Summary: Meta's GenRecs update changes how Meta understands your ads. An AI model now reads the images and words in every ad and turns them into a short set of labels, called tokens. Meta uses those tokens to decide what your ad is about and who should see it, even before anyone clicks. This guide explains GenRecs in plain English, how it fits with Andromeda and GEM, what results Meta has reported, and how to adjust your creative strategy and briefs.
What is Meta's GenRecs update?
GenRecs, short for Meta Generative Recommender, is Meta's newest ad system. It uses large language models (LLMs), the same kind of AI behind ChatGPT, to predict which ads people want to see. Meta explained GenRecs in its three new AI innovations in ads ranking update, alongside upgrades to Andromeda and GEM. AdExchanger covered the announcement on October 28, 2026.
The piece marketers need to understand is GenRecs Tokenization. Meta's engineering team explained it in an October 19, 2026 post. Tokenization is how Meta describes each ad to its AI: instead of a random ID number, every ad gets a short set of tokens based on what it shows and says.
Why it matters: before GenRecs, a new ad was mostly a blank slate until people engaged with it. Now Meta can read the ad and make an educated guess about it from day one. Meta says these updates apply to every account automatically. There is no setting to turn on.
How GenRecs Tokenization works, in plain English
GenRecs Tokenization works like a librarian filing a new book. The librarian does not wait for people to check the book out. They read the cover and the summary, then shelve it next to similar books. Readers who liked those books find the new one right away.
Meta does the same thing with your ads:
- Meta reads the ad. An AI model looks at your images, video and text, and understands general ideas such as "running shoe," "skincare" or "retirement planning."
- Meta boils it down to tokens. The ad becomes a short set of labels, a bit like tags. Meta says these tokens capture key details, such as a product's brand, color, fit and style.
- Similar ads share tokens. Ads that show similar products, people and messages end up with overlapping tokens, so Meta treats them as related.
Meta's engineering team gives the example of a new "trail shoe." Because its tokens overlap with other shoe ads, the new ad inherits what Meta already learned about that category instead of starting from zero.
The second half is where it pays off. When you walk in, the librarian looks at what you have borrowed before and what you are after today. She guesses the subject you will want next, the same way AI predicts the next word in a sentence. Then she pulls the closest books, even one that arrived this morning and has never been checked out. That is how Meta can show a brand-new ad to someone whose past clicks look nothing like it.
How GenRecs fits with Andromeda and GEM
GenRecs does not replace Meta's other ad systems. It works alongside them, and Meta upgraded all three at the same time. Here is what each one does, in marketing terms.
The key difference is in the last column. Andromeda and GEM mostly learn from what people already clicked. GenRecs adds a new input: the content of the ad. That matters most when there is little click history, such as a new ad, a new product or a new audience. For the ranking side of the system, see our guide to Meta's Adaptive Ranking Model, which Meta also pairs with GenRecs Tokenization.
GenRecs now feeds Andromeda too. Meta says GenRecs predicts the ideal ad for a person, then matches it to the real ads in its system. That is how Meta says it can reach someone who is in the market for what you sell, even when nothing in their past clicks looks like your ad.
What Meta says GenRecs has delivered so far
Meta's reported gains from GenRecs are real but modest. All figures below come from Meta's ads ranking update and are based on Meta's own data:
- Nearly 2% more conversions across targeted segments on Facebook Feed, after Meta expanded GenRecs to generative retrieval in Q3 2026.
- A 0.35% lift in conversions across surfaces from GenRecs Tokenization, paired with GEM and the Adaptive Ranking Model.
- A 1.37% conversion increase on Facebook and 0.24% on Instagram Reels in Q3 from Andromeda, after Meta made it 10x more complex. Meta says Andromeda has added 2.4% more Facebook conversions since 2025.
GEM still drives the biggest gains. Meta says GEM has delivered up to an 18.9% conversion improvement and a 12% click-through rate improvement since 2025, and it credited its ranking improvements with 8.3% more clicks and 15.7% more conversions on Facebook in Q2 2026.
Two reality checks on the GenRecs numbers
First, new ads still need time. Matt Steiner, Meta's VP of ads infrastructure, ranking and AI foundations, told AdExchanger: "You still have to learn." GenRecs shortens the learning phase but does not remove it. Second, ads cost more. Meta's average price per ad rose 14% year over year in Q3 2026, while ad impressions rose 10%, per Meta's Q3 earnings release. Wasted spend on ads Meta sees as duplicates now costs more too.
