How to appear in ChatGPT

Appearing in ChatGPT is the outcome of the assistant naming a brand, describing it or linking one of its pages inside an answer. It happens two ways. In the first, ChatGPT replies from what the model already holds and downloads nothing: it repeats whatever the sources it trained on said about that brand. In the second it searches the web during the conversation, downloads pages and cites the ones it uses.

The slot inside the answer is not for sale. OpenAI does accept ads in ChatGPT and documents an agent of its own for reviewing the landing pages submitted to it, but that is an advertising slot rather than the answer, which is earned by being citable.

For a company selling to other companies, that appearance is already an acquisition channel with numbers behind it. Between 15% and 20% of new clients at Kulturo and HeyTalent arrive through GEO, according to what they state when they close, within a 25% to 30% that arrives through inbound overall. That is two companies over one year, not a market average.

The asymmetry between the first two agents is the mistake that turns up most often in an audit. OpenAI’s crawler documentation, read on 22 September 2026, states it plainly: “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links.” The same page explains that a site may allow OAI-SearchBot in order to appear in search results while disallowing GPTBot, because the two settings are independent.

Allowing an agent in robots.txt is still not enough if the firewall stops it first. OpenAI publishes the IP ranges each agent comes from in a file of its own — the OAI-SearchBot file sits at openai.com/searchbot.json, with a creation date of 2 January 2026 — and recommends allowing those requests alongside the user agent. This is the half that gets forgotten, because robots.txt and the firewall live in different places and usually belong to different people.

The widest published measurement of that overlap comes from Ahrefs, which analysed 15,000 prompts and checked how many of the URLs cited by AI assistants appeared in Google’s top 10 for the same query. The average was 12%, and more than 80% of the citations went to pages that did not rank at all for the target query. Louise Linehan wrote it up; it was published on 11 August 2025 and updated on 31 May 2026.

The split by assistant is what changes the tactics. In that same study Perplexity is the outlier, with 28.6% of its cited URLs inside Google’s top 10, close to one in three, while ChatGPT and Gemini are the two showing the least overlap. Translated into decisions: classic SEO transfers fairly well to Perplexity and poorly to ChatGPT.

None of that makes Google rankings wasted work. These are correlations: they describe what appears alongside what, not what causes what, and being indexed by Google is still the door through which a page becomes findable. What does follow is that a rank-tracking report is no measure of presence in ChatGPT.

On Similarweb data from January and February 2026, roughly 600,000 citation events in the United States, Wikipedia accounted for 13.15% of ChatGPT citations and Reddit for 11.97%: between them, more than a quarter of the total. 5W Research published it on 11 May 2026, and in that count neither The Wall Street Journal nor The New York Times nor Bloomberg made the top twenty domains.

That share is not stable, and the risk of building a plan on a domain someone else owns sits right there. Otterly, an AI visibility tracking tool, reported on 27 August 2026, revised on 14 September 2026, that ChatGPT citations to Reddit fell by at least 73% during August 2026, across 16 brand reports. It is a vendor measuring its own book of clients, so it counts as a volatility warning rather than a market figure.

Three decisions follow from that split, and none of them is posting on Reddit at volume:

• Write one sentence defining the company — what it does, who for and where — and use that exact string on its own site, on LinkedIn, in its industry directories and in the bio of anything it publishes elsewhere. When sources agree, an assistant states things; when they contradict each other, it hedges.
• Find the third-party pages already cited in its category and get the description they give of the brand corrected. Fixing a wrong detail is the request that gets accepted most often.
• Treat no external domain as an owned channel: the only thing that does not depend on a third party is the brand’s own site.

Both timings count from the start of the work with DoGeo: February 2026 at HeyTalent and January 2026 at Kulturo. Each client stated their date at closing, so these are attributions from a sales conversation rather than an analytics measurement.

What comes afterwards looks more like a flow than a spike. HeyTalent gets between 2 and 3 demos a week from people who arrived by asking ChatGPT, according to HeyTalent’s own figures. At Kulturo it is around 3 qualified form submissions a month from people who state they found the company by asking ChatGPT or Claude, according to Kulturo’s own figures.

Two projects are not a forecast, and the timing depends on things nobody controls: how much was already being written about the company, how crowded its category is, and whether its site was downloadable on day one. All those dates support is that the timescale is weeks and months rather than days.

The technical route is web analytics. A visit arriving from a link shown on chatgpt.com lands in the referral channel of Google Analytics 4, which Google defines as the channel by which users arrive at a site via non-ad links on other sites or apps. It does not arrive as an AI channel of its own: separating it out means building that channel grouping by hand.

The route that has produced usable data is asking. In the sales conversations at Kulturo and HeyTalent, leads name ChatGPT and Gemini above all, and Claude as well at Kulturo, according to what those leads say themselves. It is qualitative and gives no split by engine, but it is the only thing that has let a customer be attributed to a specific assistant: the question “how did you find us?”.

What does not work as a measure of ChatGPT is the generative AI features report in Google Search Console, however often the industry puts it on screen. It measures impressions inside Google’s own AI answers: a different engine, a different surface and a different index.