Coaley Peak GEO Research Centre Engine guide
ChatGPT How the world's most-used AI assistant searches, cites and learns, and how to be its source.
ChatGPT is the assistant most people mean when they say AI: OpenAI says it reaches 1.2 billion people a week. It answers from OpenAI's GPT models and, when it searches, from OpenAI's own crawl and third-party search providers, with citations. This guide covers its timeline, the four layers behind its answers, its four crawlers and how to opt out of each.
Independent guide: Coaley Peak is not associated with OpenAI. Trademarks
- Owner
- OpenAI
- Launched
- 30 November 2022; ChatGPT search 31 October 2024
- Home market
- Worldwide: 208 countries and territories, not mainland China or Russia
- Languages
- English plus 59 interface languages
- Open weights
- No for ChatGPT's models; OpenAI's separate gpt-oss models are open
- Crawlers
- OAI-SearchBot, GPTBot, ChatGPT-User and OAI-AdsBot
Overview
What ChatGPT is, and who uses it
OpenAI was founded as a non-profit research lab in December 2015 and launched ChatGPT on 30 November 2022, as a chat interface to a GPT-3.5 model trained with reinforcement learning from human feedback. GPT-4 followed in March 2023, GPT-4o in May 2024 and GPT-5 in August 2025. Since October 2025 OpenAI has been a for-profit company controlled by the non-profit OpenAI Foundation.
ChatGPT became a search engine in stages. OpenAI tested a prototype, SearchGPT, in July 2024 and launched ChatGPT search that October; it reached all logged-in users in December 2024 and logged-out users in February 2025. ChatGPT now decides for itself when to search, and shows inline citations and a Sources panel. Shopping results, restaurant bookings and, since February 2026, ads below answers on the Free and Go plans have followed.
Its reach is unmatched. OpenAI says ChatGPT reaches 1.2 billion people each week, and it is offered in 208 countries and territories, though not mainland China, Hong Kong, Russia or Iran. Its interface comes in English and 59 other languages.
OpenAI is also one of the more transparent providers for site owners. It documents separate crawlers for search, training, user actions and ads, publishes an IP list for each, and tags referral traffic with utm_source=chatgpt.com. A site can refuse training and still be eligible for ChatGPT search.
Timeline
ChatGPT, from launch to today
OpenAI is announced as a non-profit AI research company.
OpenAI launches ChatGPT, fine-tuned from a GPT-3.5-series model and trained with reinforcement learning from human feedback.
OpenAI releases GPT-4, first offered through ChatGPT Plus.
OpenAI documents GPTBot and how to block it in robots.txt, letting sites opt out of training.
The New York Times sues OpenAI and Microsoft for copyright infringement.
OpenAI announces GPT-4o, reasoning across audio, vision and text.
OpenAI tests the SearchGPT prototype and introduces OAI-SearchBot, kept separate from training opt-outs.
ChatGPT search launches, with links to sources, content from news and data partners, and third-party search providers.
Italy's data protection authority announces a €15 million fine against OpenAI. In a judgment published in March 2026, the Court of Rome upheld OpenAI's challenge.
GPT-5 launches and becomes ChatGPT's default model for logged-in users.
OpenAI launches Instant Checkout and the Agentic Commerce Protocol, built with Stripe, for buying inside ChatGPT.
OpenAI completes its recapitalisation as a for-profit company controlled by the non-profit OpenAI Foundation.
Ads begin testing in ChatGPT in the US, for Free and Go users.
GPT-5.6 Sol begins rolling out as ChatGPT's flagship reasoning model for paid plans.
OpenAI introduces GPT-6 Astra, initially to a limited set of organisations.
OpenAI releases GPT-6 Sol and GPT-6 Luna, the smaller and lower-cost tiers of the GPT-6 generation. Read our news story
GPT-6 with Intelligent UI starts rolling out in ChatGPT: GPT-6 Sol on paid plans, and GPT-6 Luna on Free and Go from 8 October. Answers can mix text, visuals and interactive elements. Read our news story
Things to know
Things to know about ChatGPT
To be eligible for ChatGPT search, allow OAI-SearchBot in robots.txt and make sure your host or CDN allows OpenAI's published search IP ranges. OpenAI says placement is never guaranteed.
