How to get mentioned by ChatGPT and AI search: the practical guide with a 12-point self-test
The question companies have asked us most often since the beginning of the year is: “Why does ChatGPT name our competitors and not us?”
The answer is rarely mysterious. In most of the cases we check it comes down to two or three technical details that have been sitting unnoticed on the website for years, plus one thing that happens off the website.
This guide has three parts: first a self-test of the prerequisites you can verify yourself. Then the documented technical requirements and the editorial recommendations. Finally the question of how to strengthen your own measurement and your direct customer relationships.
If you want to understand first what sits behind terms such as GEO, AEO and llms.txt: we took that apart in the basics article llms.txt, GEO and AI visibility. If you want to know what this work looks like as a service: What a GEO agency does.
Part 1: the self-test with 12 questions
Before we get to measures, it is worth looking at the starting point. Five minutes, twelve questions, and afterwards you know where to start and which parts you can skip.
Answer each question with yes or no. Every yes counts one point. The test takes about five minutes; for three questions you will need your website open.
| # | Question | How to check |
|---|---|---|
| 1 | Do your robots.txt rules allow the crawlers you want to allow for search or for user-initiated fetches, and do they keep those apart from training? | Open yourdomain.de/robots.txt |
| 2 | Does your main content stay readable when JavaScript is switched off? | Switch JS off in the browser, reload |
| 3 | Is your Google Business Profile filled in completely and kept current? | Open the listing |
| 4 | Do name, address and phone number match factually across the relevant platforms? | Compare the imprint with the directory entries |
| 5 | Is your company mentioned outside your own domain, in directories, trade press or forums? | Search your company name, exclude your own domain |
| 6 | Do your most important pages carry a named author with a role? | Review the pages |
| 7 | Does your content carry a visible update date when the substance has changed? | Review the pages |
| 8 | Are your prices or price ranges visible without a form? | Check the pricing page |
| 9 | Do your texts back central statements with numbers and linked sources? | Read the home page and the service pages |
| 10 | Do you have a direct channel of your own, a newsletter or mailing list, with real recipients? | Answer honestly |
| 11 | Are your company data and independent mentions correct in the places that matter to your customers? | Check the business profile, directories and trade sources |
| 12 | Have you checked in the last 30 days what ChatGPT, Perplexity and Gemini say about you? | Answer honestly |
Your score
0 to 3 points: sort out the basics first. Check indexing, robots.txt by purpose and the technical accessibility of your main content. This says nothing about whether any single AI system knows you or names you.
4 to 6 points: check content and data. Back your central statements, name who is responsible and keep company data correct. No automatic selection by an AI system follows from that.
7 to 9 points: a good working basis. Check structure, data consistency and your own measurement routine. Test changes with a fixed set of questions and a period you can retrace.
10 to 12 points: many checkpoints are met. Watch the real questions your customers ask and improve content and presence where they are demonstrably missing.
Work stages for your own AI visibility
The test above is our own compilation and provides no benchmark figures. We have no reliable general market average; use these work stages instead:
| Work stage | Checkpoint |
|---|---|
| Check the basics | indexing, accessibility and useful content |
| Define your own questions | collect the real search and advisory questions |
| Measure the starting point | document answers and referrals |
| Assess changes | retrace one change per period |
The test does not replace an analysis of the actual answers. It only helps you put the verifiable prerequisites in order.
Do you know whether AI already knows you today?
The self-test shows whether your website is prepared. It does not show what ChatGPT and Perplexity actually say about you. That is exactly what we check in the free website check: technical AI readability plus a live query on whether and how AI systems name your company today, word for word, even when the answers are empty.
Part 2: how to appear in ChatGPT answers, seven priorities
You now know where you stand. Here is what to do about it, in the order in which it takes effect.
The order separates documented technical requirements from editorial recommendations. No single point guarantees a mention or a recommendation.
Priority 1: check robots.txt by purpose
robots.txt is an access decision, not a universal visibility switch. Check the purpose of every identifier: blocking training is not automatically blocking search, and granting access is no promise of a mention.
Assess these roles separately:
| Identifier | Role and decision |
|---|---|
| Googlebot | Google Search including its search features; keep it crawlable for Google visibility. |
| Google-Extended | Control for Gemini training and grounding; not a crawler of its own and not a Google Search ranking signal. |
| OAI-SearchBot | OpenAI search; assess it separately from training. |
| GPTBot | OpenAI training; allow it only if you want to permit training. |
| ChatGPT-User | user-initiated fetch; separate from automated crawler access. |
| Anthropic identifiers | The roles differ per agent; check the current documentation before you change a rule. |
The most common mix-up concerns training and search. Googlebot builds the Google search index; Google-Extended controls how content already crawled by Google is used for Gemini training and grounding. OAI-SearchBot serves search, GPTBot serves training. Google crawler documentation · OpenAI bots
If you want to prevent training, make that decision deliberately for the training identifier. Do not copy a blanket bot list: the identifiers and their effects differ from provider to provider.
