The short version
- In US consumer electronics questions, AI search drew 92.1% of its sources from earned media such as reviews and publications, an August 2025 study found (opens in a new tab).
- In a vendor dataset of brand prompts, only 2.9% of citations pointed to the tracked brand’s own site, while 75.2% went to other companies in the same space, Kumar reports (opens in a new tab).
- When asked whether a brand is legit, Perplexity cited the brand’s own site in 94.9% of answers and Google AI Mode in 17.7%, in our reputation study.
- Independent coverage was the strongest signal we measured for being recommended, in our brand entity study.
Do AI engines cite brand websites or third-party sources?
Mostly third-party sources, across the studies we reviewed. Chen and colleagues (opens in a new tab) compared Google with AI search on ranking-style product questions in August 2025. For US consumer electronics, Google’s results gave brand sites 32.9% of their sources. AI search drew 92.1% from earned media, with negligible social media.
The same tilt held for brands of every size. For well-known brands, ChatGPT drew 93.5% of its sources from earned media and 6.5% from brand sites. For niche brands, it drew 95.1% from earned media and 4.9% from brand sites.
Treat the exact figures with care. The study queried developer versions of the engines, not the consumer apps, and used an AI model to label sources. The authors also thank a GEO vendor for support.
Does the answer change with the type of question?
Yes. Brand sites gain ground when buyers are ready to act. Chen and colleagues found that on transactional questions, brand content rose in prominence across Google, ChatGPT and Perplexity.
Hotels show the same split. Zhu and Chang (opens in a new tab), who work at an AI company, audited 1,357 citations from Gemini for 156 Tokyo hotel questions in March 2026. Booking sites took 55.3% of all citations. Experiential questions drew 55.9% of citations from other sources, against 30.8% for transactional ones.
Hotel websites themselves were a small slice: 8.2% of citations for English questions and 11.0% for Japanese ones. The authors argue AI search does not prefer earned media as such. It cites whatever content type best answers the question. Whether that loosens reliance on booking sites and marketplaces is covered separately.
Banking is another case. In Chen’s bank questions, earned sources were 64.6% of citations and bank-owned sites 34.1%.
Does it depend on which AI engine is answering?
Yes, sharply. In our reputation study, we asked four engines whether 79 brands were legit. Perplexity cited the brand’s own site in 94.9% of answers. Google AI Mode did so in 17.7%.
Even when an engine cites your site, reviews shape the verdict. In the same study, 88.0% of answers cited a review or complaint platform. Claims attached only to review platforms were negative 56.5% of the time, against 6.4% for claims attached only to the brand’s own website.
Engines also differ for product questions. In the Uberti-Bona Marin audit, editorial and product-review sources made up 56.7% of the domains ChatGPT displayed, far more than in Google’s AI Overviews.
Which third-party sources carry the most weight?
Review publishers, review platforms and ranked “best of” lists. In Kumar’s vendor data, ranked list articles were 35.7% of the content pages engines cited. A single list can put a brand into many answers at once. Which kinds of sites lead depends on the question, as the types of websites AI engines rely on shows.
Some of those lists are written by the brands themselves. In our study of self-promoting lists, 24.2% of AI-cited numbered “best” lists with an identifiable publisher ranked that publisher first. Such lists were only 1.1% of all citations, so they are a minor route.
Assistants also go looking for reviews. In our hidden-searches study, ChatGPT looked for reviews in 46.2% of answers. When its search named a source such as NerdWallet or Avvo, the answer cited that source 44.0% of the time.
Does third-party coverage change whether you are recommended?
It is the strongest signal we have measured, though not proven to cause recommendations. In our brand entity study, each tenfold increase in the number of independent sites naming a brand in the cited pages went with 4.7 times the odds of being recommended.
Other researchers reach a similar view but without a direct test. Kumar and Palkhouski (opens in a new tab) argue that even high-quality pages may not be cited if they sit only on vendor blogs. They also note they did not vary where content was published, so the off-site effect remains untested in their work.
What should you do about it?
Treat third-party coverage as the main route into AI answers, and your own site as the source of record.
- Map which review sites, publications and lists the engines cite for your category.
- Earn coverage there through product reviews, analyst inclusion and press, not paid placements dressed as editorial. On paid reach more broadly, see whether advertising spend helps AI recommendations.
- Watch review platforms closely. They carry most of the negative claims engines repeat.
- Keep your own site clear and factual for transactional questions, where brand pages gain ground.
- Check Perplexity separately, since it cites brand sites far more often than other engines.
Our generative engine optimization service includes this kind of source audit.
What does the research not tell us yet?
No study we reviewed shows that adding third-party coverage causes a brand to be cited more.
- The evidence is observational: brands with more coverage may differ in other ways.
- “Earned”, “brand” and “editorial” are labeled differently across studies, often by AI models.
- Several large datasets come from vendors or companies with a stake in the answer.
- Most studies cover consumer products, hotels or software; many other industries are untested.
- We found no study measuring how much a brand-site citation is worth in traffic or sales.
Frequently asked questions
Does ChatGPT cite company websites?
Sometimes, but rarely as the main source. In one study, ChatGPT drew 6.5% of its sources from brand sites for well-known brands and 93.5% from earned media.
Why does AI cite my competitors’ pages instead of mine?
Engines often build “alternatives” answers from peer-brand pages. In one vendor dataset, 75.2% of citations went to other companies in the same space and 2.9% to the brand itself.
Do review sites like Trustpilot affect AI answers?
Yes, they shape the tone. In our reputation study, 88.0% of answers cited a review or complaint platform, and claims resting only on those platforms were negative 56.5% of the time.
Is PR more important than my website for AI search?
For discovery questions, third-party coverage matters more in most studies. Your own site gains weight on transactional questions and on Perplexity, which cited brand sites in 94.9% of reputation answers.
Sources
- Chen et al. (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.
- Uberti-Bona Marin et al. (2026), "If I Had to Buy Just ONE: Galaxy S26 Ultra": Auditing AI-Generated Product Recommendations (opens in a new tab), arXiv:2609.18729.
- Zhu and Chang (2026), The End of Rented Discovery: How AI Search Redistributes Power Between Hotels and Intermediaries (opens in a new tab), arXiv:2603.20062.
- Kumar and Palkhouski (2025), AI Answer Engine Citation Behavior An Empirical Analysis of the GEO16 Framework (opens in a new tab), arXiv:2509.10762.
- Underneath (2026), “Is this brand legit?” How AI assistants build a reputation
- Underneath (2026), Do Wikipedia and schema make AI assistants recommend a brand?
- Underneath (2026), How many “best of” lists cited by AI rank their own brand first?
- Underneath (2026), The hidden searches AI assistants run before they answer