AI advertising · Testing

Artificial Intelligence Ads vs Human-Made Ads: What Performance Tests Show

Three studies, three different answers. AI ads hold their own on clicks. On sales and memory the picture is mixed, so your own test matters more than any headline.

Ajay Kumar • •8 min read
Illustration of two bar charts comparing AI-made and human-made ads

Quick answer: do artificial intelligence ads perform better than human-made ads?

Tests so far show artificial intelligence ads match or beat human-made ads on clicks, but not reliably on sales. A 2026 study of 500 million impressions found similar click-through rates. A 2026 video experiment found more clicks and fewer conversions. Treat AI as equal on attention and unproven on persuasion, and test on profit.

What 3 real tests found

Most claims about AI ads come from tool vendors. We looked for tests with a stated method and a named publisher, and read three. Each measured something different.

TestWhat was comparedMain result
Taboola with four universities, January 2026Matched AI and human image ads in live campaigns, over 500 million impressionsClick-through about equal. AI slightly ahead in raw data
Hong Kong Baptist University, June 2026More than 150 video ads in a randomised field experimentAI ahead on clicks and retention, behind on conversions
NielsenIQ, December 2024Over 2,000 people viewing AI-made ads, with brain activity measured for about 150Viewers spotted most AI ads and remembered them less

None of these tests settles the question alone. Together they point the same way: AI ads win attention more easily than they win a sale.

If you are new to the topic, start with what artificial intelligence ads are and how they are made.

Clicks: AI ads match or beat human ads

The largest test is a field study announced by Taboola in January 2026, done with researchers from Columbia, Harvard, the Technical University of Munich and Carnegie Mellon.

It looked at hundreds of thousands of live ads on Taboola's Realize platform, with more than 500 million impressions and 3 million clicks. The researchers compared "sibling ads": an AI-made ad and a human-made ad from the same advertiser, in the same campaign, on the same day.

The study also reports that AI visuals kept or raised click-through without lowering downstream conversion.

Read the context

Taboola sells an ad platform with AI creative tools, and published the summary itself. Matching ads from the same advertiser, campaign and day is a fair way to compare, and the researchers come from four universities. Still, these are native image ads on publisher sites. Do not assume the same result for video on Instagram.

Conversions and memory: the evidence is mixed

Clicks are the easy part. Two other tests looked further down the funnel, and found problems.

Video ads: more attention, fewer sales

A paper from Hong Kong Baptist University, presented at the ISMS Marketing Science Conference in June 2026, ran a randomised field experiment with more than 150 video ads.

AI-generated video ads earned higher click-through rates, held viewers longer and drew fewer negative reactions. Yet they were less likely to produce conversions than human-created ads. The authors describe a tension between holding attention and persuading someone to buy.

Memory: weaker, even for good-looking ads

NielsenIQ research from December 2024 had more than 2,000 people watch AI-generated ads, and measured brain activity for about 150 of them.

This study is from 2024, and AI video tools have improved since. Its warning still applies: an ad can look fine and leave no trace.

Why the tests disagree

The results are not truly in conflict. They differ on three things:

How to run your own AI vs human test

Other people's tests tell you what is possible. Only your own tells you what works for your product and buyer. Keep it simple and fair.

  1. One idea, two executions. Use the same offer, the same message and the same landing page. Change only how the ad was made.
  2. Equal budget and audience. Use the platform's experiment tool so traffic is split evenly.
  3. Pick the metric first. Use orders or profit per ₹1,000 of spend. Do not use clicks.
  4. Fix the length. Decide the end date before you start, and do not stop early.
  5. Apply any required label. Test the ad as customers will see it.

On Google, Performance Max has an A/B test for assets, in beta. You choose a control set and a treatment set inside one asset group, and set the traffic split. Google recommends running it for at least 4 to 6 weeks, and locks the assets while it runs.

Read the result on profit

The trap is to stop at the first metric where one ad wins. Follow the numbers to the end.

Worked example: ₹1,00,000 behind each ad

Step 1: The click numbers

Each ad gets 2,00,000 impressions. These are example figures, not results from the studies above.

Clicks and cost per click
Human ad: 1.00% click-through = 2,000 clicks at ₹50
AI ad: 1.25% click-through = 2,500 clicks at ₹40

On this view the AI ad wins by 25%.

Step 2: The orders
Orders and cost per order
Human ad: 2,000 × 4.0% = 80 orders at ₹1,250
AI ad: 2,500 × 2.8% = 70 orders at ₹1,429
Step 3: The profit

Each order leaves ₹1,600 after product cost, shipping, tax and returns.

