AI-Generated Ads vs Human-Made Ads: What 300,000+ Ads Reveal

Do AI-Generated Ads Perform Better Than Human-Made Ads?

AI-generated ads perform as well as human-made ads on click-through rate and conversions, and the best AI ads, the ones that do not look AI-generated, perform better. On production speed, cost, and volume, AI is not close to human creative; it is in a different league.

That conclusion rests on field data from real ad delivery, backed by peer-reviewed experiments, not vendor anecdotes. The table below summarizes the strongest published evidence as of October 23, 2026.

The title's 300,000+ figure is the Taboola dataset alone: more than 300,000 live ads on Taboola's Realize platform, analyzed by the academic team. For what this evidence means for your budget, see our companion AI ad generator ROI study.

What Did the Taboola, Columbia, Harvard, TUM, and Carnegie Mellon Study Find?

The study found that AI-generated ads perform just as well as human-made ads, and that AI ads which do not look AI-generated achieve the highest engagement of any group. Taboola's report calls it one of the largest live studies of generative AI in advertising to date.

Taboola announced the results on January 28, 2026 in a report titled “AI Ads That Work: How AI Creative Stacks Up Against Humans.” The underlying academic paper, “AI in Disguise,” is by Yannick Exner, Jochen Hartmann, Ziqian Ding, Shunyuan Zhang, and Oded Netzer, and its latest version on SSRN (paper 5096969) was revised on August 12, 2026. It is also listed as a Marketing Science Institute working paper.

How the study was built

Taboola's GenAI Ad Maker, a free tool inside its Realize performance platform, lets advertisers generate ad images alongside their own human-made creative. The researchers worked with the platform's live ad data: more than 300,000 ads and, per the Harvard Business School listing, more than 16 billion impressions and 116 million clicks.

To isolate the effect of the creative itself, they used a “sibling ads” design. Each comparison pairs AI-generated and human-made ads created by the same advertiser, for the same campaign, on the same day, with the same objective and landing page. That design controls for the advertiser, timing, audience, and landing page, the variables that normally make AI-versus-human comparisons meaningless. The matched set contains 4,633 sibling ads spanning more than 369 million impressions and 2.5 million clicks.

The five findings that matter

1. AI matches human creative. In raw data, AI ads averaged a 0.76% click-through rate versus 0.65% for human ads. Under the tightest statistical controls, the two performed comparably; the latest paper reports no detectable average CTR disadvantage for AI-generated images against their human-made siblings.
2. Human-looking AI wins outright. AI-generated ads that did not “look like AI” achieved the highest engagement of all groups, significantly outperforming both human-made ads and AI ads perceived as artificial.
3. Faces build trust. A large, clear human face was one of the most important factors in making an ad feel human and trustworthy, and AI-generated ads were more likely to include that cue than human-made ones.
4. No quality trade. AI visuals increased or maintained click-through rates without reducing downstream conversion performance.
5. Early adopters cluster. Food and drink and personal finance brands were among the first to adopt AI ads.

Columbia Business School's Oded Netzer summed it up in Taboola's announcement: when AI is used to enhance human cues, “it doesn't just match human performance; it often sets a new ceiling for engagement.”

Why Do AI Ads That Don't Look Like AI Win?

AI ads that read as human-made win because people react against content they recognize as artificial, and modern AI is very good at producing images that do not trigger that reaction. Remove the tells, and AI's strengths in composition and clarity show up directly in click-through rate.

In the 2025 working version of the paper, raters judged 59% of the AI-generated ad images as human-made or were unsure, while they flagged 25% of genuinely human-made images as AI. In the model-free comparison, AI images rated as human-looking averaged a 0.79% CTR, ahead of human images that looked human (0.67%), AI images that looked artificial (0.62%), and human images that looked artificial (0.55%).

The visual signals are specific and actionable:

• Reads as human: sharper, clearer, less blurry images with larger faces. AI tends to produce exactly these features, which is how it “disguises” itself.
• Reads as AI: intense color saturation and high aesthetic scores, according to the 2026 versions of the paper.
• Getting better fast: the study's AI images came from a 2023 tool built on Stable Diffusion 2. When the authors regenerated a sample with Stable Diffusion 3 and 3.5, the images scored as less AI-looking.
• Small advertisers gain most: smaller and less experienced advertisers benefited relatively more from AI-generated images, which the authors describe as AI leveling the playing field.

