Many studies now count which websites ChatGPT, Perplexity and Gemini link to in their answers. The results are useful, but they do not agree with each other. Here is what two of them found.
A study of 366,087 citations
A study of 366,087 citations, published by Everything PR, found that each assistant leans on a different kind of source:
- OpenAI’s models concentrated on primary news sources such as wire services. In the same study, Reuters accounted for about 22.8% of ChatGPT’s citations.
- Perplexity favoured the BBC.
- Gemini surfaced Forbes consistently across consumer and business questions.
A study of “who to hire” questions
UX Continuum studied what the assistants cite when people ask whom to hire. It found large differences in how many sources each assistant uses:
| Assistant | Sources per answer |
|---|---|
| Perplexity | 20.0 |
| Gemini | 10.0 |
| ChatGPT | 7.4 |
In the same study, Gemini pointed to the contractor’s own website in 80.7% of its citations. The author describes this as a small sample, so it should not be treated as a general rule.
Why the studies disagree
Reports of this kind often give different answers because they differ in:
- The questions asked. News questions, product questions and local-service questions pull in different sources.
- The date. Assistants change how they search, and a study from last year may not describe today.
- The counting method. Counting every link, or only the first link in each answer, gives different results.
What to take from this
- Do not aim for one assistant. Each one looks in different places, so clear, accurate pages that others also mention are the safer approach.
- Check your own topics. Ask the assistants the questions your customers ask, and note which sources appear. Repeat monthly.
- Keep your own pages complete. If an assistant cites a company’s own site, as Gemini did in the hiring study, an up-to-date, detailed page on your site matters.
For how the three differ in practice, see why ChatGPT, Perplexity and Google cite different sources.
