ChatGPT SEO: How to Get Your Business Found in ChatGPT

Blog · ChatGPT SEO

ChatGPT SEO means getting ChatGPT to mention your business in its answers and link to your website as a source. There is no fixed ranking like on Google. ChatGPT either answers from its model knowledge, meaning training data up to a cutoff date, or it searches the web when needed and links the pages it relies on. The system decides from the question whether to search. For you, that means three things: OpenAI's search crawler must be able to reach your site, your content has to answer questions directly and with evidence, and other websites should confirm what you say about yourself. The only reliable way to measure it is to ask real customer questions regularly and analyse the answers.

In brief

ChatGPT answers from model knowledge or from a web search, and only web search answers link to sources.
OAI-SearchBot decides whether your pages can appear in ChatGPT search answers; GPTBot only concerns training.
Research shows that statistics, quotations and cited sources raise visibility, while keyword stuffing does not.
Measure with realistic prompts, repeated runs and a documented search mode.

Published 25 Sep 2026Nikolai Schöbel and Jeremias Burger9 min read

Nikolai Schöbel and Jeremias Burger

Co-founders of Scalableloops GmbH. Nikolai Schöbel leads online marketing and AI strategy, Jeremias Burger the AI architecture. Both build AI systems and train teams on them in their own agency work.

On this page

What is ChatGPT SEO?
How does ChatGPT build its answers and choose sources?
Why the same prompt gives different answers
Can ChatGPT actually read your website?
What kind of content does ChatGPT like to cite?
Why mentions on other websites matter
How do you measure whether ChatGPT mentions you?
Frequently asked questions
How to find out where you stand in ChatGPT
Where the information on this page comes from

Basics

What is ChatGPT SEO?

ChatGPT SEO is about your business showing up in ChatGPT's answers: as a recommendation, as an example or as a linked source. People ask differently there than on Google. Instead of two keywords, they write full sentences with their situation, budget and region, such as "Which supplier in Bavaria builds conveyor systems for small production runs?". The answer is a single piece of text in which only a few names appear. If yours is not among them, you are not considered at that moment.

ChatGPT SEO is part of generative engine optimization (GEO), the practice of optimising for AI systems that write their own answers. Much of classic search engine optimisation still helps, such as clean technology and good content. What changes is the goal: you are no longer working towards a position in a list, you are working towards being recognised as a credible answer. For the wider picture, read our article on AI SEO.

900m

weekly users were reported by OpenAI for ChatGPT in February 2026. Part of the research that used to start on Google now starts in a chat.

OpenAI, reported by ALM Corp and TechnologyChecker.io

How it works

How does ChatGPT build its answers and choose sources?

ChatGPT has two ways to answer a question. The first is model knowledge: the language model processed large amounts of text during training and writes its answer from that. This knowledge stops at a cutoff date, and the answer names no sources, because it comes from learned patterns rather than from individual pages.

The second is web search. ChatGPT has been able to search the web since the end of October 2024. The system decides from the question whether a search is needed, and users can also trigger it themselves. ChatGPT then reads current pages, summarises them and links the sources inline and in a sources panel. Questions about suppliers, prices, opening hours or recent developments often go this way, while general explanations are frequently answered from model knowledge.

The distinction matters because the two paths respond to different levers. You influence model knowledge slowly and only indirectly, through what has been published about you over the years. Web search responds to what is on your website and on other sites today, provided the search crawler is allowed to read it.

	Model knowledge	Web search
Where does the answer come from?	Training data up to a cutoff date	Pages read at the moment of the question
Are sources linked?	No	Yes, inline and in the sources panel
How current?	As old as the training data	Up to date
Which OpenAI crawler?	GPTBot (training)	OAI-SearchBot (search), ChatGPT-User for user-requested fetches
What can you influence?	Long term: mentions and reputation online	Short term: crawl access, content, third-party sources

From our measurements

Why the same prompt gives different answers

In our AI visibility analysis we ask AI systems the questions real customers ask and evaluate every answer individually. One pattern keeps coming back: the result depends heavily on whether the question was asked with or without web search. The providers' APIs answer from model knowledge unless search is switched on, while in our observation the chat interface often adds a search for supplier and comparison questions.

An example from an analysis for an industrial company: asked through the API without web search, an AI system cited no source at all. The same question with search enabled linked to the company's reference page. Both measurements were correct; they simply measured different things. Since then, every one of our reports states which mode was used.

Answers also vary between two identical requests. A single sample therefore shows a direction, not a fixed position. A finding only becomes reliable across several runs and many questions.

Technical

Can ChatGPT actually read your website?

The first check is the least glamorous. OpenAI runs several crawlers with separate jobs. OAI-SearchBot collects pages for search; according to OpenAI, sites that block it in robots.txt are not shown in ChatGPT search answers. GPTBot collects content for training future models. Blocking GPTBot is a decision about training, and your pages stay available for search. ChatGPT-User fetches pages when a user asks for something in the chat.

In our GEO audits we therefore start with robots.txt, then look at firewall and bot protection settings, which can turn crawlers away even when robots.txt is open. Next comes the question of whether key content is in the delivered HTML or only loaded later via JavaScript, and whether Schema.org structured data describes your organisation, services, articles and FAQs in a machine-readable way. In our view, structured data is basic equipment rather than a magic trick: it helps systems classify you, but it does not replace good content.

