AI in marketing: what the evidence shows about content, ads, research and search

Marketing was one of the first business functions to adopt generative AI, and it is easy to see why. So much of the work is words and pictures: product descriptions, ad variations, emails, social posts, campaign images. Tools that produce all of these in seconds look like a gift to any team that is short on time and budget, which is most of them.

But "AI can make an ad" and "AI makes better marketing" are different claims. This article sorts the evidence into four areas where AI is changing marketing (content, advertising, market research, and how customers find you), shows what the best studies found in each, and flags where the results come with real caveats. It ends with practical questions you can apply to your own team this week.

Where marketing teams are in 2026

Spending is ahead of readiness. In Gartner's 2026 CMO Spend Survey of 401 marketing leaders in North America, the UK and Europe (fielded January to March 2026, mostly at companies with over US$1 billion in revenue), CMOs said they were putting an average of 15.3% of their marketing budgets into AI initiatives. Yet only 30% reported mature or fully developed AI capabilities, and 70% said their internal processes were not yet mature enough to scale AI¹. Overall marketing budgets barely moved, at 7.8% of company revenue versus 7.7% in 2025¹. Keep in mind this is survey data from large firms, not a measure of results, and it may not reflect smaller businesses.

The platforms are also pushing AI into the workflow whether teams ask for it or not. In mid-2025 the Wall Street Journal reported that Meta planned to let brands create and target ads with its AI tools by 2026: an advertiser would supply an objective and a budget, and Meta would generate the creative and choose the audience¹³. That plan was reported, not announced as a finished product, but it captures the direction of travel. The question for most marketers is less "should we use AI?" than "where does it actually help, and where do we need a human?"

Content creation: faster, often better, but more alike

The cleanest evidence on AI and everyday marketing writing comes from a randomized experiment by MIT economists Shakked Noy and Whitney Zhang. They gave 444 college-educated professionals, including marketers, grant writers, consultants and HR staff, realistic writing tasks from their own occupations, and randomly gave half of them access to ChatGPT. Those with access finished about 10 minutes faster, a 37% reduction, and their work was rated about 0.45 standard deviations higher in quality by evaluators in the same fields². A standard deviation is a measure of spread. An improvement of almost half of one is a meaningful jump, roughly enough to move a middle-of-the-pack writer up to around the 67th percentile. The tasks were short and self-contained, which is exactly the kind of work where AI does best.

There is a catch, and it matters for brands. In a 2024 study in Science Advances, Anil Doshi and Oliver Hauser had 300 people write short stories, with some given AI-generated ideas to start from, and had 600 readers rate the results³. AI help made individual stories more novel and more useful, and the gains were largest for the least creative writers: with access to five AI ideas, their novelty rose 10.7% and usefulness 11.5%. But the AI-assisted stories were also more similar to one another than stories written without AI³. The authors frame this as a social dilemma: each writer is better off using AI, but the collective body of work becomes less diverse.

For marketing, the implication is direct. If you and your competitors prompt similar tools in similar ways, your copy will drift toward the same middle. AI raises the floor of quality, but distinctiveness, which is much of what a brand is for, still has to come from somewhere else: your customer insight, your point of view, your editing.

Advertising: AI images can win, if they don't look like AI

Visual ads are where the most striking recent evidence sits. In a peer-reviewed 2025 study in the International Journal of Research in Marketing, Jochen Hartmann, Yannick Exner and S. Domdey (Technical University of Munich) used seven image generators to create 10,320 marketing images from 2,400 real ones, then collected 254,400 human ratings. AI-generated imagery could surpass human-made images in quality, realism and aesthetics. In a field test with more than 173,000 ad impressions, AI-generated banner ads competed with professional stock photography and achieved an up to 50% higher click-through rate than a human-made image⁴. Note the "up to": that is the best case in the test, not an average.

A larger follow-up study adds an important twist. Exner, Hartmann, Oded Netzer (Columbia) and Shunyuan Zhang (Harvard) analyzed data from the ad platform Taboola, covering more than 16 billion impressions, and focused on 4,633 "sibling" ads from the same advertiser, campaign, day and landing page, some with AI images and some with human-made ones, which mimics an A/B test (a controlled comparison of two versions)⁵. On average, the AI and human images performed about the same. But AI images that people did not recognize as AI-made outperformed, and the more an AI image "looked like AI," the lower its click-through rate. Raters judged more than 45% of the AI ads as likely human-made⁵. In this setting, consumers were not told which images were AI-generated.

