What Meta’s GEM Really Wants from Your Creative in 2026
Why it Matters
Meta’s GEM now determines which ads are shown, who sees them, and the order they appear in. It rewards diverse and well-structured creative work, rather than favoring the same successful ideas. If you don’t provide GEM with high-quality, persona-driven, and varied creative consistently, you’re limiting your growth, efficiency, and long-term results in 2026.
If your creative strategy hasn’t changed since GEM quietly started running in Q2 2025, it’s time to reconsider. Meta’s Generative Ads Recommendation Model (GEM - we, too, don’t know why it’s not abbreviated as GARM) is now the central intelligence behind every ad auction on Meta ads, and it has already reshaped what “good creative” means at every stage of the process, from the creative brief to the post-launch analysis.
For media buyers and creative strategists who live inside Meta accounts every day, the practical implication is this: GEM learns from diversity. Diversity in creative type, in video length, in visuals, in angle, in personas, in audience targeting, in trying new things, and in findings things that do, and do NOT work. It doesn’t reward a single hero ad, but rather rewards accounts that give it enough creative material across enough formats, angles, and personas to actually do its job. The teams that are winning in 2026 aren’t the ones making one great ad. They’re the ones running systems that continuously feed GEM what it needs. Keep in mind though, even though GEM needs enough creative, that doesn’t mean quantity is king. We will die on this hill - quality over quantity of creative, all day, every day. If you can have quality AND quantity, you go Glen Coco. If not, stick with quality, and quantity will come as your hit rate gets higher and higher and ads are fatiguing slower and slower.
Keep reading to see how this plays out across the full creative cycle.
Start with Data
GEM evaluates sequence features, meaning it looks at patterns in user behavior over time, not just what someone did in their last session on Facebook. That makes the data work that happens before a single frame is shot more important than ever. Before briefing any creative, the most effective teams are digging into comment sections, reviews, post-purchase surveys, and customer service threads to identify the exact language, pain points, and emotional hooks that resonate with real buyers.
The goal at this stage is not to confirm what the brand wants to say; it’s to surface what customers are already saying, and then build creative around that. GEM is looking for signals of genuine engagement, and content that mirrors the language of real users tends to generate those signals at a much higher rate than polished brand messaging that’s disconnected from how buyers actually speak and think.
Data inputs to prioritize at this stage: top-performing organic content from the brand’s own Instagram and Facebook channels, competitor ad libraries, high-rated product reviews (especially the critical ones), and any first-party purchase or survey data available.
2. Account for Persona Diversity in Creative Briefs
One of GEM’s core functions is figuring out which creative matches which user. It’s not trying to optimize one ad for everyone in its target audience, but instead building a picture of which message, format, angle, and tone combination moves which person closer to conversion. That means a brief that only produces one angle or one persona may leave performance on the table, or it’s restricting horizontal scale of the account if other personas aren’t also being tested in the creative mix.
A well-constructed brief for a GEM-era campaign defines at least three distinct audience personas and maps a specific creative angle to each one. These three personas are:
The skeptic who needs social proof.
The research-driven buyer who needs to understand the mechanism.
The impulse buyer who just needs a strong hook and a reason to act now.
Each of these users exists inside a broad Meta audience, and GEM will find them, but only if the creative library includes something to serve each of them.
This is also the stage where format diversity gets planned intentionally. A brief should specify which concepts are built for video, which work as statics, and which should be shot with a lo-fi, UGC-style approach versus a higher-production treatment. GEM processes all of these differently, and accounts that supply a range of formats give the system more surface area to match against different user preferences. Some concepts could be made with each of these treatments in mind and others may be best suited for only one. It’s a good creative strategist who can determine which is which and which will likely succeed once in the ad account.
3. Create Content with GEM’s Signals in Mind
GEM is reading creative elements like visuals, themes, hooks, and language to evaluate relevance and predict engagement. That has real implications for how video is produced and edited.
On the video side, the hook carries a disproportionate amount of weight. (Remember that the hook nowadays is the first TWO (2) seconds of a video). GEM picks up on early retention as a signal, so the hook must communicate the core value proposition or pattern interrupt before a viewer has any reason to scroll. A hook that leads with a strong declarative statement, a recognizable problem, or an unexpected visual tends to hold attention better than branded intros or slow builds.
UGC and testimonial-style content continues to perform well in this environment because it reads as organic to users who are already wired to scroll past anything that looks like an ad. GEM is trained on both paid and organic interaction data, which means it understands what native-feeling content looks like and can identify when a paid ad is generating the kind of engagement typically reserved for organic posts. That is a signal worth chasing. Authentic creator testimonials and honest product walkthroughs tend to carry genuine engagement data that GEM can actually learn from.
During editing, plan for multiple cut-downs and variations from the same shoot. A 60-second long-form video, a 15-second hook cut, a 6-second version for Reels placement, all with audio (voiceover and/or music) and text on screen or subtitles are all legitimate distinct assets that GEM can deploy in different contexts for different users. Treating each of those as separate deliverables instead of an afterthought multiplies creative inventory without multiplying production cost.
