If your Meta campaigns have felt harder to stabilise over the past few months, you are not alone. Many advertisers have seen creative fatigue more quickly, results fluctuate after relatively small changes, and previously reliable tactics lose effectiveness.

This shift is not coincidental. Meta has been running a new ad prediction system called GEM, short for Generative Ads Model, since November. Rather than adjusting the existing algorithm, Meta rebuilt how ads are selected and delivered across Facebook and Instagram at a foundational level.

Understanding how GEM works is now critical for building campaigns that perform consistently.

What Makes GEM Different

The biggest shift GEM introduces is where ad delivery decisions start.

In previous systems, ads were evaluated first and then ranked against one another. GEM flips that approach: it begins by analysing user behaviour, then selects the ad that best fits that context.

The model examines long-term patterns across Meta platforms and how users engage with content, move between Feed, Reels, Stories, and Explore, and how their behaviour evolves over time. It also takes timing into account, such as whether a user is browsing, researching, or closer to taking action.

With this behavioural context, GEM predicts which type of ad a user is most likely to respond to at that moment. Ads are no longer judged solely on creative quality or past performance; they are matched to predicted intent.

This shift explains why frequent campaign changes often cause instability instead of improving results.

Why this Matters for Advertisers

With GEM, short-term fluctuations matter less than they once did. The system focuses on learning from patterns over time rather than reacting to isolated events.

Tactics like constant creative tweaks, frequent pausing, or duplicating campaigns to “reset” performance can actually harm results, interrupting the continuity GEM needs to make accurate predictions.

Optimisation is still important, but it’s shifted from reactive adjustments to creating stable conditions that allow the model to learn and perform effectively.

Building Campaigns That Work With GEM

Campaign structure now plays a larger role in stability than in control.

Simpler structures with fewer campaigns and ad sets generally perform better than fragmented setups. Campaigns should be designed to run long enough for meaningful behavioural patterns to emerge, rather than being rebuilt whenever performance fluctuates.

Consistency in objectives, budgets, and delivery helps GEM make clearer predictions. Frequent structural changes make results harder to interpret and often delay improvement rather than accelerate it.

Creative Strategy Under the GEM Model

Creative remains the primary driver of performance, but the way it is evaluated has changed.

Instead of searching for one winning ad and extending it through minor variations, advertisers benefit more from running distinct creative angles that address different motivations and levels of intent.

Some users need reassurance and proof. Others respond better to education, context, or explanation. Some are motivated by urgency or offers, while others engage more with emotional or aspirational messaging.

GEM uses behavioural signals to decide which of these messages to show to which users. The role of the advertiser is to provide genuinely different messages, not multiple versions of the same idea.

Optimising Live Campaigns Effectively

Optimising campaigns under GEM works best when you allow enough time for meaningful data to emerge and avoid impulsive changes.

Review performance in line with your product’s or service’s actual purchase cycle: short-consideration offers may convert quickly, while longer-consideration decisions can take weeks. Evaluating too early risks turning off ads that are still learning or scaling ads that only perform at the top of the funnel.

The most effective optimisations are deliberate and guided by clear trends, rather than reacting to daily fluctuations.

Improving Signal Quality

Signal quality is one of the strongest levers available under the Generative Ads Model.

Accurate conversion tracking, consistent spend patterns, stable delivery, and meaningful creative variety all help the model learn more effectively. In contrast, inconsistent budgets, excessive testing, and frequent resets introduce noise.

While this approach can feel less active than constant optimisation, it often produces stronger results over time because it improves the accuracy of the model’s predictions.

Make GEM Work for You

GEM represents a shift from ad-level optimisation toward behaviour-led delivery. Campaigns perform best when they provide consistent inputs and enough creative range to align with different user states.

Advertisers who adapt their approach tend to see more predictable scaling and fewer sharp drops in performance. Those who rely on short-term tactics often experience increasing volatility.

Understanding how the Generative Ads Model evaluates users and ads allows you to build campaigns that align with how Meta now makes delivery decisions, rather than working against them.

Ready to Take Control of Your Marketing?

The trends point to one clear direction: success comes from an integrated approach that blends data, creativity, and technology to deliver measurable business results. It’s about building trust, strengthening authority, and proving performance through meaningful metrics.

Looking ahead to 2026? At Unity Online, we make digital marketing simple and effective. We’re passionate about helping businesses of all sizes grow with strategies that actually work.

Explore some of the brands we’ve helped scale and see the results for yourself.

Our team is friendly, knowledgeable, and ready to guide you through the evolving digital landscape with confidence.

Get in touch today to boost your visibility, sharpen your strategy, and set your business up for a successful year ahead.