Meta officially introduced GEM, its new generative advertising model, which has been operational on Facebook and Instagram since November 10, 2025. Early results show an average of ~5% growth in conversions on Instagram and about 3% on Facebook Feed.
But the change is much bigger than a few percentage points – GEM fundamentally changes the way ads are targeted, optimised, and displayed to users.

What is GEM in a nutshell?

GEM (Generative Ads Recommendation Model) is the "central brain" of Meta's ads—a huge AI model that analyzes signals, behavior, creatives, and user patterns in real time.
Instead of relying on manual settings or interests, GEM learns from Meta's entire advertising ecosystem – millions of ads, creatives, and interactions.

This allows it to predict much more accurately who will convert, when and with which creative.

How GEM works (simply and clearly)

1. Replaces manual targeting

GEM does not look at demographics or interest layers.
It analyzes real signals —actions, intentions, behavior, reactions to creatives—and shows the right ad to the right person at the right time.

2. Learns from all advertisers
The model doesn't just look at your data, it learns from global patterns across the entire platform.
That's why it's more accurate, faster, and more predictable than any previous Meta algorithm.

3. It runs on supercomputer chips

It was created in partnership with NVIDIA and uses the GH200 Grace Hopper Superchip.
This allows it to process billions of signals in real time.

4. It understands user intent

GEM analyzes a sequence of actions:

• before the ad;
• during the showing of the ad;
• after the person sees it.

This determines whether the user is in:

• the awareness stage;
  
• the consideration stage;
• the purchase stage.

What the data shows (results to date)

According to Meta, after implementing GEM:

• Instagram: ~5% more conversions;
• Facebook Feed: ~3% more conversions;
• Lower CPA and higher ROAS in campaigns with diverse creatives;
• Faster ad training and smarter budget allocation. (If you work in the industry, you know that Facebook usually likes one creative and bets 80%+ of the budget on it, without having the opportunity to validate the other creatives in the ad set). In this regard, we have seen a significant improvement.
  

What does this mean for your advertising strategy?

1. Your creative is now the most important factor

GEM evaluates not only what you advertise, but also how.
Diverse creatives—different formats, styles, messages, and angles—lead to more conversions because the model can "match" them more accurately to different people.

2. Broad audiences win

GEM works better with broad targeting.
Complex structures with 10 small ad sets confuse the algorithm and slow down training.

A more consolidated structure = more data for the model = better results.

3. The focus is shifting from control to automation

Manual settings (interests, micro-lookalikes, stacked audiences) are becoming increasingly ineffective.
Automation is no longer an "option" — it is the primary mechanism for effectiveness.

4. Tracking is critically important

GEM works with data.
If the pixel, CAPI, or server events are not implemented correctly, the results will not come.

5. There is an adaptation period

Training usually lasts 2–4 weeksdepending on:

• budget;
• traffic volume;
• signal quality;
• account history.

After this period, the model "finds its rhythm" and the results stabilize.

Practical tips for getting the most out of GEM

1. Ensure a variety of creatives – static, video, carousels, different messages.
2. Eliminate excessive segmentation – fewer campaigns, broader audiences.
3. Monitor the quality of events – no missed conversions or duplication.
4. Don't stop campaigns too early – give GEM time to learn.
5. Test creatives, not audiences GEM takes care of the audience.
6. Don't get bogged down in small details about creatives. This is definitely an own goal. Logo positioning, colour, and the like. It is much more important to have a variety of conceptually different advertising approaches (emotional, rational, customer testimonial, comparison, direct pain points, etc.).
   

Conclusion

Meta GEM AI is a fundamental change in Meta's advertising system.
The model increases conversions, reduces customer costs, and makes optimization smarter and more predictable. But success only comes if you are willing to adapt your structure, mindset, and creative strategy.

The old approaches—audience stacking, micro-lookalikes, 10 ad sets—no longer work as well.
The new world is simple: automation + quality signals + strong creatives = results.

If you want to extract real value from GEM and move to a structure that works with the new algorithm – not against it – contact me.

You will receive a clear plan, specific actions, and a strategy that delivers results from the very first month.