If your Meta campaigns suddenly got more efficient between May and September 2025 without you touching anything, you weren't imagining it. On November 10, Meta finally explained what happened: they'd been running GEM (Generative Ads Model) since Q2, and it's been delivering 5% more conversions on Instagram and 3% more on Facebook Feed.

GEM is Meta's largest AI model for advertising, trained on thousands of GPUs at the same scale as ChatGPT. The main difference from previous systems is that it learns from both your paid ads and organic content across Facebook, Instagram, and Messenger, then shares those insights everywhere. Before GEM, Instagram optimization didn't talk to Facebook optimization. Now they do.

These improvements happened automatically. No opt-in required, no campaign restructuring needed. Every advertiser benefits whether you're spending $500 or $500,000 monthly.

Why These Meta Ads Updates Actually Matter

A 5% conversion increase sounds small until you do the math. At $2,000 monthly spend generating 100 conversions, that's 5 extra conversions at the same cost. At $20,000 with 1,000 conversions, it's 50 extra. Multiply that across an entire year, and the numbers add up.

Meta also claims these benefits "doubled in Q3" after they improved GEM's architecture, though they haven't published specific Q3 numbers. Just that the model became more efficient at learning from data.

But these are platform-wide averages across millions of advertisers. Some campaigns probably saw 15% improvements. Others saw zero. Your results depend on what you were already doing, what industry you're in, and how much conversion data you're generating.

Also, Q2 and Q3 2025 included Prime Day, back-to-school shopping, and other seasonal events. Separating GEM's impact from regular seasonal lift is basically impossible unless you're comparing Meta against other channels you ran at the same time. If Meta outperformed Google Ads or TikTok by 3-5% during those months, GEM probably helped.

How GEM Works Differently

Previous Meta ad models treated each placement as a separate world. Instagram Stories had its own optimization system, Facebook Feed had its own, Reels had its own. They never shared information.

GEM changes that. When someone watches 30-second product demos on Instagram but scrolls past videos on Facebook Feed, GEM learns that person prefers Instagram for video content. Your video ads show more on Instagram, less on Facebook for that user. When someone browses products during weekday lunch breaks but only buys on weekend mornings, GEM adjusts delivery timing automatically.

Meta claims GEM is 4x more efficient at driving performance gains compared to their original ranking models. The knowledge transfer system works 2x better than standard AI training techniques.

The tradeoff? Less visibility into why performance changed. When your Instagram conversion rate jumps, was it GEM's optimization, your creative, or just seasonal factors? Hard to say definitively.
‍

What You Should Actually Do About These Meta Ads Updates

Don't restructure your campaigns. GEM is already running in the background, and making major changes could disrupt its learning.

Do give campaigns more time before adjusting. Meta now recommends waiting 7+ days before making significant changes, up from their previous 3-4 day guidance. GEM needs more time because it's processing signals across multiple surfaces.

Run campaigns across multiple placements instead of restricting to just Facebook Feed or just Instagram Stories. GEM performs best when it can transfer knowledge between all Meta properties. Let the algorithm decide where your ads work best rather than manually limiting delivery.

Use broader audiences. GEM works better with larger audience pools. If you're targeting narrow 10,000-person custom audiences, consider expanding to 50,000+ or testing Advantage+ campaigns that give Meta maximum flexibility to find converters.

Still track placement-level performance in Meta Ads Manager even though GEM optimizes automatically. You need to know which placements actually convert for budget decisions. If you're managing multiple ad platforms and want to automate cross-channel reporting, Dataslayer connects Meta Ads with 50+ other platforms to Google Sheets, Looker Studio, BigQuery, or Power BI.

The Downsides Meta Didn't Emphasize

Meta's announcement focuses on wins. Some realities they gloss over:

• Some advertisers saw temporary CPA increases in Q2 2025 as GEM rolled out and recalibrated. Most campaigns stabilized by Q3, but the transition wasn't smooth for everyone. If your May-June performance looked weird, that was probably GEM adjusting.
• You have less control over what's happening. When Instagram Reels conversions jump 8% while Facebook Feed drops 2%, is GEM redistributing delivery because the net result is better? Or is something else happening? The opacity makes troubleshooting harder.
• Not all campaigns benefit equally. GEM learns from conversion patterns. High-volume e-commerce with hundreds of weekly conversions gives GEM plenty of signal. Low-volume B2B campaigns with 10 monthly conversions? Less signal, probably less benefit. That 5% improvement likely skews toward high-volume advertisers.
• Meta is systematically removing manual controls. They've been eliminating detailed targeting options and pushing Advantage+ campaigns where Meta makes most decisions. GEM accelerates this trend. If you like controlling every aspect of your campaigns, the future isn't friendly.

