What are the risks of native ads blended into AI chatbot answers?

What is a native ad inside an AI answer?

It is a paid mention written into the answer itself, so it reads like the assistant’s own advice. Erickson (opens in a new tab) contrasts two forms. A banner ad sits apart from the answer and is easy to recognize as an ad. A native ad is woven into the reply, and it may or may not be disclosed.

To show what that could look like, Erickson prompted ChatGPT to role-play a user with depressive symptoms. The prompted replies slipped in a soft drink, a brand-name antidepressant, and, in the worst case, a vodka brand as a “quick boost.” These were staged on purpose. The paper says plainly that it does not claim such responses are already happening organically.

Are ads really coming to AI answers?

Yes: ads have started appearing, and product questions are where they are most likely to show up. The audit by Uberti-Bona Marin and colleagues (opens in a new tab) reports that ChatGPT advertising expanded to 31 European markets in August 2026. Ads appear in between 10 and 14% of sessions for product recommendation queries. That figure comes from another study the audit cites, not from the audit’s own measurement. For the wider picture, see how ads are reaching AI assistants.

Product questions are not a niche use. The same audit notes that product and service recommendations already made up roughly 2% of ChatGPT conversations in a large 2025 measurement. And a survey cited by Wen and colleagues (opens in a new tab) found that 39% of 8,350 shoppers across 21 countries use AI for product discovery and related shopping tasks.

Google’s conversational search was still ad-free when Wang and colleagues (opens in a new tab) ran a field experiment with 1,100 US Google users in March 2026. Sending people to AI Mode cut the share who clicked any ad by 42.7 percentage points, simply because no ads were shown. The authors expect user satisfaction to shift once Google monetizes it.

Why are blended ads riskier than banner ads?

Because users cannot easily spot them, and AI answers already carry a tone of personal, confident advice. Erickson points to earlier research showing that people are more skeptical when they know someone is trying to persuade them. An ad that hides its purpose skips that skepticism. That is what makes it effective, and also what makes it deceptive.

The voice of the answer adds to the problem. The audit found clear differences in how three Google and OpenAI products frame advice:

These shares come from 117 real shopping questions asked from the Netherlands. A paid product that appears inside a “my pick” sentence would read as a personal endorsement.

Trust is also easy to inflate. In the experiment by Li and Aral (opens in a new tab), with 4,927 participants matched to the US adult population, people trusted AI search less than ordinary search on average. But adding reference links raised trust, even when the links were incorrect or invented. The look of credibility, in other words, can move people more than the substance.

Who is most exposed to the harm?

People in distress and people asking about health, money or other high-stakes choices. Erickson argues that native ads could be “especially corrosive” for users who are struggling and who trust a system with commercial motives. He calls this the “fake friend dilemma”: the user believes the assistant is on their side when it is serving someone else.

He singles out older adults, children and people in acute distress. He also argues for tobacco-style limits on where harmful or expert-only products, such as prescription drugs, can be advertised. These are policy arguments from a position paper, not findings from a study of real ads.

The reach is wide. In the poll cited by Wen and colleagues (opens in a new tab), 60% of 1,437 US adults said they use AI to find information at least some of the time.

What is the risk for the brand that buys the placement?

A brand that is caught hiding an ad in an answer can lose trust and invite backlash. Erickson cites earlier advertising research: when users recognize a disguised ad, trust drops and views of the brand can turn negative. Ads that cross a moral line, or personal targeting that feels like a privacy breach, can make the reaction worse. Our study of how AI assistants judge whether a brand is legit shows how that reputation is built from evidence.

Wen and colleagues (opens in a new tab) describe a second, quieter risk. If hidden promotion beats labeled ads, honest advertisers lose ground to covert ones, and the whole channel drifts toward hidden influence. They warn of “trust erosion” when such influence is later revealed.