What happens when they are told? A 2025 working paper by Hyesoo Lee and Anindya Ghose (NYU Stern) and Vilma Todri and Panagiotis Adamopoulos (Emory) combined a lab experiment with a field study on Google's display network. Ads created entirely by generative AI raised click-through rates by up to 19% in the field, while ads where AI merely modified a human design showed no significant improvement. But disclosing AI involvement reduced advertising effectiveness by up to 31.5%⁶. This is not yet peer reviewed, so treat the exact numbers cautiously.

Put together, these studies say AI creative can perform at least as well as human creative, but the advantage is fragile. It depends on quality, on whether the output looks artificial, and on what the audience knows. That last point is becoming a practical and legal question as labeling rules spread.

Personalization and research: powerful tools with sharp edges

Personalized persuasion. Tailoring messages to individuals used to be expensive. In a set of four studies with 1,788 participants published in Scientific Reports in 2024, Sandra Matz and colleagues showed that messages ChatGPT tailored to a person's psychological profile (such as personality traits or political ideology) were more influential than generic ones, across consumer products like iPhones and sneakers and across political appeals⁷. The effects were measured in self-reported attitudes, not actual purchases, and the authors explicitly flag the risks: the same capability can be used for manipulative political messaging or to encourage compulsive buying⁷. For marketers, it is a reminder that the line between relevance and manipulation is one you will need to draw deliberately.

Synthetic research panels. One of the newest uses is "synthetic consumers": asking a language model to role-play survey respondents. A 2025 preprint by researchers at PyMC Labs and Colgate-Palmolive tested this on 57 real product-concept surveys answered by 9,300 people in the US. Instead of asking the model for a number on a 1-to-5 scale, which produced unrealistic answers, they had it answer in words and mapped those words to ratings. The synthetic results reproduced the ranking of products well enough to reach about 90% of the agreement you get when you compare two halves of a real human panel⁸. That is impressive, but the limits are spelled out by the authors themselves: all 57 surveys were in one category (personal care), the method was tuned on these surveys, and results for some groups, such as by gender or region, were not reliably reproduced⁸. Synthetic panels look useful for screening many early ideas cheaply. They are not yet a replacement for talking to real customers, especially new or niche ones.

The "AI" label itself. Your customers may not share your enthusiasm. In experiments with more than 1,000 US adults across eight product and service categories, Mesut Cicek, Dogan Gursoy and Lu Lu found that describing a product as using "artificial intelligence" consistently lowered purchase intentions, by reducing emotional trust, with stronger effects for higher-risk products¹¹. Leading with "AI-powered" in your messaging may cost more than it gains.

Search and discovery: the ground is shifting

The biggest long-term change for marketers may not be how they make content, but how customers find it. Search engines increasingly answer questions with AI-written summaries. Pew Research Center tracked the actual browsing of 900 US adults in March 2025. About 18% of their Google searches produced an AI summary. When a summary appeared, users clicked a traditional result in 8% of visits, versus 15% when there was none, and clicked a link inside the summary in just 1% of visits. They were also more likely to end their browsing session right there: 26% versus 16%⁹. This is observational data from one month and one search engine, so it shows an association, not proof that summaries cause fewer clicks, but the direction is clear.

A new practice has grown up in response, often called generative engine optimization (GEO): shaping content so AI systems cite it. The idea was formalized in a 2023 research paper that found certain content changes, such as adding citations, quotations and statistics, could boost a source's visibility in AI-generated answers by up to 40%, with effects varying by subject¹⁰. Treat specific GEO tactics sold today with care. The research base is small and the systems change constantly. The durable advice is older: be the clear, credible, well-sourced answer to the questions your customers actually ask.

The guardrails: trust, law and the limits of the evidence

AI also makes some old forms of deception cheap. In the US, the Federal Trade Commission's rule on consumer reviews and testimonials took effect on October 21, 2024. It bans creating, buying or knowingly spreading fake reviews, explicitly including AI-generated reviews attributed to people who don't exist, and lets the agency seek civil penalties against knowing violators¹². The practical rule is simple: if a real customer didn't write it, don't present it as if they did.

