GEM Meta Ads: +18.9% conversions with the new unified AI model

Meta says its new AI model for advertising has increased conversions by 18.9% across all its platforms. If you have active campaigns on Facebook or Instagram, this change is already under way in your accounts, without you having done anything.

Does this Meta update change anything for my campaigns?

Yes, and automatically. GEM is already active on all Meta ad accounts: you don't need to do anything to turn it on, but you do need to know what to expect from your reports.

GEM, short for Generative Ads Model, is the generative AI model for advertising that Meta uses to decide which ad to show to which user, and when. Until recently, Facebook, Instagram and Reels each had their own separate system. Now there's a single central model that learns from all three together.

For anyone running campaigns, this means a more powerful optimization system working in the background. Meta reports an 18.9% increase in conversions and a 12% increase in click-through rate (CTR, the percentage of people who see an ad and click on it). The numbers come only from Meta and can't be verified externally, but they're consistent with the company's advertising revenue growth.

What really happened: what did Meta do with GEM?

Meta has replaced the separate optimization systems of Facebook, Instagram and Reels with a single central AI model, called GEM, which learns from all the data together and then trains the smaller models that serve ads in real time.

The structural change is significant. Before, each ad surface (Facebook Feed, Instagram, Reels) had its own prediction system. Each learned from its own data, sharing nothing with the others. GEM breaks down this separation: one huge model learns from all user behavior across every Meta app, then passes that knowledge to the smaller models that decide in real time which ad to show.

The new architecture is called Kunlun and replaces the previous one (called Interformer). According to Meta, Kunlun delivers twice the performance with the same amount of compute, and has enabled a 2.5x increase in model complexity compared to earlier versions. It's all described in the technical post published on Engineering at Meta.

GEM isn't brand new: the first public document dates back to November 2025. The current one is the third update in less than a year, and it consolidates earlier progress into a single, mature system.

One important caveat: Meta's numbers have no stated baseline. We don't know whether they come from A/B tests or internal evaluations, what period they cover, or how they're weighted. They're self-reported figures from a company talking about its own product. Read them as a signal of direction, not as a guarantee for your specific account.

To understand how to measure the true impact of paid campaigns on your business, our article on conversion lift and paid campaign measurement offers a practical method that doesn't depend on vendor claims.

How do these numbers compare with Meta's financial results?

Meta's financial results confirm real, steady growth: advertising revenue rose 27% in the second quarter of 2026, which makes GEM's impact plausible, even if it can't be verified in detail.

Meta's advertising revenue growth is documented in its official SEC filings (the U.S. Securities and Exchange Commission, which regulates financial markets). In Q2 2026, Meta reported $59.36 billion in advertising revenue, up 27% from the same period in 2025, according to the Form 8-K filed with the SEC. In Q1 2026, growth was even stronger: +33%, with $55 billion in revenue, as reported in the Q1 2026 Form 10-Q.

This growth doesn't depend on GEM alone, but the system is explicitly cited in the quarterly releases as one of the main drivers. In Q2 2026, Meta's CFO credited GEM and the user-understanding models with an 8.3% increase in ad clicks and a 15.7% increase in conversions on Facebook.

There is one discrepancy worth noting, though: the October 2026 technical post credits Facebook with a 5.8% conversion uplift for targeted users, while the Q2 2026 communication cited 15.7% by combining GEM with other models. The two figures don't necessarily contradict each other — they measure different populations, different periods and different attributions — but Meta doesn't explain the gap. If you read campaign reports, keep in mind that system-level figures don't automatically translate into +18.9% on your own account.

Three concrete scenarios: what changes for an e-commerce, a professional firm, and a manufacturing company?

GEM's impact varies depending on the campaign objective and the type of conversion you're optimizing for. Here are three illustrative scenarios to show where improvement is most likely.

Note: the calculations below are illustrative examples. The numbers are meant to show the logic of the reasoning and are not real client data.

Take an e-commerce store with a monthly Meta budget of €5,000 and a current cost per purchase of €25 (200 purchases a month). Even a 5% improvement in conversion rate would mean 210 purchases on the same budget — that's 10 extra sales without spending a euro more. This is the kind of calculation to run on your real account, not on the average figures Meta reports.

For manufacturing companies with long sales cycles, the direct impact is more limited. GEM does a good job optimizing fast, measurable conversions. If your sales process takes weeks or months, the system struggles to learn quickly enough. In that case, it's worth reading how AI agents can support campaigns with complex cycles.

What changes in practice when managing Meta campaigns?

GEM works automatically: there are no new settings to turn on. The practical change is in how you read results and how you structure campaigns to get the most out of the new system.

One aspect that's often underestimated: GEM optimizes better when campaigns use clear, measurable conversion objectives. If you're using traffic or reach objectives for campaigns you actually want to convert, you're leaving performance on the table. That's true with or without GEM, but with the new system the gap between well-configured and poorly configured campaigns tends to widen.

If you want to dig deeper into how Meta is evolving its creative tools alongside algorithmic optimization, the article on Meta's new AI tools for creative and ad insights offers a complementary view.

