How to check whether ChatGPT, Perplexity or Gemini cite your website
Four ways to find out if AI answers name your site, from asking the tools the right questions to reading your server logs for their crawlers, and what to change if the answer is no.

The quickest check: open ChatGPT, Perplexity and Gemini, ask each the five questions a customer would ask before buying from you, and read whether your domain appears in the answer or its sources. Do it with web search turned on, in the language your customers use. Fifteen minutes gives you a first answer; the other three methods below tell you whether it is representative.
AI assistants with web access do not rank pages the way Google does. They retrieve a handful of sources per question, write one answer, and cite some of those sources, sometimes inline, sometimes in a list at the end. Either you are in that handful or you are not. There is no page two. That is why "are we cited" has replaced "where do we rank" as the question worth asking, and why it is worth checking properly rather than once.
1. Ask the tools yourself
Write down the questions, not the keywords. A customer does not type "fiduciary Zurich" into ChatGPT; they type "which fiduciary in Zurich is good for a small GmbH with 10 employees". Five of those, in the language your customers use, is a realistic sample.
Then, for each tool:
- ChatGPT: make sure search is on (the globe icon, or a model that browses). Ask. Expand the sources panel; inline citations are not the full list.
- Perplexity: searches by default. Sources are numbered above the answer; check each.
- Gemini: ask, then use "double-check" or look for the source chips under the answer.
- Claude: with web search enabled, sources appear as footnotes.
Record three things per answer: were you cited, who was, and what the answer actually said about the topic. The third one is the useful one. If every tool answers "it depends, look for someone with these qualifications" and names nobody, you are not losing to a competitor; you are losing to a vacuum, and a page that answers the question directly can fill it.
The weakness of this method is that one person's session on one day is one sample. Answers change between runs, between accounts and between weeks.
2. Read your server logs for their crawlers
Before an assistant can cite you, something has to fetch your pages. The major operators identify their crawlers by user agent, and the names are public: GPTBot and OAI-SearchBot (OpenAI), PerplexityBot, ClaudeBot and Claude-SearchBot (Anthropic), Google-Extended (Google's AI training opt-out token; Gemini's grounding uses the ordinary Googlebot index). If none of them appears in your logs, nothing you publish is reaching them, and the first fix is access, not content. Check robots.txt has not blocked them by accident; some CMS and CDN presets did in 2024.
If they are fetching, note pages. Crawlers spending their visits on your tag archive and not your service pages is a structure problem you can fix in a day.


