Does ChatGPT mention your company?

Yes, ChatGPT can mention your company, but only if the model finds you at the moment it searches. It does not answer from memory when the question is about a supplier here and now: it runs a search, pulls a handful of pages and summarises them. The work of becoming one of those pages is called GEO, generative engine optimization. Here is what drives the selection, seven actions you can take yourself, how to measure the result and how much traffic it actually delivers.

What is GEO, generative engine optimization?

GEO is optimization for answers rather than for lists. Classic SEO competes for a position among ten links. GEO competes to become one of the three to five sources the model chooses to summarise and link to in its answer. The same thing is sometimes called AEO, answer engine optimization, or simply AI optimization. In practice it is the same work under three names.

The difference from SEO is not the technique but the space. In an AI answer there is no position four to slip down to: either you are among the sources or you do not exist in that answer. That makes the selection criteria more important than the finer points of ranking.

How much traffic does AI search actually deliver?

We have the numbers from a site we launched at the end of June 2026, a stone supplier in Stockholm with no ad budget. The first table shows sessions according to Google Analytics from 1 May to 18 August 2026, covering the period around and after the launch. The full background is in the client case Stonepoint.

That was the picture when this article was first published. ChatGPT was already the third traffic source, ahead of Bing and ahead of all social channels combined, but eight sessions in 110 days is not a channel. So we measured the same site again. The next table covers 19 August to 16 September 2026, 29 days. A contact attempt means a submitted form or a click on the phone number or WhatsApp.

ChatGPT went from 8 sessions in 110 days to 50 in 29. The visits came evenly, one to four on most days with no spikes, almost all from different users, and they landed mainly on the home page and the pricing page. Google is still three times larger in traffic. But of the month's contact attempts, four came from ChatGPT visitors and one from Google visitors. We cannot prove the cause, but the site has prices, dimensions and guides in plain text, which is exactly what a model can read and quote.

How the models pick which companies to name

ChatGPT searches through Bing's index and its own crawler, Perplexity has its own index, Google's AI Overviews are built on Google's ordinary index. Different pipes, same logic: pull a few pages that appear to answer the question, summarise them, cite the sources. That gives five concrete conditions.

• You have to exist in the underlying index. If you do not rank at all in Google or Bing you are not even a candidate to be cited.
• The page has to be machine readable. Text in HTML works, text baked into an image or rendered only after heavy JavaScript often does not.
• The answer has to sit near the surface. The model picks the paragraph that answers the question, not the whole page. An answer that arrives only after three promotional intros rarely gets picked.
• Mentions on other sites carry weight. Directories, comparison sites and forums get cited ahead of your own sales copy, because a third-party source looks more independent.
• The entity has to be unambiguous. The same company name, address and registration number everywhere, otherwise the model cannot tell that the mentions refer to the same business.

Seven actions that make you citable

1. Answer the question in the first paragraph. Heading as a question, answer directly below, depth after that. The format that wins featured snippets in Google wins space in AI answers.
2. Write out the numbers. Prices, timeframes, dimensions and terms in running text, not in an image and not only in a quote document. The model can only cite what it can read.
3. Add an FAQ block with real customer questions and mark it up with schema.org. The question format matches how people actually ask a chat assistant.
4. Make sure others mention you. Industry directories, comparison sites, local registers and press. This is the heaviest factor and the one that takes longest.
5. Keep the entity consistent. Identical company name, address and registration number on the website, in the Google Business Profile and in every directory.
6. Let the AI bots in, both in robots.txt and in the firewall. If you block the search bots they cannot fetch you, and then they cannot cite you either.
7. Keep the site fast and indexable. The fundamentals still decide, and they are collected in our technical SEO guide.

Robots.txt: should you let the AI bots in?

The bots come in two kinds, and the difference decides the question. Search bots (OAI-SearchBot, Claude-SearchBot, PerplexityBot) decide whether you can become a source in an answer: block them and you disappear from the answers. Training bots (GPTBot, ClaudeBot) decide whether future models know about your company without searching. Block them and you still appear in search-based answers, but the model knows less about you when it answers from memory. For a service company that wants to be found the choice is obvious: let both in, it is free exposure. For a site whose business rests on ad impressions it is reasonable to close the training bots and keep the search bots open.

The check has two steps. First open yoursite.com/robots.txt and look for Disallow rules for the bots above. Then look at the firewall, because robots.txt is only half the answer: Cloudflare and other CDN services have settings that block AI bots regardless of what robots.txt says, and web hosts can have similar blocks. We found such a block against GPTBot and ClaudeBot on synligwebb.se ourselves in September 2026, despite a fully open robots.txt.

How to measure whether you appear in AI answers

1. Analytics: filter on the sources chatgpt.com, perplexity.ai and copilot.com. Google Analytics now classifies these under the ai-assistant channel, so you can follow them together.
2. Search Console will not show it. A click from ChatGPT is a referral, not a Google search, and never appears in the search analytics. Looking there makes it seem as if the channel does not exist.
3. Manual prompt testing: ask ten questions a customer would ask, note which companies are named and which sources are cited. Repeat every month with the same questions. Blunt, but it is the only direct measurement of visibility inside the answers themselves.
4. Server logs or your CDN statistics: check whether OAI-SearchBot, Claude-SearchBot and PerplexityBot actually fetch your pages. If the search bots fetch nothing, the content does not matter.

GEO does not replace SEO, and the proportions show why

147 against 50. Even in the strongest AI month we have measured, three times as many visits came from Google, and on the other sites we follow the gap is far wider. Rebuilding your whole strategy around AI search would therefore mean betting on the smaller channel before the larger one is solved. Dismissing it no longer works either: in that same month more contact attempts came from ChatGPT than from Google.

The good news is that the work is largely the same work. Indexability, readable text, clear answers, mentions and an unambiguous entity are exactly what makes you findable in Google. GEO is therefore not a separate discipline but one more reason to do the fundamentals properly. The ongoing work sits in search engine optimization, and how long it takes before anything shows up at all is something we measured on a real site in How long does SEO take?.