Photo: Yael Yañez / Unsplash
CUTTING THROUGH
THE AI HYPE
Everyone's demoing AI. Almost no one's measuring it.
Open LinkedIn on any given day and you'll see another "look what AI made" post. Every platform has a generative tool now. Every vendor has a reel. The demos keep getting slicker — and they keep dodging the only question a performance marketer actually needs answered: does it work, at what cost, compared to what you were doing before?
That's not a knock on the technology. It's a gap in the conversation. A demo shows you what's possible once. Running paid media needs to know what's true at scale, across a thousand ad accounts, next quarter as well as this one.
So we went and found the actual data. Not vendor reels — named studies, platform-published results and peer-reviewed research, cross-checked against primary sources, not just search-result summaries. Where the industry's hype answer and its data answer disagree, this report sides with the data, chapter by chapter: image generation, video generation, creative testing at scale, and the real gap between what AI promises marketers and what it's currently delivering.
Consider this the companion to At the Breaking Point — same rigor, narrower question. Not "is AI happening." Is it actually paying off, and for whom.
Photo: Octavian Rosca / UnsplashTHE IMAGE
EQUATION
AI image generation is everywhere. Ad creative production is one of the last places it's actually landed.
Marketers have embraced AI image tools in general. Ad creative production specifically is still catching up.
Marketers trust AI to optimize a media budget. They're still hesitant to let it touch the creative itself.
That 48-point gap between optimization and creative production is the whole story of this chapter. The tools are mature. Trust in what they produce for actual ad creative isn't — yet.
AI image generation for advertising is a real, fast-growing category. It's just not a huge one yet, in absolute dollars.
The global AI image generator market was valued at $349.6M in 2023 and is projected to reach $1,081.2M by 2030 — a 17.7% compound annual growth rate, with advertising named alongside e-commerce and gaming as a major adopter industry.
Real performance data on AI-generated images is scarce. What exists says: yes, when it's done well.
A large-scale study by Taboola, with researchers from Columbia Business School, Harvard, TU Munich and Carnegie Mellon, compared matched pairs of AI-generated and human-made ads across 500M+ impressions and 3M+ clicks on Taboola's network. AI-generated ads slightly outperformed on click-through rate — and performed comparably once the tightest statistical controls were applied.
Ads where GenAI was used seamlessly landed in the top tier for branded cut-through more than 40% of the time — but GenAI ads overall showed lower branding on average, dragged down by cases where the AI use was obvious.
Kantar, "Rethinking AI-generated advertising," Nov 2025Consumers are more bothered by AI-generated ads than marketers themselves are: 41% of consumers say AI-generated ads bother them, versus just 29% of marketers who feel the same.
Kantar, Media Reactions 2024AI doesn't remove the work of image production. It relocates it.
45% of ad professionals spend 1–4 hours a week reviewing AI-generated outputs; 6% spend more than 12. That's not a reason to avoid AI image generation — it's a reason to build review into the workflow instead of pretending it isn't there. It's exactly the layer QuickAds builds around image generation from day one: not a raw model, a production system.
TripleLift, May 2026"Pure magic for marketers — the kind of tool you can't imagine working without."
Photo: Jakob Owens / UnsplashTHE MOTION
MULTIPLIER
Video was AI creative's hardest problem. It's now its fastest-growing use case.
Two years ago, AI video generation was a novelty act. It's now closer to standard practice than not.
Brand use of AI for video production more than doubled in a single year — and it's starting to show up in performance, not just usage.
Advertisers using Meta's video generation tools saw a 3%+ lift in conversion rates in large-scale testing. Zoom out further and McKinsey estimates generative AI could lift marketing productivity by 5–15% of total marketing spend — worth roughly $463 billion a year.
Meta, Q1 2026 earnings call · McKinsey, "How generative AI can boost consumer marketing," Dec 2023This is a real QuickAds-generated ad — the same motion-and-image pipeline behind the stats in this chapter.
"We've worked with many creative and performance partners, but what stood out here was the speed and precision." — 3.5× conversion rate increase, 68% faster creative turnaround.
Ramji Sundararajan, President, Udemy — QuickAds customer
Photo: Agence Olloweb / UnsplashTikTok Symphony took American Eagle from brief to six campaign-ready AI-generated videos across 33 SKUs in under 24 hours — driving 60% higher traffic ROAS and 8,000+ conversions.
TikTok's own disclaimer: results are brand-reported, not independently verified by TikTok. Included here as a platform case study, not a QuickAds result.
Video-specific AI spend is still a fraction of total digital video — but it's the fastest-growing slice of a market that's already large.
Photo: Ruchindra Gunasekara / UnsplashTHE SCALE
TEST
One good ad was never the hard part. A hundred good variants is.
The brands winning right now aren't just making ads faster. They're making dramatically more of them, and testing nearly all of it.
When Amazon turned on AI-driven creative optimization across Performance+ and Brand+, every core metric moved in the same direction.
This is what "AI at scale" looks like when it's measured properly, on a platform's own advertiser base rather than a single case study.
Photo: Luis Quintero / UnsplashL'Oréal ran the same search campaigns with AI Max turned on. The results weren't incremental.
On the creative-ranking side, Meta's Generative Ads Model (GEM) delivered a 5% increase in conversions on Instagram and 3% on Facebook Feed after rollout — evidence that the AI doesn't just generate more creative, it also gets better at knowing which creative to show whom.
