Does advertising spend help our brand get recommended by AI, or does something else matter more?

Does ad spend predict which brands AI assistants recommend?

Not on its own, in the most direct study so far. Malthouse and colleagues (opens in a new tab) at Northwestern University measured how often, and how high, six AI models recommended brands in five categories, from cordless drills to cruises. They then compared that with five measures of each brand’s presence in the market.

Across 82 brands, Google search interest was the strongest and most consistent signal, followed by online conversation. Advertising, news mentions and Wikipedia views contributed relatively little once all five were considered together. A stricter second check dropped advertising, news and Wikipedia entirely and kept only search interest and conversation.

How sure can we be about this?

Not very sure yet: it is an exploratory pattern in 82 brands, not proof of cause. The authors say plainly that their results should not be read causally. All five measures rise and fall together, and each one on its own was linked to more prominent recommendations, advertising included.

So the finding is narrower than “ads do not matter”. Ads are part of a brand’s general market visibility. They simply added little extra once search interest and conversation were known.

The measurement has limits too. Advertising was measured over the 12 months to May 2026, conversation volume was estimated by Brandwatch from a 5% random sample of mentions, and the AI models answered with web search switched off. Results with live web search may differ.

Why would search interest matter more than advertising?

The research does not say, and the authors warn against assuming searches cause recommendations. Both could reflect the same underlying thing: genuine consumer interest in a brand, which also shows up in what people write and discuss.

What other studies do show is that AI assistants lean on third-party sources. Chen and colleagues (opens in a new tab) at the University of Toronto sorted the sources behind AI answers into brand-owned, earned and social. For niche brands, 95.1% of ChatGPT’s sources were earned media, such as publications, reviews and expert sites.

Our brand entity study points the same way. Across ChatGPT, Gemini, Perplexity and Claude, the signal most closely linked to being recommended was independent coverage: how many other websites named the brand in the pages the assistants cited.

What do AI answers draw on instead of brand messaging?

Mostly other companies’ pages, review lists and coverage, rarely the brand’s own site. Kumar, a co-founder of the AI visibility company Ranqo, classified 149,912 citations from his company’s tracking data. Only 2.9% pointed to the tracked brand’s own domain, while 75.2% pointed to other companies in the same space. The wider evidence is in our guide on whether AI cites your own site.

Among other sources, YouTube was cited most, at 4.2% of citations. Ranked “best of” lists were the single most cited kind of content page, at 35.7% of content citations. This is a vendor’s data from its own customers, so treat the exact shares with caution.

The assistants also go looking for outside verdicts. In our hidden searches study, ChatGPT ran a mean of 3.7 web searches per buyer question. When one of those searches named a source such as NerdWallet or Avvo, the answer cited that source 44.0% of the time.

Can you simply pay to appear in AI answers?

Not inside the organic answer, under current published policy. Chu and Hou (opens in a new tab) note that in February 2026, OpenAI began testing ads in ChatGPT for US users on its Free and Go plans. The paid placements appear below the organic answer and are labeled, and OpenAI states that organic answers are generated independently. Why such labels matter to advertisers is covered in our guide to labeling ads in AI search.

Their experiments also suggest that classic sales language does little. In skincare tests on three AI models, evidence-style claims beat the leading brand 50% to 73% of the time, while urgency and scarcity copy did so only 10% to 13% of the time. Authority claims were worth the same as +0.17 rating points of real product improvement.

One warning applies. Those authority claims were deliberately invented, including fake clinical trials, to find the upper limit. The authors class made-up claims as potential false advertising and recommend only real, verifiable evidence.

What should you do about it?

Keep advertising for what it does well, and fund the things AI assistants actually read.

1. Do not expect ads to buy recommendations. On current evidence, spend shows little direct link to AI prominence once interest and conversation are counted.
2. Watch search interest and conversation. Track them next to AI visibility as early signals, while remembering the link is not proven to be causal.
3. Invest in earned coverage. Reviews, rankings, awards and “best of” lists from credible publishers are what assistants search for and cite.
4. Write evidence, not slogans. Real certifications, published results and clear specifications moved AI choices far more than urgency copy.
5. Test before reallocating budget. Change one input, then measure AI answers over several weeks and runs. It helps to settle first which team owns AI search visibility.

If you want help planning that mix, see how we approach generative engine optimization.

What does the research not tell us yet?

The research describes associations, and the key experiment has not been run.

• Nobody has varied ad spend and measured AI answers. The Northwestern study observed brands as they were.
• The sample is small and specific. It covers 82 brands in five US-oriented categories, with models answering from training only.
• The effect of ChatGPT ads is unknown. Testing began in 2026, and no study yet measures how paid placements interact with organic recommendations.
• The direction of cause is open. Whether rising search interest leads to more AI recommendations is a question the authors propose testing next.

Frequently asked questions

Does advertising help a brand show up in ChatGPT?

Not directly, on current evidence. In a study of 82 brands, ad spend added little to AI recommendation prominence once search interest and online conversation were taken into account.

What predicts whether AI assistants recommend a brand?

Search interest and online conversation in one study, and independent coverage in ours. Brands named by more independent websites had 4.7 times the odds of a recommendation for each tenfold increase.

Should we move budget from ads to PR for AI search?

Not wholesale, because the evidence is correlational. A safer step is to fund earned coverage alongside ads and test whether AI visibility moves.

Are there ads in ChatGPT answers?

OpenAI began testing labeled ads below organic answers for some US users in February 2026, and says organic answers are generated independently of them.

Sources

• Malthouse, Lee, Yang, Pal and Feng (2026), Evaluating Brand Retrieval and Ranking in Large Language Model Recommendations (opens in a new tab), arXiv:2609.16304.
• Chen, Wang, Chen and Koudas (2025), Generative Engine Optimization: How to Dominate AI Search (opens in a new tab), arXiv:2509.08919.
• Kumar (2026), Generative Engine Optimization at Scale: Measuring Brand Visibility Across AI Search Engines (opens in a new tab), arXiv:2606.20065.
• Chu and Hou (2026), Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems (opens in a new tab), arXiv:2606.17443.
• Underneath (2026), Do Wikipedia and schema make AI assistants recommend a brand?
• Underneath (2026), The hidden searches AI assistants run before they answer