The practical lesson is clear. Generic, oversaturated AI art underperforms; AI creative anchored in your real brand colors, real products, and real people outperforms. That is the design principle behind brand-trained creative, and each new generation of image models makes human-looking AI creative easier to produce. For how to turn one idea into several distinct, testable concepts, read our guide to AI ad angles vs variations.

What Do Other Field Experiments and Peer-Reviewed Studies Show?

Other rigorous studies point the same way: AI marketing images can beat professional human images in real ad delivery, better-trained AI keeps raising performance, and generative AI makes professionals dramatically faster.

International Journal of Research in Marketing (2025)

In “The power of generative marketing,” Jochen Hartmann, Yannick Exner, and Samuel Domdey prompted seven text-to-image models, including DALL-E 3, Midjourney v6, Firefly 2, and Imagen 2, to create 10,320 marketing images from 2,400 real human-made images. Across 254,400 human evaluations, AI-generated imagery surpassed human-made images in quality, realism, and aesthetics. When AI models and commissioned human freelancers received identical creative briefs, the best synthetic images also excelled on ad creativity, ad attitudes, and prompt following. In a field study with more than 173,000 impressions, AI-generated banner ads competed with professional stock photography and achieved up to 50% higher click-through rate than a human-made image.

Meta's AdLlama experiment (arXiv, 2025)

Meta researchers trained an ad-text model on historical ad performance and tested it against Meta's earlier imitation-trained model in a randomized 10-week experiment from February 16 to April 25, 2024, covering 34,849 U.S. advertisers and about 640,000 ad variations. Advertisers given the performance-trained model, AdLlama, saw a 6.7% increase in click-through rate (from roughly 3.1% to 3.3%) and created 18.5% more ad variations. This compares two AI models rather than AI and humans, and it shows a key point: AI ad performance is a moving target that rises as models learn from results.

Science (2023)

In a preregistered experiment published in Science, Shakked Noy and Whitney Zhang gave 453 college-educated professionals, including marketers, consultants, and grant writers, occupation-specific writing tasks. Participants with ChatGPT cut average completion time by 40% and raised output quality by 18%, as graded by experienced professionals who did not know which work used AI.

What Do Meta, Google, and OpenAI Report About AI Creative Results?

The major ad platforms report that their AI creative and automation tools lift conversions and are being adopted by millions of advertisers. These are platform-reported figures about each platform's own tools, so read them as strong directional evidence alongside the independent research above.

The 50 AI advertising statistics for 2026 add the newest platform numbers, including Meta's October 2026 GEM results and Google's AI Max adoption.

• Meta, 2024: more than 1 million advertisers used Meta's generative AI tools to create more than 15 million ads in a single month, and Meta estimated that businesses using Image Generation were seeing a 7% increase in conversions (Q3 2024 transcript).
• Meta, 2026: more than 8 million advertisers were using at least one generative AI ad creative tool by the first quarter, and advertisers using Meta's video generation feature saw more than 3% higher conversion rates in tests (Q1 2026 transcript). By the second quarter, more than 9 million small businesses used at least one AI creative tool, and image generation adoption more than doubled in the quarter (Q2 2026 transcript).
• Meta, October 2026: at Advertising Week New York, Meta began testing agentic actions in its Ads Manager business assistant that plan and create campaigns, generate images and copy, update targeting, and change budgets, and said advertisers adopting the assistant see 5% more conversions at the same cost per conversion (MediaPost). Every agent action depends on advertiser permissions and explicit approval before anything is published (BestMediaInfo).
• Google: Google says advertisers that activate AI Max in Search campaigns typically see 14% more conversions or conversion value at a similar CPA or ROAS, based on 2025 internal data for non-retail advertisers (Google Ads Help). AI Max combines search term matching with AI text customization and final URL expansion, so it measures automation broadly. An independent analysis of 250+ campaigns by Smarter Ecommerce found a median 13% revenue gain at a median 16% higher CPA (Search Engine Land).
• OpenAI: on October 5, 2026, OpenAI said ChatGPT reaches 1.2 billion people each week and published early partner results: DV Rockerbox found WeightWatchers' attributed CPA on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark, WorkMagic reported statistically significant lift for Dose with 67% of incremental purchases from net-new customers, and Triple Whale found 93% of Portland Leather's ChatGPT Ads visitors were new. These measure ChatGPT Ads as a channel, the newest place where AI-built creative competes.