Can you trust Meta's numbers, or should you take them with a grain of salt?

The numbers are self-reported and can't be verified externally, but they are consistent with the official financial data. Read them as a signal of direction, not as a promise for your specific account.

The original source, PPC Land, says it plainly: Meta doesn't specify the measurement period, the method (A/B test or internal evaluation), or how the numbers are weighted. The 18.9% is an aggregate average across all surfaces and all objectives — which means some campaigns will have seen more, others less, and some nothing at all.

That said, there are signals that make the numbers plausible. Meta's advertising revenue growth is documented and verifiable. The average price per ad rose 12% in Q2 2026 and impressions rose 14%: if the system didn't work, advertisers would stop spending.

Our practical advice: don't wait for Meta to tell you whether GEM is working for you. Look at your own data. Compare conversions from the last 4 weeks with the same period last year, accounting for seasonality. If you see an improvement, great. If you don't, the problem may lie in your campaign structure, creatives or tracking — not in Meta's system.

Why it's in the news now: is the digital advertising market really growing?

Yes, and significantly. The Italian digital advertising market is worth €6.2 billion in 2025 and is heading toward €7 billion in 2026. If you invest in Meta, you're in a fast-growing market where AI systems like GEM are rewriting the rules of the game.

Here's the most important number for anyone managing a Meta budget: 83% of online advertising spend in Italy goes to international platforms like Meta and Google, according to the Osservatorio Internet Media at Politecnico di Milano. That means Meta isn't just one option among many. For most Italian companies, it's the main channel. So any improvement to Meta's optimization system has a direct impact on a huge share of the country's ad spend.

Across Europe, the market reached €131 billion in 2025, up 10.5%, according to IAB Europe. Retail Media, meaning advertising inside e-commerce platforms, grew 16.7%. It's a market in acceleration, where AI systems are becoming the main factor separating those who get results from those who don't.

How to get started: the concrete steps to get the most out of GEM

GEM is already live, but there are specific steps you can take to put your campaigns in the best position to benefit from it.

One step that's often overlooked: creatives. GEM optimizes ad delivery, but it can't make up for weak creatives. An ad that doesn't grab attention in the first 2-3 seconds won't be saved by the algorithm. For Christmas and year-end campaigns, Meta has published specific guides, which you'll find in our article on Meta's 2026 holiday guides and AI playbook for Christmas campaigns.

What are the most common mistakes to avoid with GEM?

The main mistakes aren't about GEM itself, but about how you set up the campaigns GEM has to optimize. A powerful system on weak foundations won't deliver results.

How can you tell if GEM is really improving your campaigns?

There's no 'GEM active' indicator in the Meta dashboard. You need to measure your campaign results with the right KPIs and compare them with previous periods.

The most useful comparison is year over year, not week over week. Short-term swings are often statistical noise. If after 4-6 weeks your cost per conversion has dropped and CTR has risen without you changing creatives or budget, GEM is probably contributing. For a more rigorous way to measure campaign impact, our article on conversion lift and measuring paid campaigns explains how to isolate advertising's real contribution from external factors.

So, is GEM really worth anything to your business?

Worth it if you have active Meta campaigns with clear conversion objectives, working tracking and enough budget to generate at least 50 conversions a week. In this case GEM is already working for you, and you should see an improvement in cost per conversion over the coming weeks.

Somewhere in between: you have active Meta campaigns but a limited budget or mixed objectives (awareness + conversion). GEM can still help, but the impact will be smaller. It's worth optimizing your campaign structure before expecting big performance jumps.

Not yet if you have no active Meta campaigns or very long, complex sales cycles (industrial B2B, large contracts). GEM is optimized for fast, measurable conversions: in these contexts, other channels or approaches may work better.

How we see it at SHM

For us, GEM is an example of how AI is becoming invisible in the daily work of people who run campaigns. The system optimizes in the background, with nothing for the ads manager to do. This is exactly the kind of automation that makes sense: it frees people from the routine of manual optimization and gives back time for the work that really matters: strategy, creatives, reading the data.

We believe AI should be integrated to free companies from routine, not to replace human judgment. GEM is a good example of this principle in action: the system does the heavy lifting of optimization, but the strategic decisions — which objective to pursue, which message to communicate, how much to invest — stay in people's hands.

The risk we see, though, is that the numbers Meta reports get taken at face value without verification. An average +18.9% across all campaigns worldwide says nothing about your specific account. Our approach with clients is always to look at real data, not the vendor's promises.

If you run Meta campaigns and want to know whether GEM is really improving your results, we can help with an analysis of your account. We work on Meta campaigns and Google Ads campaigns with an approach based on real data, not system averages. If you want to understand how to integrate AI into your marketing strategy in a structured way, our strategic AI integration service is the place to start. You'll find more insights on campaigns and performance in our Marketing and ADV performance section.

Frequently Asked Questions

GEM isn't a promise of results: it's a more powerful system that amplifies both well-built campaigns and configuration mistakes.