A test conducted by Hugo Sivignon on four AI assistants shows that ChatGPT, Perplexity, Gemini, and Claude rarely cite company websites when a client is looking for a service provider. In his column published on Journal du Net, he bases his findings on 24 responses obtained on October 29th, with web search enabled.
- ChatGPT, Perplexity, Gemini et Claude citent rarement les sites des entreprises.
- Les annuaires et plateformes comptent davantage que son propre site pour apparaître.
- Les informations obsolètes persistent, il faut corriger le site, les annuaires et les réseaux sociaux.
- Il faut vérifier le robots.txt, soumettre le site à Bing Webmaster Tools et interroger les clients sur leurs sources IA.
The protocol is simple. He asked five commercial queries, such as ‘freelance SEO consultant Lyon,’ ‘cheap showcase website for artisan,’ or ‘website creation Bourg-en-Bresse,’ followed by a question about his own name. Each question was asked only once, even though these tools never respond in exactly the same way. The author emphasizes: it’s ‘a snapshot, not a measurement.’ However, it allows for identifying some trends.
Directories and platforms rather than websites
For the commercial queries, his website appeared in only 2 out of 20 responses. The most frequent sources were freelance platform category pages, directories like Codeur, Sortlist, or the France Num activator directory, comparison articles published by agencies, and LinkedIn profiles.
For a small business, appearing in the responses therefore depends as much on what other sites say about the company as on its own website. A well-filled and up-to-date listing in a sector-specific directory can carry more weight than an additional page on one’s own site.
An outdated version and specific facts
To the question ‘Who is Hugo Sivignon?’, ChatGPT attributed a position he had left, taken from an old version of his LinkedIn profile, even though it had been updated two weeks earlier. The other assistants cited his website. If a company has changed its address, name, or activity, old information can therefore continue to circulate, and it is necessary to correct the website, the Google My Business listing, social media, and directories.
When the responses cited his website, they picked up concrete details: a starting price, a delivery time, a service area. According to him, generic statements provide nothing to latch onto.
The avenues discussed
The author first recommends asking your clients’ questions yourself to the main assistants and noting the sources cited. He then advises checking that the robots.txt file, which tells robots what they can crawl, does not block AI robots. He cites OpenAI’s documentation, which states that sites refusing its search robot, OAI-SearchBot, do not appear in ChatGPT’s search responses.
He also suggests submitting your site to Bing Webmaster Tools, Microsoft’s equivalent of Search Console, which is free and, according to him, takes about ten minutes. The Bing index feeds Copilot, and OpenAI indicates it partly relies on third-party search providers. Finally, he proposes adding ‘ChatGPT or another AI’ among the responses in the ‘How did you hear about us?’ form.
He considers, however, the llms.txt file, presented as a way to ‘talk to AIs,’ less of a priority: to date, none of the major players have announced using it to select their sources. The author plans to repeat the same test in six weeks.