Meta Engineering blog, Nov 2025Even with all this volume, performance still concentrates — which is exactly why testing matters more than raw output.
The top 2% of creatives still drive 53% of total ad spend in gaming, versus 43% in non-gaming — a widening gap that signals broader testing, not just more assets, is what's actually shifting outcomes.
AppsFlyer, April 2025Over 80% of marketers agree that early-stage creative research and consistent testing lead to better campaign outcomes; more than 70% value AI's ability to predict which creatives are worth testing in the first place.
And the companies actually capturing AI's value share a trait: McKinsey found high performers are nearly 3× more likely than others to have fundamentally redesigned their workflows when deploying AI — the single strongest correlate of business impact out of 31 factors tested.
Kantar, "US Media Reactions 2025" (n=974 marketers) · McKinsey, "The State of AI in 2025," Nov 2025 (n=1,993)This is QuickAds' entire premise — not "generate an ad," but know which one to generate before you spend a dollar testing it.
QuickAds' Creative Intelligence layer runs on 30M+ ads analyzed across virtually every platform and format, powering 500K+ ads generated for 30,000+ brands across 120 countries to date.
"QuickAds turned our ad workflow from a grind into a growth loop."
72% less time spent on ad creation, freeing the team to test more, not just ship faster.
Photo: Thilak Mohan / UnsplashTHE HYPE
GAP
Where AI creative actually stands, between the demo reel and the disillusionment.
95% of enterprise generative AI pilots fail to show measurable financial impact.
Only about 5% achieve rapid revenue acceleration — the rest stall with little to no measurable P&L impact. This is the headline finding of MIT's most-cited 2025 study on enterprise AI, and it applies well beyond marketing. But marketing has a particular stake in it: more than half of generative AI budgets are directed at sales and marketing tools — while the biggest realized ROI so far has actually shown up in back-office automation, not marketing.
Abandonment rates aren't shrinking as the technology matures. They're growing.
Gartner predicted at least 30% of generative AI projects would be abandoned after proof-of-concept by the end of 2025, citing poor data quality, unclear ROI and escalating costs. A separate study found the share of companies abandoning most of their AI initiatives jumped to 42% in 2025, up from just 17% the year before.
Despite average spend of $1.9M on GenAI initiatives, fewer than 30% of AI leaders say their CEOs are happy with the return.
88% of organizations report regular AI use in at least one business function, up from 78% a year earlier — but only 39% report any enterprise-level EBIT impact from it.
Gartner's own Hype Cycle already made this call. Generative AI moved from the Peak of Inflated Expectations in 2023 into the Trough of Disillusionment in 2024 — and by 2025, the hype itself had moved on, to AI agents and AI-ready data, as GenAI matures past being the industry's shiny new thing.
Marketers are candid about it. 51% say they're still in the "testing" phase of their own AI approach. And when Marketing Brew asked sitting CMOs to name 2025's most overhyped marketing trend, several pointed straight at AI-as-strategy itself.
People are talking a lot about AI as the strategy, but AI isn't the strategy.
None of this means AI creative can't pay off. It means it pays off unevenly — and the clearest wins are documented, not anecdotal.
Forrester's Total Economic Impact study of Adobe's Firefly-powered Creative Solutions found a composite enterprise organization achieving 461% ROI over three years in the medium-impact scenario, with hero-asset creation up to 60% faster and asset-variant production scaled 70–80%. This is a vendor-commissioned, composite-organization projection, not an audited industry average — but it's the most rigorous public ROI modeling available for this category.
Separately, Gartner found early GenAI adopters self-report averaging a 15.8% revenue increase, 15.2% cost savings, and 22.6% productivity improvement.
It's undirected AI versus AI pointed at a specific, measured workflow. That's the entire design premise behind QuickAds: hybrid AI-and-human, built around ad performance data rather than general-purpose generation — not chasing whatever's at the peak of the hype cycle this quarter.
Photo: Fabian Kleiser / UnsplashTHE
RECEIPTS
Three ways to actually land on the right side of the gap.
Treat AI as a volume unlock, not a one-shot bet.
The brands seeing the clearest gains treat AI as a way to produce dramatically more tested creative, not a faster way to make the same one ad. AppsFlyer's benchmark of 2,365 creative variations per quarter, and Amazon's advertisers running 5× more products with 2× the images per product, are the new baseline — not the ceiling.
"Five weeks in, CAC is down 55%, repeat purchase is up 2.4×."
Redesign the workflow. Don't bolt AI onto the old one.
McKinsey's finding is the clearest signal in this whole report: high performers are nearly 3× more likely than others to have fundamentally redesigned their workflows when deploying AI — the strongest correlate of real business impact, ahead of model choice, budget, or headcount.
QuickAds helped us translate 150 years of craftsmanship into creative that actually performs on Meta and Google — without losing the soul of the brand.
Keep humans on judgment. Let AI handle production.
Kantar's finding holds the whole report together: seamless AI wins, obvious AI underperforms. And the "review tax" — 45% of ad pros spending real hours a week checking AI output — isn't a failure of the technology. It's the cost of doing this responsibly, and it's smaller than the cost of getting it wrong in public.
The Taboola/Columbia study's core finding is the closest thing to an industry-wide, rigorously verified "yes, it works" data point available today: AI-generated ads matching or slightly exceeding human-made ads on real performance, across 500M+ impressions. That's the hardest evidence in this report, and the reason it exists.