Guide
What it takes for a page to be found, read and cited by ChatGPT, and by Google’s AI Overviews and AI Mode, taken from what OpenAI and Google publish about their own systems and from nothing else. Two lines in robots.txt, one check at the firewall, and the same page Google has always asked for.
OpenAI’s own figure for a robots.txt change to reach ChatGPT search. Everything else on this page is a decision you can make today.
Two different questions share the phrase. One is whether ChatGPT will find your page, read it and cite it when somebody asks a question your business answers. The other is whether ChatGPT can do your SEO for you. This page is about the first, because it is the one with a documented answer. The second gets a section at the end.
Everything below comes from four pages the vendors maintain themselves: OpenAI’s overview of its crawlers, OpenAI’s help article on searching the web with ChatGPT, Google’s page on AI features and your website (updated December 2025) and Google’s overview of its crawlers and fetchers (updated June 2026). Where they are silent, this page is silent. Neither vendor publishes a ranking factor list for its AI answers, and anybody who sells you one is guessing.
OpenAI’s help article says ChatGPT may search the web automatically when a question would benefit from current information, and that a person can also start a search by choosing Search from the tools menu. The part that matters for a website is what happens to the question next. When ChatGPT search works with another search provider, and OpenAI says it sometimes does, it typically rewrites the question into one or more targeted queries and sends those. OpenAI’s own example turns a researcher’s long question about drugs that target CCR8 into the query “CCR8 immunotherapy drug development 2025”, reads the results, and then sends more specific follow-ups. The providers the article names, by way of their privacy policies, are Microsoft and Shopify.
Location goes into the rewrite too. The article says ChatGPT takes a general location from the IP address, so “what are some good restaurants near me” from San Francisco becomes “top restaurants San Francisco”, and a saved memory that the person is vegan makes it “good vegan restaurants San Francisco”. So a page is matched against short, specific queries in plain nouns, with a place and often a year, and never against the sentence the person typed. A page that says what it is about and where, in the words people use, is what a rewritten query finds. A page that speaks in slogans is not.
The answer may carry citations, and the reader can open each one. OpenAI’s article warns its own users that search results and citations can be incomplete, outdated or incorrect, and tells them to open the cited source, check when it was published or updated, and prefer an authoritative source when accuracy matters. That is the reader you are writing for: somebody who arrives to check, with the answer already in hand.
On ranking, the article says this much and no more: ChatGPT ranks search results using multiple factors intended to help users find relevant, reliable information, placement is not guaranteed, and to make a website eligible for inclusion you allow OAI-SearchBot to crawl it and confirm that the host or content delivery network allows traffic from OpenAI’s published searchbot IP addresses. Every other claim about ChatGPT’s ranking that you will read anywhere is somebody’s inference.
One more thing in the article matters to a restaurant. Reservation availability shown in ChatGPT comes, depending on the restaurant, from OpenTable, Resy or Yelp. A restaurant that takes bookings through none of them can still be found and cited from its website, but the booking part of the answer is drawn from those partners.
OpenAI’s crawler page describes the user agents in the figure above and says each robots.txt setting is independent of the others. Its own example is a site that allows OAI-SearchBot in order to appear in search results while disallowing GPTBot to say its content should not be used in training. Where both are allowed, OpenAI may use one crawl for both purposes rather than fetching the site twice. For search, it says a robots.txt change can take about 24 hours to take effect.
OAI-SearchBot is the one that surfaces websites in ChatGPT’s search results. A site that opts out of it will not be shown in ChatGPT search answers, though OpenAI says it can still appear as a navigational link. GPTBot crawls content that may be used to train OpenAI’s models, and disallowing it is how a site says no to that. ChatGPT-User is different in kind: it visits a page when a person asks ChatGPT or a custom GPT something, it is not used to crawl the web automatically, robots.txt rules may not apply because a person triggered the visit, and OpenAI says it is not used to determine whether content appears in search. OAI-AdsBot visits only pages submitted as advertisements.