4. Upload with Structure and Patience
How creative gets loaded into the account matters. GEM needs a mix of time, quality, and creative volume to identify patterns, and it needs time to distinguish between signal and noise. The accounts that see the most consistent improvement are the ones that consolidate campaigns, upload multiple diverse creatives/ads per ad set, run across multiple placements rather than limiting delivery, and then leave things alone long enough for GEM to actually learn. One of our favorite tactics is to load in a handful of new creatives every 3-4 days (depending on the size and spend of the account), and have a “queue” of new creatives ready to be loaded in at any time and when the account needs fresh creatives. By letting the data in the account direct when to add in new creatives, it allows for the GEM system to give each creative a real chance instead of being overwhelmed with too many ads at once to test and learn, OR having too few and forcing spend into a creative that it knows won’t work, or it doesn’t want to deliver.
The old habit of launching one ad, watching it for 48 hours, then either killing it or duplicating the ad set is actively counterproductive in GEM’s system. GEM is tracking sequences across placements and over time. Constant resets interrupt that learning and keep the model from building reliable predictions. The updated best practice is to give campaigns at least seven days before making any significant structural changes. If early numbers look rough, that discomfort is part of the learning phase and not necessarily a signal to act. Caveats to this are small spending accounts that cannot waste 7 days’ worth of spend in a test, or very large spending accounts that can get to full statistical significance in spend and data before the 7-day mark.
From a creative standpoint, aim to upload at least four to six distinct and diverse creatives per ad set. Think about genuinely different angles, hooks, and formats that represent the persona diversity mapped out in the brief.
5. Analyze Performance Through GEM’s Lens
Because GEM is optimizing across sequences and not individual placements, traditional creative analysis needs to evolve. Evaluating an ad purely on its click-through rate or its cost-per-click in isolation misses the point. A video that generates low direct conversions but consistently appears early in the sequences that lead to conversion is doing something valuable that a surface-level report will bury.
The more useful questions to ask in analysis:
Which creatives are holding their performance over time rather than spiking and dying?
Which formats are getting the most delivery across multiple placements?
Are UGC and testimonial formats outperforming polished brand content on a per-conversion basis?
Where in the funnel is each creative contributing?
Is hook retention (three-second video views relative to impressions) telling a different story than downstream conversion data?
Tracking creative performance over at least a two-week window — rather than optimizing day by day — gives a more accurate picture of what GEM is actually using and why. The patterns that emerge from that kind of analysis become the foundation for the next round of creative development.
6. Build Iteration Concepts from What You Learn
The creative cycle does not end at creative analysis. GEM is continuously learning, which means the competitive advantage in this system belongs to the teams that iterate faster and smarter, and not the ones that find one winning ad and stop their entire creative cycle process.
When a format is working, the next step is not simply to run more of the same ad. It’s to identify why it’s working and build adjacent concepts that test a specific variable, while still making the new concepts truly diverse and clearly a net-new ad asset.
If a UGC testimonial featuring a specific customer concern is outperforming everything else, the iteration brief might test three different concerns from three different customer segments using the same UGC format. If a particular hook structure is holding attention in the first three seconds, test that same hook applied to a different persona’s problem.
When something is underperforming, that’s equally valuable data. A static image that failed to generate meaningful delivery signals that the format or angle may not match the audience GEM is finding in that campaign. That informs both the next creative brief and potentially the campaign structure itself.
The iteration step feeds directly back into the data phase, completing the loop. New concepts get researched, briefed, shot, uploaded, and analyzed again — and GEM gets progressively better at optimizing for each account because the creative library it’s working with gets progressively more refined.
The Creative Strategist’s New Job Description
GEM has not replaced the creative strategist or the media buyer. It has redefined what the job actually is. The manual lever-pulling of the ad buyer of audience exclusions, placement restrictions, constant bid adjustments are no longer where the edge in media buying lives. The edge lives in the quality and diversity of the creative system being fed into the algorithm.
In practical terms, that means the most valuable skill set in a GEM-era Meta account combines three things: the ability to translate customer data into specific creative concepts, the ability to produce or brief content across multiple formats (with particular emphasis on video hooks and authentic UGC-style content), and the discipline to analyze performance at the pattern level rather than reacting to daily fluctuations.
GEM is already running. It is already making decisions about which ads get shown to which users in which order. The only question is whether the creative library it’s working with is diverse enough, well-researched enough, and refreshed frequently enough for it to actually do its best work. For the teams willing to build that system, the performance data — including reported gains of 5% in Instagram conversions and 3% in Facebook Feed conversions since GEM’s rollout — suggests the upside is meaningful.
The creative process has always mattered on Meta. GEM has just made it the entire game.
If this blog has been helpful, imagine how helpful 24/7 access to some of the world’s best marketing minds would be. That’s exactly what you’d have in Foxwell Founders.