How GEM Fits Meta's Bigger AI Push

GEM works alongside two other AI systems Meta built:

• ‍Meta Lattice handles ad ranking with trillions of parameters. It consolidated hundreds of separate models into one system, which Meta says improved ad quality by 8%.‍
• Meta Andromeda determines which specific ads each user sees based on their engagement history. When advertisers enabled Advantage+ creative features, Meta reported a 22% ROAS increase.‍

GEM acts as the "central brain" feeding insights to both systems while learning from organic and paid content across the entire Meta ecosystem.

This explains why Meta's Q2 2025 ad revenue hit $46.6 billion, up 21% year-over-year. CEO Mark Zuckerberg directly credited AI integration in advertising for the growth.

Common Questions About GEM

What exactly is GEM and when did it launch?

GEM (Generative Ads Model) is Meta's largest advertising AI, trained on thousands of GPUs at the same scale as large language models like ChatGPT. Meta announced it November 10, 2025, though it had been running since Q2 2025. Unlike previous models that optimized each placement separately, GEM shares learnings across Facebook, Instagram, Messenger, and other Meta properties while analyzing both organic content and paid ad signals.

‍

Are conversions really increasing 5%?

Meta reports 5% on Instagram and 3% on Facebook Feed starting Q2 2025, with improvements doubling in Q3. But these are platform-wide averages. Your results depend on industry, creative quality, existing optimization level, and conversion volume. High-volume e-commerce likely sees more benefit. Low-volume B2B campaigns probably see less. Also, Q2-Q3 included major shopping events, so separating GEM's impact from seasonal lift is difficult.

‍

Do I need to change my campaign settings?

No mandatory changes. GEM works automatically. You can potentially maximize effectiveness by running campaigns across multiple surfaces instead of restricting to single placements, allowing 7+ day learning phases instead of 3-4 days, using broader audiences instead of narrow 10,000-person segments, and avoiding constant manual bid adjustments. But none of that's required.

‍

Why is Instagram's improvement higher than Facebook's?

Facebook Feed has been Meta's primary ad surface for over a decade with mature optimization. Instagram, especially Reels and Shopping, had more room for improvement because previous models were less sophisticated. GEM's cross-surface learning found optimization opportunities specific to Instagram user behavior that previous models missed.

‍

Can I measure GEM's impact on my campaigns?

Sort of. Compare Q2 vs Q3 2025 performance while accounting for seasonal factors. Break down by placement to see Instagram vs Facebook differences. More importantly, compare your Meta performance against other ad channels during the same period. If Meta outperformed by 3-5%, GEM likely contributed. But isolating GEM's impact from creative changes, audience shifts, and seasonal variables is nearly impossible.

‍

What happens to manual targeting going forward?

Meta is reducing manual controls systematically. They've removed detailed targeting options and push Advantage+ campaigns where Meta controls most decisions. GEM accelerates this. Manual bid adjustments and placement restrictions may hurt performance by preventing GEM from optimizing. The shift: focus on creative quality and proper conversion tracking, let GEM handle delivery. If you like controlling campaigns completely, 2026 won't be great.

‍

Should I trust Meta's performance claims?

Meta has incentive to report positive results to convince advertisers to spend more and cede control to automation. The 5% and 3% figures are real platform data, but they're averages including campaigns with 15% improvements and campaigns with 0% improvements. Take them as directional, not guarantees. Best validation: compare your own Meta performance against other channels during the same period.

What This Means for 2026

GEM signals Meta's direction: more automation, less manual control, AI making decisions that used to require human judgment.

Expect Advantage+ to become the default for most campaign types. Creative diversity will matter more than detailed audience targeting. Provide multiple video formats, image ratios, and messaging angles, then let GEM decide what works where. Real-time optimization should improve as GEM learns faster from trending topics.

Advertisers who constantly adjust bids or restrict placements will likely see worse results than those who provide quality assets and trust the algorithm. Whether that trade-off works for you depends on your tolerance for giving up control.

If you're managing multiple ad platforms and want to see how Meta performs compared to other channels, try Dataslayer free for 15 days to automate cross-platform reporting in Google Sheets, Looker Studio, or your data warehouse.