The claim on Meta’s business site
Meta has published a case study from Nest Commerce, a UK performance agency, saying its in-house tool Nest AI Creative converted 20% better than non-AI creative. A separate cohort figure puts revenue growth at 70%.
The page is dated 19 October, with no year shown, but the site footer points to 2026. What the page doesn’t show matters more: no measurement period, no baseline for the revenue figure, and no definition of the benchmark the brands are said to beat.
What the numbers say – and don’t say
- 68% video share of spend against a 38% benchmark – an input, not an outcome.
- 93% vertical-first video adoption, with no denominator stated.
- 20% higher CVR for AI creative versus non-AI creative, with no conversion counting or attribution method.
- 70% revenue growth, with no period or revenue baseline.
- One health brand: 57% AI share of spend, 5% lower cost per result, 17% higher CVR.
- Another unnamed brand: 56% higher ROAS with no comparison group named.
There is also a selection problem. When Meta’s delivery system allocates spend across ads, the ads that rise to 40% or 57% of spend may already be favoured by the system. Higher conversion rates on those ads could reflect placement and audience selection as much as creative production. The page reports no holdout, randomised split or lift test.
The page adds that the cohort ran hundreds of active ads per account, but it never states the brands’ spend levels, so readers can’t tell how close they were to Meta’s ad caps.
Why this matters for performance teams
Meta has spent nearly two years making creative diversity a central message: give the ad system varied assets and let it choose. In its Q1 2026 results, Meta said more than 8 million advertisers used its AI creative tools, and a large-scale video generation test showed a 3% CVR gain. Nest’s 20% is dramatically larger, but it comes from a single agency’s own data.
The broader shift makes the missing controls more important. Meta’s Andromeda retrieval engine, announced in December 2024, narrows millions of ad candidates to thousands per auction, so the ad system decides far more than it used to. If you build creative strategy around that system, a holdout is the only way to know whether you are buying a real lift or simply following the algorithm’s allocation.
Recent incidents add context. REI’s AI-altered bicycle image and ODDITY Tech’s algorithm-driven cost spike show how automated creative and delivery systems can create review and dependency risks.
A simple test before you shift budget
Treat the Nest page as a direction of travel, not proof. Ask three questions of any vendor case study: what period, what baseline, where’s the control? Then run your own holdout. Keep one account or campaign on non-AI creative while testing AI-generated assets, and compare incremental conversions from the same window.
Creative diversity may genuinely improve results. Just don’t let a vendor’s own numbers replace a controlled experiment.
Source: PPC Land
