Meta GEM Lifted Conversions 18.9%, No Baseline Given

Meta says its GEM ads model lifted conversions 18.9% and click-through rate 12% across its ad surfaces, and it did not publish a baseline.

If you run high-ticket ads on Facebook and Instagram, this is Meta grading its own homework. Your numbers in Ads Manager may improve while your bank balance does not, and Ecommerce Paradise readers selling $1,500 and up products can’t afford to confuse the two.

The claim comes from a technical post on Engineering at Meta dated October 13, 2026. I’ll go through what Meta actually reported, how the number compares to earlier GEM claims, and what I’d do with a Meta account before the company reports earnings on October 21. If you are still deciding whether to build around paid social at all, start with my guide to what high-ticket dropshipping is.

Your ad account deserves a real company behind it, so form your LLC with Northwest Registered Agent, the best LLC formation service for ecommerce owners. Northwest has been doing this for roughly 25 years, and that tenure is why I trust it over cheap formation sites that upsell you every renewal. Form your LLC with Northwest →

Meta Says GEM Lifted Conversions 18.9% and CTR 12%

Meta published the post on Engineering at Meta on October 13, 2026, under seven named engineers including Huayu Li and Reva Srinivasan. According to the post, GEM delivered an 18.9% conversion improvement and a 12% ad click-through rate improvement across Meta’s ad surfaces and objectives.

The post does not state a baseline or a measurement window for either figure. It also does not say whether the numbers come from A/B tests or offline evaluation.

It does give a second set of figures with more context. For “targeted users” over the past year, GEM added a 5.8% conversion uplift on Facebook, 4.1% on Instagram and 1.6% on off-site conversions, per the same post. Meta does not define “targeted users,” and the post does not explain how those numbers sit alongside the 18.9% headline.

The rest of the post is about machinery. Meta describes a new architecture called Kunlun that it says doubles performance over the previous design at the same compute, and a 2.5x increase in model complexity at the largest pre-training scale. Training compute grew 4x, and the long engagement history the model reads doubled in production.

GEM now trains as one central model across Facebook, Instagram, Reels and Stories, instead of separate models per surface. It predicts click-through rate across surfaces and conversion rate across major objectives, including awareness, consideration, conversions, off-site conversions and app installs.

Meta’s pitch to advertisers is general. The company wrote that the result is “simultaneous ad performance gains across every surface and funnel stage,” and that it will “unlock better ad relevance and help advertisers connect with the right audiences more efficiently.”

What the post does not contain matters as much. There is no new Ads Manager setting, no pricing change and no new advertiser tool. The effects arrive through automated delivery, which means you can’t switch GEM on or off.

PPC Land covered the post on October 14 and flagged that the figures are vendor-reported and unaudited. Meta’s own earnings call is the next checkpoint: the company said it will release third-quarter results after market close on October 21, with a call at 4:30 p.m. Eastern, per its investor relations page.

GEM’s Claims Keep Changing: 5% in 2025, 18.9% Now

This is the third long GEM document in under a year, according to PPC Land, after posts on November 10, 2025 and August 3, 2026. Each one came with a bigger number.

The November 2025 post reported that GEM’s second-quarter launch lifted ad conversions 5% on Instagram and 3% on Facebook Feed. Meta called it the largest foundation model for recommendation systems in the industry and said it trained across thousands of GPUs.

PPC Land’s timeline of later disclosures shows how the figures moved. In January 2026 Meta reported a 3.5% lift in Facebook ad clicks and more than 1% more Instagram conversions. In April it credited GEM and a second system called Lattice with a landing page view conversion rate more than 6% higher.

In July 2026, per PPC Land, Chief Financial Officer Susan Li credited GEM and user understanding models with 8.3% more Facebook ad clicks and a 15.7% Facebook conversion uplift. The new post says 5.8% for Facebook conversions over the past year. PPC Land notes the post does not reconcile the two.

Here is the cleanest way to read it. Different teams at Meta are measuring different things over different windows, and the headline keeps growing because the scope keeps widening. One number is an A/B result on a launch. Another is an all-surface average.