A frequent question concerns llms.txt. It is a proposal by Jeremy Howard from 2024, not an adopted standard. Google's John Mueller has said more than once that no AI system uses the file. It does no harm, but it does not belong at the top of your list.

Content

What kind of content does ChatGPT like to cite?

The most thorough study so far is the GEO paper by researchers from Princeton and other institutions, presented at KDD 2024. They changed content in targeted ways and measured how much it then appeared in AI answers. Visibility rose by up to 40 percent. The strongest methods were adding statistics, quotations from credible sources and source citations. Classic keyword repetition did not help and in some tests performed worse than the unchanged text. One limitation: most measurements were made on simulated answer engines, not directly on ChatGPT.

This matches what we see in our analyses. Pages that answer a question in the first paragraph and back it up with a number or a source are easier to cite than pages with long introductions. Four building blocks make content usable for AI systems:

01
Answer
Up front
The key statement comes in the first two or three sentences, the reasoning follows.
02
Evidence
Numbers with sources
Concrete values, standards, time frames and references instead of vague adjectives.
03
Structure
Questions as headings
Subheadings in your customers' words, plus tables and an FAQ block.
04
Freshness
Visible date
A clear last updated date and regular reviews of important pages.

Third-party sources

Why mentions on other websites matter

An AI system checks what it reads about a company. A statement that only appears on your own website is a claim. If it also appears in a trade article, an industry directory or customer reviews, it becomes confirmed information. This idea underpins our white paper on visibility in AI systems, and it shows clearly in our analyses: where brands are named, the linked sources often come from other sites rather than from the company's own website.

In practice: keep your listings in your industry's directories consistent in name, address and service description. Work towards coverage in trade media, case studies on partner sites and genuine reviews. A link is welcome, but even an accurate mention in a credible context helps, because it confirms what your website says.

Measurement

How do you measure whether ChatGPT mentions you?

Classic metrics such as position and click-through rate fall short here, because there is no fixed ranking and many users get their answer without clicking. Our AI visibility analysis therefore measures along three questions: Are you mentioned? How are you described? Where do you stand against competitors? It is based on four types of prompts that real customers use: recommendation questions, problem questions, questions from a specific role and questions about standards or specifications. We deliberately leave your brand name out of the prompts, because we want to know whether the AI brings you up on its own. More in our article on how to measure AI visibility.

On your website, part of the visits from ChatGPT can be attributed. ChatGPT often appends the parameter utm_source=chatgpt.com to linked URLs, which shows up as a source in your analytics. Some visits, however, arrive without referrer information and are counted as direct traffic. The analytics figure is therefore a lower bound. Rising searches for your brand name and more direct enquiries by phone or email are additional signals.

Metric	What it shows	How we collect it
Share of mentions	How many relevant answers include your business, compared with competitors	Count across all prompts and AI systems
Mention or citation	Whether you are only named or linked as a source	Review of linked sources per answer
Position in the answer	Whether you appear as the first, second or last option	Rank within lists
Accuracy	Whether details about services, location and products are correct	Check against your verified data
Sentiment	Whether the description is positive, neutral or critical	Rating of every mention
Visits from ChatGPT	How many users come to your site from answers	Analytics, source chatgpt.com, as a lower bound

Frequently asked questions

Can you rank number one in ChatGPT?

There is no fixed ranking like on Google. ChatGPT writes every answer anew, and the same question can produce different results. A better goal is to be mentioned in as many relevant answers as possible and linked as a source.

Where does ChatGPT get its information?

Either from model knowledge, meaning training data up to a cutoff date, or from a web search at the time of the question. Only web search answers link to the pages used as sources.

Do I need to allow GPTBot in robots.txt to be found in ChatGPT?

For search, OAI-SearchBot is what counts. GPTBot concerns model training. You can block GPTBot and allow OAI-SearchBot, and your pages remain available for ChatGPT search answers.

Do I need an llms.txt file for ChatGPT?

Not as a first step. llms.txt is a proposal, not a standard, and according to Google no AI system currently uses it. Crawler access, clear content and third-party sources come first.

How can I see visitors who come from ChatGPT?

Many of them appear in your analytics with the source chatgpt.com, often via the parameter utm_source=chatgpt.com. Some arrive without referrer information and are counted as direct traffic, so the figure is a lower bound.

How long does it take for changes to show in ChatGPT?

It depends on the path. Changes to your website and robots.txt can show up in web search answers relatively quickly. Model knowledge only changes with new model versions. Repeated measurements over several weeks give reliable results.

Next steps

How to find out where you stand in ChatGPT

01
Collect prompts
Write down ten to twenty questions your customers actually ask, without your brand name.
02
Check crawlers
Review robots.txt and bot protection to make sure OAI-SearchBot can read your pages.
03
Review answers
Ask the prompts with web search on and record mentions, sources and accuracy.
04
Sharpen content
Update your key pages with a direct answer, sourced figures, a table and an FAQ.
05
Measure again
Repeat the same prompts at fixed intervals and compare the results.