So the decision for a business that wants to be cited and does not want to train a model is two groups:
# ChatGPT search: allowed, so the site can be cited
User-agent: OAI-SearchBot
Allow: /
# Model training: your decision, and independent of the group above
User-agent: GPTBot
Disallow: /ChatGPT-User gets no group, because OpenAI says robots.txt rules may not apply to a fetch a person triggered and that blocking it changes nothing about search inclusion. Our robots.txt generator writes these groups with a named choice for each AI crawler, and checks the file it produces before you copy it.
Two details from the same page help you read your own server log. The OAI-SearchBot and GPTBot user-agent strings carry a version number that may change, so match on the name rather than the whole string. And when either fetches robots.txt it may add a robots.txt marker to its user-agent string, so that a log without paths still shows which requests were for that file.
OpenAI’s eligibility statement has two halves and most sites only do the first. The second is to confirm that the website host or content delivery network allows traffic from the published searchbot IP addresses, which OpenAI lists as JSON files on its own domain: searchbot.json for OAI-SearchBot, gptbot.json for GPTBot and chatgpt-user.json for ChatGPT-User. A bot-protection rule at the CDN that challenges every automated visitor challenges OAI-SearchBot too, and what it then reads is the challenge page rather than yours. Nothing in robots.txt can override a firewall.
Google’s crawler overview adds a rule that catches sites which never meant to block anyone. Google says its crawlers leave primarily from IP addresses in the United States and only attempt other countries when they detect a site blocking requests from there, so a country rule written to stop spam from abroad can stop Googlebot. The same page says Google’s crawlers identify themselves three ways, by the user-agent header, the source IP address and the reverse DNS hostname, and that is how to check a blocked request was really Google before you let it through.
Google’s page on AI features says there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisation. Its eligibility rule is one sentence: to be shown as a supporting link, a page must be indexed and eligible to be shown in Google Search with a snippet. It adds the caution it attaches to everything, that meeting every requirement does not guarantee crawling, indexing or serving.
The useful part is how the answer is built. Google says both features may use what it calls query fan-out, issuing multiple related searches across subtopics and data sources to develop a response, and that while the response is being generated its models identify more supporting pages, so the set of links shown is wider and more varied than a classic result page. It also says AI Overviews are shown only when its systems judge them additive to classic search, and so often do not trigger. The rewriting OpenAI describes and the fan-out Google describes point the same way: a page is matched on the specific question it answers, under a heading that names it.
Google’s practices for AI features are the ones it has published for search all along, and it lists them: crawling allowed in robots.txt and by any CDN or hosting infrastructure, content findable through internal links, a good page experience, important content available as text, images and video that support the text, structured data that matches the visible text, and Merchant Center and Business Profile information kept up to date. Then the sentence that saves a lot of money: you do not need to create new machine-readable files, AI text files or markup to appear in these features, and there is no special structured data to add.
The controls are the ordinary ones. Google says the robots.txt rules for Googlebot are the control for how a site is crawled for search, and that to limit what is shown from a page you use nosnippet, data-nosnippet, max-snippet or noindex; Google-Extended is a separate control for training and grounding in some of Google’s other systems. Traffic from AI Overviews and AI Mode is counted in Search Console’s Performance report under the Web search type rather than shown apart, and Google says clicks from result pages that carry an AI Overview tend to be higher quality, meaning people spend longer on the site.
Three plain facts from Google’s crawler overview round this off. Its crawlers read the first 15 MB of a file by default and ignore the rest, so a fact buried at the end of a huge page or PDF may never be read. They support ETag and Last-Modified for caching and Google recommends ETag, which is how a corrected page gets its correction fetched cheaply. And crawling over HTTP/2 saves resources on both sides but brings no ranking benefit, in Google’s own words.