The counterpoint on measurement

Independent measurement suggests the platform dashboard and the real world don’t line up neatly. Haus, a measurement vendor, analyzed 640 Meta incrementality tests run by its customers since the start of 2024 and published the results in July 2025.

Haus found Meta drove an average lift of about 19% to brands’ primary business metric across account-wide studies. For click-only seven-day direct-to-consumer attribution, it found Meta under-reported incrementality by about 15% on average. So Meta’s dashboard was, on average, conservative on that window.

The same analysis cut the other way for Advantage+. Advantage+ campaigns showed 2.4% higher platform ROAS than manual campaigns but 12% lower incremental ROAS, at 18% lower daily spend, and 58% of brands saw higher incremental ROAS on manual setups. Haus added an editor’s note in July 2026 that a re-run found Meta’s Incremental Attribution setting now outperforms standard attribution.

Caveats apply. The average advertiser in the Haus sample spends about $14 million a year on Meta, tests ran about 18.6 days on average, and the sample comes from Haus customers. A high-ticket store spending $10,000 to $50,000 a month is not that advertiser.

What Meta’s 18.9% Means for a High-Ticket Meta Account

My read is that GEM is real and probably helps, and that the 18.9% tells you almost nothing about your account. Here is how I’d think about it.

First, the model improves who sees your ad and when. That should help on high-ticket products over time, because a $2,000 purchase depends on finding the few people in market this month. Meta’s job is prediction, and better prediction is worth paying for.

Second, better prediction raises reported results before it raises sales, because the same system that picks the audience also does the counting. When one company delivers the ad and reports the conversion, you want a second opinion. I covered the tracking side of this in my breakdown of Google’s conversion lift tests, and the logic holds on Meta.

Third, high-ticket stores have an extra problem: the sale usually happens on the phone or days after the click. Meta sees the click and sometimes the order, and misses the quote request and the follow-up call. That is why I push people to log phone sales with a source field, using something like HubSpot and my free HubSpot setup guide.

Hypothetical math: when reported ROAS flatters you

This example is hypothetical, not from any real store. Say you spend $15,000 a month on Meta and Ads Manager reports 25 purchases at an average order of $1,800. That is $45,000 in reported revenue and a 3.0 ROAS.

Now suppose a holdout test shows only 60% of those purchases were incremental, meaning the other 40% would have bought anyway through search, email or a direct visit. That leaves 15 incremental orders, $27,000 in revenue and a 1.8 incremental ROAS.

At a 22% gross margin, a typical figure for furniture-type high-ticket products, those 15 orders produce about $5,940 in gross profit against $15,000 in spend. You would be losing roughly $9,060 a month while the dashboard says 3.0. Change the 60% to 100% and the same account makes $39,600 in gross profit on $45,000 of revenue, which is a very different business.

The Haus data says Meta under-reported on some windows, so your real number could land better than the dashboard. The point is that you don’t know until you test, and a bigger platform number makes the test more important, not less.

Scenarios and thresholds

If your Meta account has under $5,000 a month in spend, I would not touch your structure because of this announcement. The model change is automatic, and your sample size is too small to see an 18.9% effect through the noise.

Between $5,000 and $20,000 a month, run one clean holdout or geo test this quarter. Above $20,000, you should already be testing, and the GEM news is a reason to repeat the test with your current setup. After Meta’s past-buyer exclusion rollout and the AI ad agent test, the account you measured six months ago is not the account you run today.

One more consideration is where Meta fits. In my experience, Google Shopping and search capture buyers who are already looking, and paid social creates demand. Both matter, but their numbers should not be compared at face value. My guide on why high-ticket Google Ads need a complete conversion system covers the other half of that stack.

If your ads are going to a store that is half built, no model improvement will rescue it. That is the case for letting someone else build the base. My team can build and run the whole store through the turnkey done-for-you service, so the ad spend lands on a page that converts.

