework and inspiration, but don't hand them a script. The brands winning with creator content are the ones giving creators room to be themselves, operating within their world rather than trying to redirect it.

What made this session particularly practical was the emphasis on how creators communicate, specifically what Meta calls the "language of Reels." The creative elements that make creator content thrive are: a strong hook in the first two seconds, visual dynamism through fast cuts and transitions, deliberate audio choices, text overlays, and genuine human presence on screen. These aren't stylistic preferences; they're the signals Meta's AI is reading to determine creative relevance and reach.

On the tooling side, a major update is coming to the Partnership Ads Hub (source), which is being redesigned to make the entire influencer workflow significantly easier. The new hub moves away from the manual code-based approval process and introduces a streamlined discovery and approval experience. Critically, it will also surface organic mentions — so brands can see which creators are already talking about them, and either promote that content directly or use it as a starting point for a formal partnership (source).

The AI powering the hub is worth noting: Meta uses over 130 data points to predict how well organic creator content will perform as a paid ad before you ever put budget behind it. The "Recommended" tab in the hub is designed around exactly this, surfacing the organic content most likely to convert, ranked by the model's prediction of partnership ad performance. That's a meaningful shift from the current workflow of manually combing through creator content and guessing what will translate (source).

‍

What This Means for Net Conversion Clients

The shift is clear: feed the machine better inputs, simplify your structures, and measure more rigorously. Creative diversity, rich catalog data, and consistent incrementality testing are no longer advanced tactics — they're the baseline for competitive performance on Meta in 2026.

We're already applying these principles across client accounts. If you want to talk through what this means for your specific campaigns, reach out.

‍