Disallowing GPTBot opts out of training only. It does not remove a site from ChatGPT search; OAI-SearchBot controls that.
ChatGPT-User, which visits pages for user actions, is not used to decide search eligibility, and OpenAI says robots.txt may not apply to it.
Allow about 24 hours after a robots.txt change before checking how ChatGPT search behaves.
Blocking OAI-SearchBot does not guarantee invisibility: a URL ChatGPT learns of from a third-party search provider can still appear as a bare link and title unless the page carries noindex.
Referral traffic from ChatGPT carries utm_source=chatgpt.com, so it can be separated in your analytics.
From 8 October 2026 Free and Go users move to GPT-6 Luna, whose published knowledge cut-off (for its API version) is 18 May 2026; newer facts reach them mainly through search.
OpenAI's help page on search points users to Microsoft's and Shopify's privacy policies for its third-party search providers.
Product carousels are organic, not ads. OpenAI says partnerships and Instant Checkout do not influence product ranking; Shopify merchants are included through Shopify Catalog, and other merchants can apply to send product feeds.
Ads on the Free and Go plans sit below answers, are labelled as sponsored and, OpenAI says, do not affect answers. Self-serve ads are available in the UK.
The four layers
How ChatGPT builds an answer
Every AI answer is made in four layers: what the model learned in pre-training, how it was tuned, what it retrieves live, and how it puts the answer together. Here is each one for ChatGPT, and what it means for a brand that wants to be cited.
Layer 1
Pre-training: the base (foundation) model
ChatGPT runs on OpenAI's closed GPT models. From 7 October 2026 GPT-6 Sol powers Chat on the paid plans and GPT-6 Luna reaches Free and Go from 8 October, replacing GPT-5.6 as the default; GPT-6 Astra is offered as 'GPT-6 Pro' on Pro, Business and Enterprise, and GPT-5.6 Sol stays selectable. OpenAI's published knowledge cut-offs are 20 April 2026 for GPT-6 Sol and 18 May 2026 for GPT-6 Luna (both for their API versions), 30 April 2026 for GPT-6 Astra and 16 February 2026 for GPT-5.6. Its training data, OpenAI says, is freely accessible public web content (not known paywalled sources), data from partners, information from users and trainers, and increasingly synthetic data.
What it means for you: Anything about your brand published after those cut-offs reaches ChatGPT only through search, and what GPTBot could crawl before them shapes what ChatGPT simply knows. Allowing GPTBot is a choice about whether you want to be part of that knowledge.
Layer 2
Post-training: tuning and alignment
ChatGPT was originally trained with reinforcement learning from human feedback. OpenAI publishes a Model Spec, a living document of how its models should behave, last updated in August 2026; a 2025 update moved refusals towards 'safe completions'. GPT-5.6 adds trained-in safeguards and real-time monitoring for high-risk biology and cybersecurity requests. Consumer conversations may be used for training unless users switch off 'Improve the model for everyone'.
What it means for you: The Model Spec favours balanced, sourced answers over promotion, so ChatGPT tends to present several options and cite them. Brands are named when trustworthy sources name them, not because their own pages assert it.
Layer 3
Retrieval: live search
ChatGPT decides when to search, or the user chooses Search. It rewrites the prompt into targeted queries and draws on OpenAI's own crawl by OAI-SearchBot, third-party search providers, and content from partners such as news and data providers. Shopping uses structured product data, Shopify Catalog and merchant feeds. ChatGPT-User fetches individual pages live when a request needs them. OpenAI does not publish the size or make-up of its own index.
What it means for you: This is where ChatGPT visibility is won. Allow OAI-SearchBot, keep key facts in crawlable HTML, rank well in the major search engines that feed third-party results, and supply clean feeds for products. OpenAI says placement depends on several relevance and reliability factors.