The pattern across all three platforms is the same: AI creative is now standard infrastructure, and the advertisers who feed these systems more distinct, on-brand creative get more out of them.

Do AI-Generated Ads Hurt Conversion Rates?

No. The best available evidence shows AI-generated creative does not reduce conversion performance, and platform data shows AI creative tools lifting conversions.

In the Taboola study, AI-generated visuals increased or maintained click-through rates without reducing downstream conversion performance. The paper's authors tested ads with a purchase objective and found no significant effect of AI-generated images on conversion rate, measured as conversions per click. Taboola's Realize summary adds that the data showed no evidence of “curiosity clicks,” the fear that novel AI images attract clicks that never convert. On Meta, the platform's own estimate puts the conversion lift for Image Generation users at 7%.

Conversion rate depends on what happens after the click: the landing page, the offer, and how closely the page matches the ad's promise. That is where most teams lose the gains their creative earns. A dedicated, ad-matched landing page for every ad and audience closes that gap; our guide to custom landing pages and conversion rates shows how.

How Much Faster and Cheaper Is AI Creative Than Human Creative?

AI creative is dramatically faster and orders of magnitude cheaper than human creative. Controlled research shows a 40% time saving on professional writing, and the cost gap runs from dollars per AI asset to $100 or more per hour of agency design time.

• Speed: in the Science experiment, participants with ChatGPT finished tasks about 11 minutes faster than a control group that averaged 27 minutes, a 40% reduction.
• Cost: the IJRM authors conclude generative AI lets advertisers produce marketing content “not only faster and orders of magnitude cheaper” but at superhuman effectiveness levels. In G2's 2026 report, six of seven surveyed vendors said AI image generation is significantly less expensive than traditional design workflows or stock imagery.
• Human price benchmarks: Clutch reports most graphic design companies charge $100 to $149 per hour, and Upwork lists median freelance graphic designer rates of $15 to $35 per hour. Lapis Basic is $99 a month with a 25-credit allowance, about $4 per credit.
• Volume: Meta's advertisers produced more than 15 million AI-made ads in one month in 2024. According to IAB, 83% of advertising executives say their company has deployed AI in the creative process, up from 60% in 2024.

Volume is where speed and cost turn into performance. A human team that can afford three variants tests three ideas; an AI-equipped team tests a broad slate of distinct angles, audiences, and offers, then scales the winners. Our guides on why ad creative volume testing matters and AI ad creative testing at scale cover how to run that system without drowning in noise.

Where Does Human Creative Still Win?

Human creative still wins where AI output looks artificial, where a highly optimized human pipeline already exists, and in brand-building work where audience sentiment toward AI is skeptical. Each of these is a direction-and-review problem, not a reason to abandon AI.

• When AI looks like AI. In the Taboola paper's model-free data, AI images that looked artificial averaged 0.62% CTR, below the 0.67% of human images that looked human. The penalty is for the AI look, not for AI itself.
• When a human pipeline is already optimized. Ad arbitrage advertisers, who run volume-based businesses with established automated workflows for human-made ads, did better sticking with human images in the study period.
• When the audience is primed to notice. NielsenIQ's 2024 neuroscience study of more than 2,000 participants, about 150 measured with EEG, found consumers identified most of the AI-generated ads they were shown and rated them more annoying, boring, and confusing than traditional ads, with weaker memory activation even for ads perceived as high quality. NIQ noted the one ad consumers did not immediately flag was made by an advertising professional through considerable iterative editing.
• When sentiment is the risk. IAB's January 2026 research with Sonata Insights found 82% of ad executives believe Gen Z and Millennial consumers feel positive about AI-generated ads, while only 45% of those consumers do, and Gen Z is nearly twice as likely as Millennials to feel negatively (39% vs 20%) (IAB).

