How Facebook’s New Algorithms Work

Driving traffic from Facebook isn’t easy — complex promotion algorithms make it difficult to target the right audience. To avoid driving losses, experts from 888STARZ Partners will remind you what algorithms appeared on Facebook in 2025 and how to use them to promote top offers.

Meta GEM

One of the most essential algorithms on Facebook is Meta GEM (Generative Ads Recommendation Model), which is responsible for showing relevant ads to users. Essentially, the Meta GEM neural network works as an "archivist" on Facebook, parsing the entire pool of ads on Facebook and matching them to user interests, after analyzing their behavior on the platform. For example, if you liked a photo of your friend from a nature trip, the algorithm will show you ads for camping gear, hiking tours, etc. And if you engage with such an ad, the number of banners will increase further. Moreover, your creativity can be shown even if it is imperfect. Meta GEM can display plain banners if they match the target audience’s interests. For the Meta GEM algorithm to notice your banner, it must meet these criteria:

• The ad must be tailored to a specific audience — universal solutions won’t work.
• Adding storytelling with a clear message for the target audience is appropriate. You can write a story about using the promoted product or explain how it will benefit the audience.
• The visual component of the creative must also have thematic content matching the audience’s interests — banks and money for financial offers, exotic countries for tours, etc.

Meta Lattice

In the past, Meta used a whole arsenal of different algorithms and AI models to perform various tasks on Facebook — some tracked traffic, others assessed the relevance of Reels, and so on. For a while, this structure worked, but the main drawback was the isolation of databases when training the AI tools, which reduced algorithm efficiency. When the problem became obvious, Meta combined these tools into Meta Lattice, a unified system that tracks all campaign goals (e.g., traffic, conversions). Now, affiliates can set any goals for their ad campaigns, and all metrics, such as traffic volume and conversions, will be tracked and analyzed by a unified system.

As with Meta GEM, the work of the Meta Lattice algorithm cannot be influenced directly — the system will automatically promote your ad based on its performance and campaign duration. To get a promotion boost from Meta Lattice, you need to have a successful and long-term ad campaign already. For this, use less aggressive creatives and work on increasing user engagement. Otherwise, the algorithm will distribute your ads and competitors’ ads equally.

Also, note that the offer shouldn’t be too niche, otherwise Meta Lattice won’t be able to identify the right audience. As the company states, the Lattice system can increase conversions by 6% and ad relevance by 12%.

Meta Andromeda

The Meta Andromeda neural network, launched in 2024, has significantly improved the efficiency of displaying ad creatives to users. Meta Andromeda can identify general user interests and deeper nuances, such as relevant product brands and their price ranges. For the Andromeda neural network to work most effectively, consider the following factors:

• You must have a large pool of creatives, as this increases the chance of ad impressions and reaching the target audience.
• A creative approach is mandatory, as the more interesting and unique the ad materials you create, the higher the likelihood of appearing in user recommendations. Use different angles, emotions, colors, styles, themes, etc.
• Using Facebook’s internal tools, such as Advantage+ creatives and DCO, positively impacts the performance of the Andromeda algorithm.

Thanks to the Meta Andromeda neural network, even non-obvious ad combinations often achieve good reach and conversions. If done correctly, Andromeda can increase ad relevance by 8%, boosting metrics like clicks and conversions. However, remember that this neural network only works when promoting white offers — cloaking for betting, gambling, or other gray verticals will only worsen relevance.

Sequence Learning

The Sequence Learning algorithm analyzes user behavioral factors before and after conversion, optimizing ad delivery based on this data. For example, suppose a user purchases a joint ointment through an offer. In that case, they will see ads not for other ointments but for offers like joint-strengthening vitamins, braces, wellness equipment, and other thematic products. Sequence Learning acts as personalized retargeting, building logical chains from product to product, which helps create deep sales funnels. To effectively use Sequence Learning, work with multiple offers from one advertiser or select offers from different advertisers within the same niche.

Conclusion

Meta continues to actively push the integration of advanced AI functionality into Facebook, making the platform more user-friendly while simplifying life for advertisers. But to profit, simply using an effective traffic channel isn’t enough — you also need to find a promising region for promotion. For example, did you know that the annual increase in Facebook users in Nigeria has nearly reached 10 million, with projections of 69.4 million active social network users in the country by 2026? Tempting.

It gets even more tempting when you learn about Nigeria's Goldmine campaign by the 888STARZ Partners platform. All 888STARZ partners participating in the campaign can earn up to $400 per FTD referred from Nigeria, plus bonus payouts for every repeat deposit using the promo code NIGERIAGOLD25. Learn more about the campaign terms at this link.