Profit after ad spend
Human ad: 80 × ₹1,600 − ₹1,00,000 = ₹28,000
AI ad: 70 × ₹1,600 − ₹1,00,000 = ₹12,000
Step 4: Is the gap real?

As a rough check on equal spend, the gap in orders should be more than twice the square root of the total orders.

Rough noise check
2 × √(80 + 70) = about 24 orders. The gap is 10.

So this test is not finished. At the same rates you would need about ₹6 lakh behind each ad before the gap is clear.

Our view

The AI ad may have cost ₹10,000 less to make. That saving is smaller than the ₹16,000 profit gap on just ₹1 lakh of spend. Production cost matters far less than what the ad does with the media budget.

For the full method of turning platform numbers into profit, see measuring profit, not platform ROAS.

What most guides get wrong about AI ads vs human ads

Articles on this topic usually pick one study and stop. Here is what that misses.

A working split between AI and people

The tests suggest a division of labour, not a winner. Give each side the work it does well.

JobBetter done byWhy
Sizes, backgrounds and versions of a proven adAIVolume at low cost, small risk
First-click image ads for cold audiencesAI or eitherClick results are level
Customer stories, demos and founder videosPeopleTrust drives the sale
The idea, the offer and the final checkPeopleAI cannot know what is true

Three habits help whichever way you lean:

If an AI-made ad shows a realistic person or scene, it may also need a label. Our guide to AI advertising ethics and disclosure covers when. For ideas worth testing in either form, browse the free ad examples library.

ScaleWin reports profit for every ad after product cost, shipping, tax and returns, so an AI-made ad and a filmed one are judged on the same line.

Key takeaways

  • On clicks, AI-made ads are at least level. One large study recorded 0.76% click-through for AI ads against 0.65% for human ads, and a tie under strict controls.
  • On conversions the studies disagree. One found no loss, and one video experiment found AI ads less likely to convert.
  • AI ads that do not look like AI did best. A large, clear human face was the strongest trust signal.
  • A 2024 brain-response study found viewers spotted most AI ads and remembered them less.
  • Judge your own test on profit per rupee of spend, and wait until the gap is bigger than normal variation.

Artificial intelligence ads: frequently asked questions

Do AI-generated ads get more clicks than human-made ads?

Often about the same, sometimes more. A 2026 field study on Taboola covering more than 500 million impressions recorded 0.76% click-through for AI ads and 0.65% for human ads in raw data. Under strict statistical controls the two were comparable.

Do AI ads convert as well as human-made ads?

The evidence is mixed. The Taboola study reported no drop in downstream conversion for AI image ads. A 2026 randomised experiment with more than 150 video ads found AI ads won on clicks and retention but were less likely to produce conversions. Test on your own sales.

Can people tell when an ad is made by AI?

Often, yes. In NielsenIQ research from December 2024, viewers identified most AI-generated ads without being told, and rated them as more annoying, boring and confusing. Tools have improved since then, but ads with an obvious artificial look still perform worse.

What makes an AI ad perform well?

Not looking like AI. In the Taboola study, AI ads that did not appear artificial had the highest engagement of any group. A large, clear human face was one of the strongest trust signals. Real product images and a specific offer help as well.

How long should an AI vs human ad test run?

Long enough for the gap to exceed normal variation. For asset tests in Performance Max, Google recommends at least 4 to 6 weeks. As a rough check, the difference in orders should be more than twice the square root of total orders before you call a winner.

Should I replace my human-made ads with AI ads?

No. Use AI for versions, sizes and quick tests of a proven idea, where the click evidence is good and risk is low. Keep people on customer stories, demonstrations and the final check, where trust decides the sale. Then let profit per ad decide the budget.

Sources

Ajay Kumar
Founder, ScaleWin · 12+ years in performance marketing

Ajay is the founder of ScaleWin. He works hands-on with Meta Ads, Google Ads and SEO for Indian D2C and service brands, and built ScaleWin so that the daily work of a performance marketer - reading numbers, pausing losers, scaling winners - happens automatically.

Drafted with AI assistance. How these articles are written

Let ScaleWin read your ad data every day

ScaleWin checks your Meta and Google Ads daily, judges every ad on profit and tells you exactly what to change.

Start free, no card

Get marketing tips that grow your business

One practical email a week on ads, SEO and growth. No spam, no fluff. Unsubscribe anytime.