Ten checks, each with the vendor page it comes from. The first three take an afternoon and are where most sites that “do not show up in ChatGPT” actually fail.
robots.txt allows OAI-SearchBot, and says what you want about GPTBot
The two lines are independent. Allowing the first is what makes the site eligible for ChatGPT search; disallowing the second says the content may not train OpenAI models, and it changes nothing about search.
The host and the CDN let OpenAI’s published IP ranges through
OpenAI’s eligibility statement has two halves, and the second is the firewall. A bot rule that challenges every automated visitor challenges OAI-SearchBot too, and a challenge page is not your page.
Traffic from the United States is not blocked
Google says its crawlers leave primarily from addresses in the United States, and only try other countries when it detects a site blocking them. A country rule written for spam takes Googlebot with it.
The page is indexed by Google and eligible for a snippet
That is the whole of Google’s eligibility rule for a supporting link in AI Overviews and AI Mode. Check it in Search Console’s URL inspection before anything else.
The important content is text, and the page says what it is about in plain nouns
Google lists textual content among its practices for AI features. OpenAI says ChatGPT rewrites the person’s question into targeted queries before searching, so the page is matched against specific words: the subject, the place, the year.
Other pages on the site link to it
Internal links are on Google’s list. A page reachable only from the sitemap is crawled and then sits at the bottom of the site with nothing pointing at it.
The structured data says what the visible text says
Google asks that structured data match the text on the page. A schema that claims what the page does not show is the one thing on this list that can make things worse.
The Business Profile and any Merchant Center listing are current
Both are on Google’s list for AI features, and for a local business the profile is the thing the answer draws hours and reviews from.
The page shows when it was published and when it was last updated
OpenAI tells its own users to open a cited source and check when it was published or updated before relying on it. A page with no date fails that check in the reader’s hand.
The server sends ETag and Last-Modified
Google’s crawlers use both to avoid fetching an unchanged page again, and Google recommends ETag. Cheap recrawls are how a corrected page gets its correction read.
It does not name a ranking factor for ChatGPT, because OpenAI does not, beyond “multiple factors” and a statement that placement is not guaranteed. It does not offer a file to submit, because neither vendor describes one for its AI answers. It quotes no study of how often ChatGPT cites anybody, because a study is somebody else’s sample and this page promised you the vendors’ words. And it cannot tell you that the checks above will make ChatGPT cite you, because OpenAI itself warns that its citations can be incomplete, outdated or incorrect. What the checks do is remove every documented reason for a page to be ineligible or unreadable, which is the whole of what a site owner controls.
The same search that brought you here is also made by people who want ChatGPT to do the SEO, so here is how it is used in this studio. It drafts well: an outline for a page, ten title variants to measure against a 60 character budget, a keyword list grouped by what the searcher wants, a first draft of a FAQ answer, a LocalBusiness schema to validate. It does not know your search volumes, it cannot read your Search Console, and it will state a figure with the same confidence whether it is right or invented.
So the rule is that nothing it wrote goes on a page unchecked. Every figure is traced to a source you can open, every title is measured in the SERP preview rather than counted, every schema goes through a validator, and the page is read aloud once before it ships, because the one thing a drafted page reliably carries is a cadence a reader has learned to skip. The FAQ schema generator and the LocalBusiness schema generator on this site build the markup from what the page actually says, which is the check Google’s list asks for.
Getting the visibility is the half everybody sells. Keeping it is the half that decides whether businesses still have customers arriving from Google in three years, and almost nobody writes it down, so here it is.
Not usually an algorithm update, whatever the industry press implies. Core updates move positions; these remove you.
By a distance the most common cause of a real collapse. URLs change and nothing redirects, or the new site ships with the staging robots.txt still in it. Both are entirely preventable and both are usually discovered weeks later.
One checkbox marked "discourage search engines", one SEO plugin defaulting to noindex on a template, one caching layer serving a stale page. Small change, total effect.
