The Andromeda galaxy, a tilted spiral of blue and gold dust, with two small companion galaxies among stars

AI

What did Meta's October 2026 ads ranking upgrade (Andromeda, GEM and GenRecs) change, and should B2B advertisers do anything differently?

By The Tofu team ·

On October 28, 2026, Meta said three ads ranking upgrades reach accounts automatically, with no setting to turn on. Andromeda, which picks the shortlist of ads, now routes each ad to specialized "expert" models. A new generation of GEM reads twice as much engagement history. And GenRecs, which describes each ad in tokens, expanded to retrieval on Facebook Feed in Q3 2026. B2B advertisers don't need to rebuild anything. Meta says results still depend on what you optimize for, the data you share and how much optimization you hand over, so work on those three.

As of November 2026.

What changed

BeforeNow
Andromeda (retrieval: which ads make the shortlist)Introduced in 2024. In December 2025 Meta said it "quickly scans tens of millions of ads and selects a few thousand strong candidates."Meta says Andromeda "now routes each ad to specialized 'expert' models," "scores every eligible ad" and surfaces top candidates "at over one billion ads per second." Its engineers say it replaced a single model of the person (a "user tower") with several specialized ones. In Q3 2026 Meta made it 10X more complex "without increasing computing costs or hardware needs"
GEM (the ranking foundation model)In March 2025 Meta said GEM had raised conversions "by up to 5%" in its first launch, on Reels"The foundation model behind ads recommendations across Facebook, Instagram, Reels and off-Meta-conversions," for "every major advertiser goal." The new generation reads "2x longer engagement history through a 2.5x more complex model"
GenRecs (generative recommender)Not in Meta's 2025 post on its ranking systemsLarge language models "describe the ad in semantic tokens." In Q3 2026 Meta "expanded GenRecs to generative retrieval on Facebook Feed"
What you have to do"There's no setting to switch on, no new campaign type to adopt, no migration to plan."

Meta's own figures, from the October 28 post:

  • All ranking improvements: "8.3% more ad clicks and a 15.7% uplift in conversions on Facebook" in Q2 2026.
  • Andromeda: the Q3 changes "increased conversions by 1.37% on Facebook and 0.24% on Instagram Reels." Since 2025, "a 2.4% increase in conversions on Facebook."
  • GEM: since 2025, "up to a 18.9% conversion improvement and a 12% ad click-through rate improvement."
  • GenRecs: nearly 2% more conversions "across targeted segments" from generative retrieval on Facebook Feed, and a 0.35% lift across surfaces when paired with GEM and Meta's Adaptive Ranking Model.

Meta doesn't say whether these are averages, and some cover only "targeted segments." It doesn't break them out for lead ads, B2B or small accounts. The October 28 post gives no baseline or method for any of them (it links the Q2 figures to Meta's Q2 2026 earnings call), the Andromeda engineering post gives none for the 2.4%, and the GenRecs post dates its 0.35% to internal data from February 2026.

What is GenRecs, and how does it pick which ad someone sees?

GenRecs turns each ad into a short set of tokens that describe it. Meta says large language models read an ad's text and images and "describe the ad in semantic tokens that capture its key attributes," such as "a product's brand, color, fit and style." Meta's October 19 engineering post says this replaces keeping "a dedicated ID for every item." A new ad can then borrow what the system already knows about similar ads, which Meta's engineers say helps recover "cold-start performance."

GenRecs feeds the retrieval step. Meta says it "enriches Andromeda": it "generates the semantic ID for the ideal ad for a person," then matches that to the ads available. Ranking models, which GEM teaches, then make the final pick. The effect Meta describes for advertisers is that "new ads, refreshed creative and new products get on a faster track to performing."

Meta's examples are retail ones. It doesn't say how well the tokens describe a software screenshot, a webinar invite or a whitepaper ad.

Who it affects

  • Every Meta advertiser, with no opt-in. Meta says "every one of these improvements reaches your account automatically" and "they're already in the system that's delivering your ads today." It doesn't say whether every part has reached every account, surface and country yet.
  • Facebook Feed most directly, for GenRecs. It's the only surface Meta names for generative retrieval. Andromeda's engineering post says the new retrieval system scores ads across Facebook and Instagram.
  • Low-volume accounts, in Meta's telling. Meta says "campaigns with few conversions now benefit from what GEM learns about other clicks data." That describes many B2B ad sets, but Meta gives no lead-gen figure.
  • Who can relax: anyone expecting a forced migration. There isn't one. Meta also names no way to turn these systems off.

What to do now

Meta lists three things that still decide your results. Each maps to a B2B fix.

  1. Don't rebuild campaigns for this. There's no new campaign type and no migration. Keep ad sets that already produce qualified leads.
  2. Optimize for the outcome you actually want. Meta's first lever is "the outcomes you tell it to optimize for." For B2B, a form fill isn't the outcome; a qualified lead is. If you have the volume, send CRM lead stages back to Meta and optimize on them. See how to send HubSpot or Salesforce lead stages to Meta and what to optimize for when you only get a few leads a week.
  3. Send better data. The second lever is "the quality of the data you share - including how you measure performance." Set up the Conversions API and check match quality. See setting up the Conversions API without a developer and raising Event Match Quality for a B2B form.
  4. Choose how much to hand over. The third lever is "how much of the optimization you let the models handle." Meta names no settings here; in practice it means choices like Advantage+ audience. For B2B, customer exclusions and tight audiences can still be worth some lost reach. Weigh it with whether to turn off Advantage+ audience for B2B and how to turn off Advantage+ on a Leads campaign.
  5. Give the system distinct creative. Meta says the upgrades mean "more of your ads, creative variations and products are considered for each person." Its December 2025 post still holds: ads that "look or feel alike" are "variations of the same creative." A few visually different concepts per ad set do more than many small edits. See Meta's creative diversity rating and how to test new B2B ads fairly.
  6. Judge it on pipeline, not Meta's averages. The gains Meta reports came in over Q2 and Q3, so there's no single date to compare before and after. Track cost per qualified lead by month. If it's rising, see why B2B cost per lead on Meta keeps going up.

Dates to watch

DateWhat happens
2024Meta introduces Andromeda
March 27, 2025Meta describes four ranking systems, including GEM and Andromeda
December 16, 2025Meta says look-alike ads are treated as "variations of the same creative"
Q2 2026Ranking improvements add 8.3% clicks and 15.7% conversions on Facebook, per Meta
Q3 2026Andromeda gets 10X more complex; GenRecs generative retrieval expands to Facebook Feed
October 19, 2026Meta's engineers explain how GenRecs tokens work
October 28, 2026Meta announces the three upgrades; its engineers publish the new Andromeda design
No date yetMeta's engineering post expects "further gains" as more optimizations reach production. Meta has announced no advertiser-facing change

Sources

Ask Tofu next

Pick a question and Tofu Ads Agent answers it in chat.

Related answers

More on

Let an agent build your next campaign.

Tofu Ads Agent reads your CRM, plans the funnel, builds the audience and creative, and stages full campaigns across LinkedIn, Google and Meta for your approval.

Try it free Get a demo