Priority 2: make content visible without JavaScript
For Google the rule is: Googlebot can process JavaScript when it is not blocked, but JavaScript SEO stays more complex. Deliver your main content reliably and check it with the Google tools. That says nothing about the rendering ability of every other AI product. Google: AI optimisation guide
Self-test: check the rendered page and its indexability with the Google tools. Switching JavaScript off indicates technical robustness, but it is not a complete test for every crawler or AI system.
Priority 3: appear outside your own domain
Company data and independent information outside your own domain help people place an offer. At the same time Google says: do not chase artificial mentions; they are not sustainable optimisation. Google: AI optimisation guide
Your own page explains your offer. An industry directory, a specialist article or a community answer can add independent information. Which source a single system uses is not predictable.
In concrete terms:
- Google Business Profile: fill it in completely and keep it current. Google names Business Profiles as one way to provide local information in search. → Guide to the Google Business Profile
- Keep company data consistent: check name, address and phone number on the website and in the relevant directories regularly.
- Relevant industry and review portals: maintain information only where it helps customers with a real decision.
- Specialist articles and communities: genuine participation instead of bought mentions. Google warns against inauthentic mentions.
How many mentions are enough?
A reliable threshold for the number of mentions is not documented. Assess the quality, relevance and correctness of the sources instead of chasing a monthly number:
| Checkpoint | Question |
|---|---|
| Company data | Is it correct and current on the pages that matter? |
| Independent sources | Do they help people decide? |
| Your own content | Does it answer the questions your audience actually asks? |
| Measurement | Can a change be retraced over time? |
In practice that means: single mentions are no reliable proof of success. Work continuously on information that is useful and verifiable for your audience.
Where these mentions come from
There is no reliable general ranking of channels for AI visibility. Choose formats by where your customers expect reliable information:
| Channel | Worth it when |
|---|---|
| Specialist article | you can contribute your own experience or verifiable expertise |
| Industry directory | your customers look there for suitable providers |
| Community | you can answer a question openly and helpfully |
| Customer review | it is genuine, voluntary and traceable |
Do not choose by which channels look unoccupied. What matters is whether the information there is correct, useful and findable for your audience.
Priority 4: back statements instead of asserting them
The GEO study compared nine interventions in a controlled experiment with model answers. Its visibility figures cannot be read as a general effect on websites, recommendations, leads or revenue. To the study
| Measure | What it did in the experiment |
|---|---|
| Quote sources | improved the experimental visibility metric |
| Statistics with evidence | improved the experimental visibility metric |
| Expert quotes with name and role | improved the experimental visibility metric |
Verifiable statements help readers place what they read. An example: in May 2026 Ahrefs observed across 137,000 domains that 97% of the llms.txt files present received no fetch in the month studied; that measures fetches, not later effect. Ahrefs
The figure of up to 115% named in the study refers to visibility in the experimental setup for sources at rank 5. It is no promise of rankings, mentions, leads or revenue for a single website.
Priority 5: show author and date
A visible author and a correct date help people place origin and currency. A general rate for AI citations by update date is not reliably documented.
Name a responsible person on important content and update the date when the content actually changes.
Since 2 August 2026 this line has had a second reason that has nothing to do with visibility: anyone writing texts with AI support can do without a notice only if a named person carries the editorial responsibility. One measure, two reasons → AI labelling obligation: what you have to label.
Important addition: do not update out of reflex. Maintain content when facts, offer or guidance require it. A new date without a change to the content is not a documented visibility lever.
Priority 6: publish prices so people can read them
Publish prices or comprehensible price ranges where that fits your offer. It helps people decide; an automatic selection by AI systems does not follow from it.
You do not have to name fixed prices. A public page with price ranges, packages and scope of work is enough and is machine-readable.
Priority 7: structure that can be extracted
Write clearly structured paragraphs so that people grasp the information quickly. Google recommends understandable sections and headings, but does not require special answer blocks. Google: AI optimisation guide
- The direct answer belongs at the start of a section, not at the end.
- State the core points completely and understandably.
- Write headings around the question your audience actually asks.
- Comparisons belong in tables, processes in numbered lists.
- Maintain structured data as part of normal SEO work; special markup is not required for Google's AI features.
A generally valid ranking of content formats for AI citations is not proven. Choose the format that answers your audience's question most reliably.
What you can skip
Keyword repetition. Write for meaning and readability. Google explains that its systems understand synonyms and context; no general percentage AI penalty can be derived from the GEO study. Google: AI optimisation guide
llms.txt as a visibility measure. Google ignores llms.txt for Google Search including its generative features. Ahrefs observed 97% of files without a fetch in the month studied across 137,000 domains; SE Ranking reports no correlation with AI citations across 300,000 domains. These are observational data, not proof of effect. Details in the basics article.
Chopping content into artificial pieces. Google explains that small text fragments are not required for generative search. Write so that people understand you; mass-produced content aimed at manipulating rankings falls under the spam policies regardless.