The Business Profile owned by a former agency, Search Console verified to somebody who has left, the domain registered to a developer. None of this hurts until the relationship ends, and then all of it does at once.
For a business that serves an area this is worse than any algorithm update, because it removes you from the map entirely and on the same day.
Nothing dramatic happens. Reviews stop, hours drift, a competitor keeps going, and a year later the position has quietly gone. This is the most common ending and the least noticed.
Every one of these still gets sold, usually by email, usually cheaply, and usually to somebody who does not know it is being done in their name. All of them work for a while.
Guest post networks, paid directories that exist only to sell links, anything offered by email with a price per link. These are the clearest thing Google acts on manually, the action can arrive years later, and recovery means removing what you paid for and then waiting.
Bought, incentivised, written by staff, or traded with another business. Google removes them, and where the pattern is obvious it suspends the listing rather than the reviews. It is also the one that damages you with customers if it is ever visible.
It moves the map pack for a while, which is exactly why it keeps being sold. It is also the single most common cause of a suspension, and the suspension takes the whole listing rather than the extra words.
A page per town you have never worked in, a page per service you do not offer, content generated to fill a content calendar. It does not rank and it drags down what would have.
Nobody controls Google’s results. A guarantee means either the terms are meaningless or the method is one of the four above.
Twenty minutes a month, and it is the difference between noticing a problem in week one and noticing it in month six.
The domain, the hosting, the Google Business Profile, Search Console and Analytics should all be registered in the business’s name with the business’s own email, and anybody working on them should have access granted rather than ownership transferred. It costs nothing to set up that way and it is close to impossible to unwind afterwards. If any of those are currently in somebody else’s name, that is worth fixing this week whether or not anything is wrong.
The ten checks above run on your site and reported from your server log and your Search Console rather than from a dashboard, then fixed, with the robots.txt and the CDN rules written for you.
No. OpenAI states that each robots.txt setting is independent: a site can allow OAI-SearchBot, so it can be shown in ChatGPT search answers, while disallowing GPTBot, so its content is not used to train models. OAI-SearchBot alone decides the search half. OpenAI also says a robots.txt change takes about 24 hours to reach its search systems.
Google says you do not need to create new machine-readable files, AI text files or markup to appear in AI Overviews or AI Mode, and that there is no special structured data to add. OpenAI’s eligibility statement for ChatGPT search names two things only: allow OAI-SearchBot in robots.txt, and let its published IP addresses through at the host or CDN. Neither vendor names any other file.
OpenAI publishes no ranking factor list; it says results are ranked on multiple factors and that placement is not guaranteed. What it does publish is the eligibility rule and the query rewriting, so the checks are: is OAI-SearchBot allowed and getting through, does the page name its subject and place in the plain words a rewritten query would use, and is there a visible date on it. The competitor’s page usually passes one of those and yours does not.
A person asked ChatGPT something and it fetched your page to answer them. OpenAI says ChatGPT-User is not used to crawl the web automatically, that robots.txt rules may not apply because a person triggered the visit, and that it is not used to decide whether a site appears in search. It is the closest thing to a citation you can see in your own logs.
Not as its own report. Google says clicks and impressions from AI Overviews and AI Mode are included in the Performance report under the Web search type, together with the rest of search. Google also says clicks from result pages that carry an AI Overview tend to be higher quality, meaning people stay longer.
OpenAI’s crawler documentation describes OAI-AdsBot, which checks pages submitted as ads on ChatGPT, so advertising in ChatGPT exists and is labelled as such. An advertisement is not a citation, and OpenAI says the data OAI-AdsBot collects is not used for training.
A 30-minute call in your working hours. Tell us what you sell and where, and we will tell you what we would do first and what it involves. Prefer email? Send your website through the form and the Passieon Team will reply within two working days.
Message us
Ask anything about your site or where your enquiries are going. It goes straight to a person, not a bot. Leave an email and you get a reply.