What GenRecs changes for your creative strategy
GenRecs means your creative is no longer just something Meta tests. It is something Meta reads. Below is what that changes, in practical terms. Meta has not published creative guidance for GenRecs, so these points are our take on how the system works.
Your ad now tells Meta who it is for
GenRecs reads the people, setting, product and words in your ad. Those details become part of how Meta decides who sees it. An ad showing a woman in her 60s talking about joint pain carries different signals than one showing a 25-year-old at the gym, even for the same product.
Ads that look alike get filed together
If you run 10 ads with the same product shot and 10 different headlines, Meta may file them as roughly the same ad. That matches what advertisers already see with Meta's creative Entity ID and the Creative Similarity score. Meta has not said these systems are connected, but they point the same way: small tweaks do not make a new ad.
New audiences and launches get a head start
A new ad aimed at a new customer type has no clicks yet. Under GenRecs, what is on the ad is the first thing Meta has to go on. Clear signals on the ad give Meta a better starting guess. A vague ad gives it very little. Meta says new ads, refreshed creative and new products now get on a faster track to performing.
Past campaigns carry over, good and bad
Meta says its stable token vocabulary lets learning carry across campaigns, such as running shoe ads from one season to the next. Relaunches should ramp faster. But if last season's concept flopped, a near-copy of it likely inherits that weak history too.
How to brief creative for GenRecs
GenRecs rewards ads that are clear about who they are for, what they sell and how they differ from your other ads. Use these five checks in your next round of briefs. They build on the basics of a creative strategy.
1. Show or say who the ad is for
Put your ideal customer on the ad. Use a creator who matches them, show their setting, or call them out directly, as in "Are you 60+ and dealing with knee pain?" This is especially useful when you test a new customer profile with no click history. Treat it as a hypothesis, not a rule: some accounts find that ads without a callout still win, so test both.
2. Make the product and problem obvious early
Show the product in the first seconds, name it in the copy or on-screen text, and state the problem it solves. Meta's AI understands general concepts, so plain, specific signals help it place your ad next to the right shoppers.
3. Change the concept, not just the headline
New creators, settings, formats and angles change what Meta reads. A new headline on the same video usually does not. Build briefs around distinct concepts and test them with a structured creative testing framework.
4. Judge winners by concept, not single ad
Because learning now spreads across similar ads, the real question is which concepts, angles and creators work, not which single ad won. Tag every ad by concept and review results at that level, the core of creative analysis.
5. Give new concepts time and clean data
New ads still need a learning period, so avoid killing a fresh concept after one day. Keep your Meta Conversions API set up correctly so the click and purchase side of the picture stays accurate. Meta says results still depend on three things you control: the goal you optimize for, the quality of the data you share and how much of the optimization you let its models handle.
Frequently Asked Questions About Meta GenRecs
What is Meta GenRecs?
Meta GenRecs (Meta Generative Recommender) is an ad system that uses large language models to read the images and text in every ad and predict who will want to see it. Meta detailed GenRecs on October 28, 2026, alongside upgrades to its Andromeda and GEM ad systems.
What is GenRecs Tokenization?
GenRecs Tokenization is how Meta describes each ad to its AI. Instead of a random ID number, every ad gets a short set of tokens based on what it shows and says, so similar ads share tokens and new ads can borrow what Meta has already learned about similar ones.
Do new Meta ads still go through a learning phase with GenRecs?
Yes. Matt Steiner, Meta's VP of ads infrastructure, ranking and AI foundations, told AdExchanger that new ads still have to learn. GenRecs gives new ads a better starting point by reading their content, which shortens the learning phase, but it does not remove it.
How should marketers change their creative for GenRecs?
Marketers should make each ad clear about who it is for and what it sells, build genuinely different concepts instead of headline variations, judge results by concept rather than single ads, and give new concepts enough time and budget to learn.
Do advertisers need to turn on GenRecs?
No. Meta says GenRecs and the Andromeda and GEM upgrades reach every account automatically. There is no setting to switch on, no new campaign type and no migration to plan.
The bottom line on GenRecs
GenRecs makes your creative the first signal Meta reads, especially for new ads, new products and new audiences. The gains Meta reports so far are modest, and new ads still need time to learn. But the direction is clear: brands that brief distinct, clearly targeted concepts and track results by concept will give Meta's AI the most to work with.