Read together, these findings say the same thing as the Taboola data: quality and authenticity decide the outcome. AI creative built from real brand assets and refined by a human editor captures the efficiency without the penalty.

Does Labeling an Ad as AI-Generated Hurt Performance?

Not for most audiences. IAB found that 73% of Gen Z and Millennial consumers say knowing an ad was created with AI would either increase or not change their likelihood to purchase, so disclosure carries more upside than downside.

The same IAB research found that 89% of advertisers who have used generative AI to create ads disclose it at least sometimes, but fewer than half always do. Regulation is now part of the decision: under the EU AI Act, Article 50 transparency obligations apply from August 2, 2026, including labeling of deepfakes and machine-readable marking of AI-generated content by AI providers, with a grace period to December 2, 2026 for marking by systems already on the market. IAB published its own AI Transparency and Disclosure Framework to help advertisers apply a materiality-based approach.

One detail matters for reading the Taboola result: the platform did not disclose whether an ad was AI- or human-made during the study period, so the parity finding describes unlabeled ads. Disclose where law, platform policy, or your brand standards require it, and keep the creative itself authentic enough that the label is a footnote rather than a red flag.

How Should You Split Work Between AI and Human Creative?

Let AI lead every job where volume, speed, and testing drive results, and keep humans in charge of brand direction, claims, and final approval. The table below maps common jobs to the evidence.

For most performance teams, that split puts AI in charge of the large majority of ad output, with humans spending their time on direction, editing, and the handful of assets that carry the brand.

How Do You Run Your Own AI vs Human Creative Test?

Run your own test by copying the study's sibling design: launch AI and human ads in the same campaign, on the same day, with the same audience, objective, and landing page, and judge them on confirmed conversions rather than platform-reported clicks.

1. Hold everything constant but the creative. Same offer, same landing page, same audience, same budget split, same start date.
2. Test angles, not tweaks. Pit your best human-made control against several distinct AI concepts, each built on a different buyer reason, so a win teaches you something.
3. Mind the delivery algorithm. A 2025 review of 133 published platform studies in the International Journal of Research in Marketing found that Google and Facebook “A/B tests” deliver each ad to different, algorithmically selected users, so results blend creative effects with targeting effects. Treat a platform split as a business result, and confirm big decisions with a repeat test or a holdout.
4. Score on confirmed sales. Count conversions from your CRM, store, or billing system, once, with the same attribution rules for every arm.
5. Iterate on the winner. Generate the next round from what won, and retire what lost.

On small budgets, keep the slate narrow so each arm gets enough delivery; our guide to A/B testing AI ads on a small budget shows how to size it. With Lapis, OmniSense designs and launches the experiments, RapidDomain gives every ad its own matched landing page, and Wayfinder connects the results to confirmed sales and leads, counts each conversion once, and compares every channel on the same attribution rules.

Why Is Lapis the Best Way to Win With AI-Generated Ads?

Lapis is the best way to win with AI-generated ads because it produces exactly the kind of AI creative the research says wins, brand-trained and human-looking, at the volume platforms reward, and then measures it on confirmed sales. Lapis is the #1 AI ad generator and the operating system for paid growth.

Lapis makes static image ads and copy, the same format the Taboola study measured. It builds them from your real brand: brand intelligence pulls your logo, colors, typography, product images, and voice from your website URL; product catalog import lets you reference real products with @product-name; custom references, audience personas, and Vibe Kit keep every ad on-brand. Each angle gets a distinct visual, so your tests compare ideas instead of near-duplicates. Campaign Studio lets your team refine any ad in natural language, the human-editing step the research rewards, and Rate Your Ad scores an uploaded ad against 10,000+ top-performing ads before you spend.