Want my team to build and run your high-ticket store, with ad tracking set up from day one? See the turnkey done-for-you service →

Audit Your Meta Account Before Meta’s Oct. 21 Earnings

Here are five moves I’d make this week, in order of effort.

Export 90 days of Meta purchases and compare them to orders by source in Shopify. If Ads Manager claims 120 purchases and Shopify shows 70 orders with a Meta source, you have a 1.7x over-count to explain before you trust any model improvement.
Run a holdout test for two to three weeks using Meta’s built-in conversion lift tool, or a geo split where you pause ads in a few states. Pick a window without a major sale, and write down the expected lift before you start so you don’t talk yourself into the result.
Add a “how did you hear about us” question to your order confirmation flow in Klaviyo. Self-reported attribution is messy, but on a $1,800 order it catches phone and word-of-mouth sales that pixels miss.
Track profit per channel, not revenue per channel. A tool like Finaloop can pull ad spend and costs together so you see margin after ads, and you can hand the weekly reconciliation to a VA from OnlineJobs.ph.
Cap budget increases at about 20% a week through Black Friday. Better delivery can pull spend up fast, and my Black Friday marketing strategy guide covers how to stage budgets before the peak.

Paste the Meta export and the Shopify export into Claude and ask it to flag days with high spend and no matching orders. I wrote up how I use it for this kind of work in my guide to building a leaner store with AI.

Creative still decides most of your results. If you need fresh ads fast, my walkthrough on making UGC-style ads without filming pairs well with Canva for static variations. If you want a second set of eyes on your numbers, join the community below and share your Ads Manager and Shopify exports.

Frequently Asked Questions

What is Meta’s GEM model?
GEM is the Generative Ads Recommendation Model, which Meta describes as its largest foundation model for ads recommendation. It decides which ads people see on Facebook and Instagram, per Meta’s 2025 introduction.

Does the 18.9% apply to my ad account?
No. Meta reported it as an average across surfaces and objectives without a baseline. Your result depends on your creative, offer, price point and tracking, so test it yourself.

Can I turn GEM on or off?
No. The October 13 post announces no Ads Manager setting, and the effects arrive through Meta’s automated delivery. What you control is budget, creative, audience signals and how you measure.

Should I raise my Meta budget because of this?
Not on this announcement alone. Wait for a holdout result, and watch what Meta says on its October 21 earnings call. My piece on budget as the new quality score explains why spend level now shapes delivery.

How is this different from TikTok or Google ads?
Each platform grades itself with its own attribution. Meta recently banned TikTok ads on its apps, which I covered in what store owners must do now, so compare channels with your own order data.

Where do I find high-ticket niches worth advertising?
Start with my free niches list, then check margins with the high-ticket niches breakdown before you spend a dollar on traffic.

What if my ads send traffic to chargeback-prone orders?
Better targeting can bring in more first-time buyers, and that can raise disputes. Read my guide to chargeback prevention for high-ticket stores and set up evidence collection first.

Want to compare your Meta numbers with other store owners and me inside the community? Join the Skool community →

Meta will keep publishing bigger numbers, and some of them will help your account. Your job is to know which ones reach your bank. Subscribe to the YouTube channel for daily breakdowns. More breaking news coming through the day.

If this was useful, these go deeper:

Trevor Fenner

Trevor Fenner is an ecommerce entrepreneur and the founder of Ecommerce Paradise, a platform focused on helping entrepreneurs build and scale profitable high-ticket ecommerce and dropshipping businesses. With over a decade of hands-on experience, Trevor specializes in high-ticket dropshipping strategy, niche and product selection, supplier recruiting and onboarding, Google & Bing Shopping ads, ecommerce SEO, and systems-driven automation and scaling. Through Ecommerce Paradise, he provides free education via in-depth guides like How to Start High-Ticket Dropshipping, advanced training through the High-Ticket Dropshipping Masterclass, and fully done-for-you turnkey ecommerce services for entrepreneurs who want a faster, more hands-off path to growth. Trevor is known for emphasizing sustainable, real-world ecommerce models over hype-driven tactics, helping store owners build scalable, sellable, and location-independent brands.