It is also worth being honest about the evidence base. Much of what circulates about "AI in marketing" is vendor marketing or industry surveys of adoption, not measured outcomes. The strongest studies above are mostly short tasks, single platforms, or self-reported attitudes. Few track long-run brand effects, and none can tell you how your specific audience will react. That is a reason to test, not a reason to wait.

What this means for you

1. Start with high-volume, low-risk content. Product descriptions, ad variations and first drafts are where AI has the clearest productivity evidence². Keep humans on strategy, positioning and final edits.
2. Protect distinctiveness. Give the AI your own customer insights, examples and brand voice rather than generic prompts, and have an editor check whether the output could have come from any competitor³.
3. A/B test AI creative against your best human work. Measure clicks and conversions, not impressions of quality, and screen out images that look obviously synthetic⁴⁵⁶.
4. Decide your disclosure policy before regulators or customers decide it for you. Know the labeling rules in your markets and the evidence that disclosure can reduce response⁶¹¹.
5. Use synthetic panels to narrow options, then validate with real customers before big decisions⁸.
6. Audit how AI search describes you. Ask a few AI assistants the questions your customers ask and see whether, and how, your brand appears⁹¹⁰.

Sources

1. Gartner. (2026, May 11). Gartner 2026 CMO Spend Survey finds CMOs allocate 15.3% of marketing budgets to AI, but only 30% are ready to scale AI capabilities (press release, survey of 401 marketing leaders, January to March 2026). gartner.com newsroom, 11 May 2026
2. Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187–192. doi.org/10.1126/science.adh2586. Figures cited from the authors' working paper version (March 2, 2023): economics.mit.edu/sites/default/files/inline-files/Noy_Zhang_1.pdf
3. Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28). doi.org/10.1126/sciadv.adn5290
4. Hartmann, J., Exner, Y., & Domdey, S. (2025). The power of generative marketing: Can generative AI create superhuman visual marketing content? International Journal of Research in Marketing, 42(1), 13–31. doi.org/10.1016/j.ijresmar.2024.09.002
5. Exner, Y., Hartmann, J., Netzer, O., & Zhang, S. (2025, January 14). AI in disguise: How AI-generated ads' visual cues shape consumer perception and performance (manuscript). columbia.edu, AI in disguise (PDF)
6. Lee, H., Todri, V., Adamopoulos, P., & Ghose, A. (2025). The impact of visual generative AI on advertising effectiveness (SSRN working paper, posted November 3, 2025). doi.org/10.2139/ssrn.5638311
7. Matz, S. C., Teeny, J. D., Vaid, S. S., Peters, H., Harari, G. M., & Cerf, M. (2024). The potential of generative AI for personalized persuasion at scale. Scientific Reports, 14, 4692. doi.org/10.1038/s41598-024-53755-0
8. Maier, B. F., Aslak, U., Fiaschi, L., Rismal, N., Fletcher, K., Luhmann, C. C., Dow, R., Pappas, K., & Wiecki, T. V. (2025). LLMs reproduce human purchase intent via semantic similarity elicitation of Likert ratings. arXiv:2510.08338. arxiv.org/abs/2510.08338
9. Chapekis, A., & Lieb, A. (2025, July 22). Google users are less likely to click on links when an AI summary appears in the results. Pew Research Center. pewresearch.org, short read, 22 July 2025
10. Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2023, v3 2024). GEO: Generative engine optimization. arXiv:2311.09735. arxiv.org/abs/2311.09735
11. Cicek, M., Gursoy, D., & Lu, L. (2024). Study of the term "artificial intelligence" in product and service descriptions and purchase intentions. Journal of Hospitality Marketing & Management. doi.org/10.1080/19368623.2024.2368040, summary: Washington State University (2024, July 30), news.wsu.edu press release
12. U.S. Federal Trade Commission. (2024). Trade regulation rule on the use of consumer reviews and testimonials, 89 FR 68034 (published August 22, 2024, effective October 21, 2024). federalregister.gov, 89 FR 68034, press release (August 14, 2024): ftc.gov press release
13. Kirkland, C. (2025, June 2). Meta plans to automate all ad creation by 2026: Report. MediaPost (reporting the Wall Street Journal). mediapost.com/publications/article/406315