Layer 4
The answer: what people read
Search answers carry inline citations, previewed on hover on desktop, and a Sources panel listing cited and related links; images link to their source, and mobile answers may include a map. Since October 2026 GPT-6 can build answers from text, visuals and interactive elements such as maps, comparisons and charts (Intelligent UI), and can start answering before it finishes thinking; OpenAI has not said how citations sit inside those elements. Shopping questions produce product carousels with merchant lists and review summaries, some with Instant Checkout. Ads on Free and Go sit below answers, labelled as sponsored. Outbound links carry utm_source=chatgpt.com.
What it means for you: Citations are visible and measurable. Track utm_source=chatgpt.com visits and the questions behind them, and write pages whose key facts can be quoted cleanly, so ChatGPT has a reason to put your link beside its answer. Prices, specifications and locations stated plainly in HTML are also easier to place in a comparison or on a map.
Crawlers
ChatGPT's crawlers and user agents
The user-agent tokens to look for in your server logs and to name in robots.txt.
| User-agent token | What it does | robots.txt |
|---|---|---|
| OAI-SearchBot Documentation | Surfaces websites in ChatGPT's search features. Sites that disallow it are not shown in search answers, though they may still appear as navigational links. Its user agent ends 'compatible; OAI-SearchBot/1.4; +https://openai.com/searchbot'. | Follows robots.txt; changes take about 24 hours to take effect. |
| GPTBot Documentation | Crawls content that may be used to train OpenAI's foundation models. Independent of OAI-SearchBot, though OpenAI may use one crawl for both purposes when both are allowed. User agent: Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.4; +https://openai.com/gptbot. | Follows robots.txt: disallowing it signals that content should not be used for training. |
| ChatGPT-User Documentation | Visits pages for individual user actions in ChatGPT and custom GPTs. Not used for automatic crawling or for search eligibility. User agent: Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; ChatGPT-User/1.0; +https://openai.com/bot. | Not guaranteed: OpenAI says robots.txt may not apply to user-initiated actions. |
| OAI-AdsBot Documentation | Checks landing pages submitted as ChatGPT ads for policy compliance and relevance. It visits only submitted ad pages, and its data is not used for training. | Not stated by OpenAI. |
ChatGPT Work's cloud browser has no robots.txt token. It signs its requests with HTTP Message Signatures (Web Bot Auth), with Signature-Agent set to https://chatgpt.com, and OpenAI explains how to verify and allow it.
Crawl sources
Where ChatGPT crawls from, and how to verify it
OpenAI publishes a JSON IP list for each crawler: openai.com/searchbot.json, gptbot.json, chatgpt-user.json and adsbot.json. It asks site owners to make sure their host or CDN allows the search ranges, because a firewall that challenges them makes a site ineligible for search in practice.
OpenAI publishes no reverse DNS hostnames or networks, so verification is by IP list: match the user-agent token, then check the address against the current JSON file. The lists change, so refresh them rather than hard-coding addresses.
ChatGPT Work's cloud browser is verified differently, by cryptographic signature. Its public keys are published at chatgpt.com/.well-known/http-message-signatures-directory, and CDNs including Akamai, Cloudflare and Vercel can check them for you.
Opting out
How to opt out of ChatGPT, or limit what it uses
Stay in search, refuse training
Disallow GPTBot and allow OAI-SearchBot. Disallowing GPTBot signals that your content should not be used for training, without removing you from ChatGPT search.
Official pageLeave ChatGPT search
Disallow OAI-SearchBot. Your pages will not be shown in search answers, though they may still appear as navigational links. Allow about 24 hours for the change.
Official pageUse noindex to disappear completely
If a page blocked from OAI-SearchBot is found through a third-party search provider, ChatGPT may still show its link and title. A noindex tag stops this, but OAI-SearchBot must be allowed to fetch the page to read it.
Official pageUser fetches: network controls only
OpenAI documents no separate opt-out for ChatGPT-User and says robots.txt may not apply to it. It publishes the user agent and IP list, so blocking at the network level is the only firm control.