Guaranteed placements. AI answers are not deterministic. Anyone who promises them cannot keep the promise.
Part 3: why your own channel and your own measurement still matter
The third part is the uncomfortable one. It is about why the measures from Part 1 are necessary but not sufficient.
The calculation no longer adds up
The simplified model runs: visibility → click → visit → enquiry → revenue. In practice the attribution was already incomplete before AI search, for instance with several touchpoints or a change of device.
In AI search this chain breaks in two places.
First: answers can appear on the results page. For individual queries that can change the click paths. How often it happens and what it means for your website is something you have to check in your own data.
Second: measurement stays patchy. Google announced the generative AI performance report on 3 June 2026; the rollout was completed on 31 August 2026. The report covers Google features, not mentions in ChatGPT, Perplexity or Claude. Google: announcement and rollout
Direct attribution can be incomplete in individual cases. Assess visibility, qualified enquiries and revenue together instead of looking only at sessions.
General percentages for traffic and revenue cannot be transferred reliably to a single company. What counts is how your own enquiries and closed deals develop.
What makes AI traffic different
AI referral traffic and conversion differ by offer, query and measurement method. Check in your own analysis which referrals lead to qualified enquiries.
A single referral says nothing about purchase intent. Only the enquiries and deals that follow show its business value.
The consequence: you need something that belongs to you
Your own channels and a clearly managed brand stay valuable even when visibility arises on external platforms. Three practical building blocks are:
1. Your brand name as a search query. Brand searches can show whether people are looking for you specifically. They still depend on search systems and on demand.
2. A direct channel. A newsletter or a mailing list people subscribe to voluntarily adds to your contact routes. Delivery, unsubscribes and responses stay measurable; here too, reach is not guaranteed.
3. Presence on the right platforms. Maintain information where your audience makes sensible use of it. A general, fixed source preference of individual models is not documented.
This shift also affects what you report. What a metrics logic looks like that holds up in front of management and in reality is in What remains of SEO and what is new.
What this means in practice
- Measure brand and business results in addition to clicks. Track brand searches in Search Console and check manually every month whether ChatGPT, Perplexity and Google AI Overviews name you for your twenty most important questions.
- Build the mailing list before you need it. The best time was two years ago, the second best is today.
- Accept fuzzy attribution. Anyone who spends only where attribution is complete will switch off exactly the channels that will carry them in future.
- Be where your topic is discussed, not only where you broadcast yourself.
That is the real shift: the website is no longer the centre, the brand is. The website becomes one of several pieces of evidence that you exist and what you stand for.
See for yourself, no obligation
Free, without obligation, facts instead of promises: start the website check or book a first conversation. Within two business days you receive your findings by email. Then you will know whether working together is worth it.
Frequently asked questions
How long does it take until AI systems name my company?
In our experience several months. Technical corrections take effect faster; mentions on third-party sites and brand awareness need time. Anyone who promises faster results cannot back them up.
How do I get cited by Perplexity AI?
Start with indexing and technical accessibility for Google. With other providers, Perplexity included, keep search, user-initiated fetches and training apart; granting access is no promise of a mention.
Is there an AI visibility checklist for websites?
The twelve questions in Part 1 are our own compilation and cover the prerequisites you can verify yourself. Good content alone is not enough: your own content, correct company data and independent information serve different purposes, and a general weighting between them is not reliably documented.
How many mentions do I need?
There is no reliably documented threshold. Work on correct, helpful and relevant information instead of chasing a monthly number.
Can I see AI visibility in Search Console, or do I need ChatGPT brand monitoring?
For Google, yes. The generative AI performance report in Search Console measures Google features; other AI products have to be watched separately, which in practice is what ChatGPT brand monitoring means.
Does it do damage if I block AI bots?
That depends on the identifier and its purpose. Googlebot concerns Google Search; Google-Extended controls Gemini training and grounding. With OpenAI, OAI-SearchBot separates search from GPTBot for training. Check the provider's current documentation before you change anything.
Do I have to update my content more often?
Only when the substance changes. Maintain facts, offers and instructions; a new date without a change to the content is not a documented visibility lever.
Why is a newsletter relevant in an article about AI search?
A direct mailing list creates a route to people who subscribed to it voluntarily. You still have to measure its reach and effect yourself.
What does it cost to make a website readable for AI?
That depends on the starting point. Check indexing, technical accessibility and the content that is decisive for your customers first.
Isn't this just classic SEO?
To a large extent, yes. For Google the SEO fundamentals remain decisive: helpful, original content, technical accessibility and correct information. The GEO study provides no general percentage statement about keyword optimisation.
How to read the evidence: the GEO study is a controlled experiment with model answers. The Google documentation describes Google products. Neither supports guaranteed mentions, rankings, leads or revenue for an individual website.
Sources: Princeton GEO Study · Google: optimising for generative AI search · Google crawlers · Google: GenAI performance report · OpenAI bots · Ahrefs llms.txt study · SE Ranking llms.txt study
