Meta Ads, Premium Content

Meta Ads Insider Exposes How The Algorithm ACTUALLY Works

What Meta's algorithm interview means for advertisers

In this conversation, Meta's Matt Steiner explains ad retrieval, ranking, creative diversity, and auction volatility. The practical lesson is to give the system useful creative choices, then use performance data to develop the next set of ideas.

1. Separate retrieval from ranking

Steiner describes Andromeda as the personalized retrieval stage that selects candidate ads. Ranking then evaluates which candidate to show. This distinction helps explain why the creative options available to the system matter before the final auction decision.

2. Aim for useful diversity

Different people can respond to different versions of the same idea. The interview supports a reasonable range of creative candidates, rather than producing an enormous number with no meaningful difference. Uneven delivery across those versions does not automatically mean the lower-spending assets are bad.

3. Understand Lattice and GEM

Steiner explains Lattice as model consolidation: combining training data from previously separate models. GEM is described as a larger recommendation model that informs ranking systems. These are related parts of delivery, rather than interchangeable names for a single account setting.

4. Put volatility in auction context

The discussion connects delivery swings to a real-time auction with many advertisers changing bids and spending levels. That context matters when interpreting fluctuations. A change in results does not by itself identify which setting, creative, or competitor caused it.

5. Keep the human work focused on ideas

The interview emphasizes the advertiser's role in developing novel messages, forming hypotheses from results, and creating variations to test. Use the delivery system's speed while keeping judgment in the creative process. Review why a message resonated and turn that learning into the next brief.