Official pageRemove personal data from answers
Anyone, with or without an account, can ask OpenAI through its Privacy Portal to stop personal data appearing in ChatGPT's responses. Requests are assessed under laws such as the GDPR and need identity verification.
Official pageCorrect product information
Report wrong product details from the product menu in ChatGPT or through OpenAI's report form. Merchants can supply product feeds; Shopify merchants are covered through Shopify Catalog.
Official pageKeep your team's chats out of training
Turn off 'Improve the model for everyone' under Settings, then Data controls. Business, Enterprise, Edu and API data are not used for training by default.
Official page
Disallowing GPTBot affects future training only: models already trained keep what they learned. Leaving ChatGPT search means giving up one of the largest sources of AI referrals, so for most brands the right setting is to allow OAI-SearchBot and decide about GPTBot on its merits.
How we help
How we get brands cited in ChatGPT
The Coaley Peak GEO Research Centre researches and implements GEO for SEO agencies and in-house teams, with foreign markets as our speciality.
ChatGPT visibility audit
We ask ChatGPT the questions your customers ask, across markets and languages, and record whether you are cited, how you are described and which sources it trusts instead.
Crawl access that keeps you eligible
We check robots.txt, CDN and firewall rules against OpenAI's published IP lists, so OAI-SearchBot can reach you, with GPTBot set as you decide.
Pages ChatGPT can cite
We restructure key pages so their facts can be quoted on their own: clear claims, named products, prices and specifications in crawlable HTML, in every language you sell in.
Product feeds and shopping
For retailers, we get product data into the feeds ChatGPT's shopping results draw on, accurate and consistent with your site, so your products appear in its carousels.
Measurement
We segment utm_source=chatgpt.com traffic, track your citation share against competitors over time, and report it market by market.
More on our research, services and team at the Coaley Peak GEO Research Centre.
FAQ
ChatGPT GEO: questions
How does ChatGPT decide which sites to cite?
ChatGPT decides when to search, rewrites the question into targeted queries, and draws on OpenAI's own crawl, third-party search providers and partner content. OpenAI says placement is never guaranteed and ranking uses several relevance and reliability factors.
What is the difference between GPTBot and OAI-SearchBot?
GPTBot crawls content that may be used to train OpenAI's models; OAI-SearchBot surfaces sites in ChatGPT search. They are independent, so you can disallow GPTBot to refuse training and still allow OAI-SearchBot to stay in search.
Does blocking GPTBot remove my site from ChatGPT?
No. It signals that your content should not be used for training. To leave ChatGPT search you must disallow OAI-SearchBot, and to stop bare links appearing you also need noindex.
How can I see traffic from ChatGPT?
ChatGPT adds utm_source=chatgpt.com to the links in its answers, so visits from it can be segmented in your analytics.
What is ChatGPT's knowledge cut-off?
It depends on the model. For the API versions of the GPT-6 models now rolling out in ChatGPT, OpenAI publishes 20 April 2026 for GPT-6 Sol (paid plans) and 18 May 2026 for GPT-6 Luna (Free and Go), and 30 April 2026 for GPT-6 Astra. GPT-5.6 stops at 16 February 2026. Anything newer reaches ChatGPT through search.
Do ads in ChatGPT affect which brands are cited?
OpenAI says not. Ads on the Free and Go plans appear below answers, labelled as sponsored, and OpenAI says they do not influence responses. Product carousels are organic, not ads.
Want a free chat about ChatGPT?
Tell us which market and brand you are working on. If we can help, we will set up a free, no-obligation chat with Stephen or Ethan.
More engine guides
Related engines
Independence and trademarks. ChatGPT, GPT and OpenAI are trademarks of OpenAI. Other names are trademarks of their respective owners. Coaley Peak is independent: we are not associated with, endorsed by or a partner of OpenAI. This guide is our own research from public documentation and reputable reporting, checked in October 2026.
Reviewed by Stephen Grindley, research centre lead. Last checked .
Document reference: ISO_webpage_geo_engine_chatgpt_v1
Last modified: 7 October 2026
GEO Research Centre·Engine guides·ChatGPT