There is also a risk to the platform, and so to the value of the placement. In the Google experiment by Wang and colleagues (opens in a new tab), being moved to AI Mode cut trust in information on Google by 0.34 points on a 7-point scale. It also drove an 11.2% increase in use of competing search engines. That test was about AI Mode, not ads, but it shows users do react when an experience feels worse.

A modeling study by Zhang and colleagues (opens in a new tab) points the same way. In simulations of 500 users over 20 periods, ad-heavy policies raised early revenue but shrank the user base and slowed paid sign-ups. This is a theoretical model, not observed behavior.

What should you do about it?

Treat any paid placement inside an AI answer as a reputational decision, not just a media buy. Practical steps:

1. Only buy AI placements that the platform labels clearly as sponsored, and ask the platform to show you exactly how the label will appear.
2. Avoid placements next to sensitive questions, such as mental health, medication or debt, unless you are confident the context is appropriate.
3. Write ad copy that could survive a screenshot: no claims you could not defend if a user, journalist or regulator saw it out of context.
4. Keep your earned visibility honest. Clear, verifiable facts on your own site and in independent coverage are the content AI answers can cite without paid help. That earned base is also the core of a long-term AI search strategy.
5. Check what AI assistants say about your category regularly, since the products they recommend change from run to run.

If you want help making your brand easier for AI assistants to describe accurately, see our generative engine optimization service.

What does the research not tell us yet?

The research has not yet measured how real users react to real ads inside AI answers. Specifically:

• The best-known paper on native ads in chatbots uses staged examples, not observed ads.
• The 10 to 14% ad frequency for ChatGPT is a figure one audit cites from another study; it is not independently confirmed here.
• No study we reviewed measures how often users notice a blended ad, or how a brand’s reputation changes after one is exposed.
• The trust experiments tested citations and AI labels, not ads; the link to advertising is an inference.
• The modeling study of ad-heavy versus ad-free answers is a simulation with assumed user behavior.

Frequently asked questions

Are there ads in ChatGPT answers now?

Yes, in some markets. An audit reports that OpenAI expanded ChatGPT advertising to 31 European markets in August 2026, most often on product recommendation questions.

Can people tell an ad from an AI recommendation?

Often they cannot, unless it is clearly labeled. Erickson cites earlier user research showing that people may not notice ads woven into AI search answers unless the ads are explicitly disclosed.

Is native advertising in AI answers legal?

The law is unsettled. Erickson notes that it is unclear how deceptive advertising rules apply to conversational search. Wen and colleagues call for extending FTC-style disclosure rules to AI answers.

Should my brand advertise inside AI answers?

Only where the placement is clearly labeled and the context is not sensitive. The research suggests disguised promotion carries a backlash risk that labeled ads avoid, though this has not yet been tested with real AI ads.

Sources

• Erickson, J. (2025), Fake Friends and Sponsored Ads: The Risks of Advertising in Conversational Search (opens in a new tab), arXiv:2506.06447.
• Uberti-Bona Marin, L. G., Bertaglia, T., Astante, G., Rijsbosch, B., van Dijck, G., Hannák, A., Spanakis, G. and Kollnig, K. (2026), "If I Had to Buy Just ONE: Galaxy S26 Ultra": Auditing AI-Generated Product Recommendations (opens in a new tab), arXiv:2609.18729.
• Li, H. and Aral, S. (2025), Human Trust in AI Search: A Large-Scale Experiment (opens in a new tab), arXiv:2504.06435.
• Wen, Y., Zhang, N., Yuan, H., Chen, X., Zhang, H. and Guo, H. (2026), Position: Generative Engine Optimization Creates Underexamined Risks, Governance Must Target Concentration, Disclosure, and Academic Blind Spots (opens in a new tab), arXiv:2606.12439.
• Wang, S. T., Gleason, J., Bart, Y., Wilson, C. and Metaxa, D. (2026), AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence (opens in a new tab), arXiv:2608.18352.
• Zhang, L., Jiao, C., Li, B. and Xiong, C. (2026), An Economic Framework for Generative Engines: Advertising or Subscription? (opens in a new tab), arXiv:2603.29071.