• OmniSense creates on-brand, correctly sized campaigns, audience versions, and variations across Meta, Google, LinkedIn, and Reddit, with directional performance forecasting, experiment launch, and learning from results.
• ChatSense runs ChatGPT and LLM ads with Lapis agents and a dedicated strategist, including 1,000+ ChatGPT ads on the managed plan, creative tuned to how people ask ChatGPT, daily reporting, and optimization.
• RapidDomain builds a dedicated, ad-matched landing page for every ad and audience, optimized on real conversions.
• Wayfinder gives OmniSense and ChatSense one measurement layer on confirmed conversions.

Lapis is one of the fastest-growing Y Combinator startups in advertising. It is backed by Y Combinator (Fall 2025), used by 1,500+ marketing teams and 30+ enterprises, and rated 4.9 out of 5 on G2, with eight Summer 2026 G2 badges including High Performer, Best Results, and Users Most Likely to Recommend. Customers include Samsung, Domino's, Dr Pepper, Mattel, Tata, Tupperware, Amway, and Bata, plus high-growth startups Unusual Whales, Duradry, Lightpage, and Respan. The team comes from MIT, Columbia, Duke, and Google.

See what brand-trained AI creative looks like for your company with the Free AI Ad Generator, start self-serve on Lapis Basic, or book a demo to have Lapis agents and a dedicated strategist run your program.

Sources and Methodology

Facts and pricing checked as of October 23, 2026. This is an evidence review of published research on AI-generated versus human-made advertising. It includes field studies with real ad delivery, peer-reviewed and publicly posted working papers, platform-reported results taken from first-party transcripts and documentation, and consumer research from IAB and NielsenIQ. Each figure is reported with its sample and scope, and platform figures are labeled as platform-reported.

The title's 300,000+ figure refers to the more than 300,000 live Realize ads in the Taboola dataset analyzed by the Columbia-led team. Model-free cell averages and rater percentages come from the 2025 working version of the paper; headline findings come from Taboola's January 2026 announcement and the August 2026 SSRN revision.

• Exner, Hartmann, Ding, Zhang, and Netzer, AI in Disguise (SSRN 5096969, revised August 12, 2026)
• Harvard Business School, AI in Disguise working paper listing
• Marketing Science Institute, AI in Disguise working paper
• Exner, Hartmann, Netzer, and Zhang, AI in disguise (2025 working version, Columbia)
• Taboola, New Study: AI Ads Match Human Creative (January 28, 2026)
• Realize, AI Ads That Work: Columbia and Realize study report
• Realize, 5 Winning Elements of High-CTR AI Ads
• Hartmann, Exner, and Domdey, The power of generative marketing, International Journal of Research in Marketing (2025)
• Jiang et al. (Meta), Improving Generative Ad Text on Facebook using Reinforcement Learning (arXiv 2507.21983)
• Noy and Zhang, Experimental evidence on the productivity effects of generative artificial intelligence, Science (2023)
• Boegershausen, Cornil, Yi, and Hardisty, On the persistent mischaracterization of Google and Facebook A/B tests, International Journal of Research in Marketing (2025)
• Meta, Q3 2024 earnings call transcript
• Meta, Q1 2026 earnings call transcript
• Meta, Q2 2026 earnings call transcript
• MediaPost, Meta Unveils Enhanced AI Business Assistant, Automated Ad Tools (October 7, 2026)
• BestMediaInfo, Meta tests agentic ad assistant (October 7, 2026)
• Google Ads Help, About AI Max for Search campaigns
• Search Engine Land, AI Max increases revenue 13% but drives higher CPA
• OpenAI, Building advertising for the way people use AI (October 5, 2026)
• NielsenIQ, NIQ Research Uncovers Hidden Consumer Attitudes Toward AI-Generated Ads (December 12, 2024)
• NielsenIQ, Will AI-generated advertising disrupt the creative industry?
• IAB and Sonata Insights, The AI Ad Gap Widens (January 2026)
• IAB, AI Transparency and Disclosure Framework (August 2026)
• European Commission, Transparency obligations under Article 50 of the AI Act
• G2, State of AI Image Generation in 2026 (August 1, 2026)
• Clutch, Graphic Design Company Pricing Guide (updated September 21, 2026)
• Upwork, Graphic Designer cost
• Y Combinator, Lapis company profile
• G2, Lapis reviews