Why ChatGPT does not see the site: reasons and step-by-step verification

08.06.202610 min read

If ChatGPT does not mention your company, does not show the site in sources, or links to competitors, the problem is usually in one of four places: the OpenAI robot cannot retrieve the page, the desired page does not answer the question well, the facts about the company are not confirmed, or the site has not yet accumulated enough clear signals on the topic.

There is no separate form “add a site to ChatGPT”. OpenAI recommends allowing bypass for OAI-SearchBot, do not block published robot IP addresses on the hosting or CDN side and make public pages searchable. But one permit per robots.txt not enough: availability gives the right to participate in the search, and does not guarantee a mention or citation.

This instruction solves one specific problem: find the reason why the site does not appear in ChatGPT Search responses and create a fix plan. The general strategy of AEO and GEO is broken down in separate material, and the Google AI Overviews requirements are in Google AI search guide.

First, determine what exactly isn't working.

The phrase “ChatGPT does not see the site” can describe different problems. Record the symptom before making edits.

Symptom	Probable Cause	First check
ChatGPT doesn't even find the company name	there is no stable data about the brand, the site is new or closed for crawling	ask about a company with and without a domain, check `robots.txt`
The site opens upon direct request, but does not appear in responses to the topic	the page is accessible, but loses in relevance and evidence	compare the page's answer with the user's question
The title and link are shown, but the page content is not cited	`OAI-SearchBot` closed or not receiving content	check the rules for the bot, HTTP response and server logs
ChatGPT provides outdated information	There are conflicting data on the website and external sites	check prices, addresses, services, team and update dates
The answers include competitors, but not your company	competitors have a better match of intent, more verifiable facts or independent confirmation	analyze the sources of a specific answer, and not the entire competitor’s site
There are no referrals from ChatGPT, although the site is sometimes cited	the user receives a response without clicking or analytics does not highlight the source	check `utm_source=chatgpt.com`, referrals and conversions

Don't start with a massive rewrite of texts. First, separate technical inaccessibility from a situation where ChatGPT can read a site but does not select it to respond to.

Step 1: Check OAI-SearchBot access

For search answers, OpenAI uses OAI-SearchBot. This is a separate robot: it is responsible for the presence of the site in the search functions of ChatGPT. GPTBot used for content that can be used to train models, and ChatGPT-User can access the page based on user action. These settings do not need to be mixed.

If a company wants to participate in ChatGPT Search, but does not want to allow the content to be used for training, the rules can be divided:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

OpenAI indicates that after the change robots.txt It may take up to a day for systems to take into account the new rule. There is no point in checking the result after five minutes.

What to check besides robots.txt

Returns the required URL 200 OK, not 301, 403, 429 or 5xx.
Not on the page noindex and conflicting X-Robots-Tag in the HTTP header.
Canonical points to the same indexed page, not a different URL.
CDN, WAF and bot protection pass IP ranges OAI-SearchBot, published by OpenAI.
The body text is present in the HTML and does not appear only after a complex action in the browser.
The page is accessible without authorization, cookie wall and mandatory region selection.

It is important to check the server logs. Successfully opening a page in your browser does not prove that the robot is receiving the same response. A common situation: for an ordinary user, the server gives 200, and to an unfamiliar user-agent - 403 or the endless Cloudflare check.

Step 2: Make sure there is a page for the user's question

ChatGPT is not looking for “the best company website in general,” but rather a source for a specific part of the answer. The main page is rarely able to equally well confirm the cost of implementing a CRM, the time frame for developing an online store, and the SEO features of a medical clinic.

Take 15-20 real customer questions and match each question to the best page on the site.

User question	Which page should respond
How much does turnkey CRM development cost?	landing services with cost ranges and evaluation factors
How to transfer data from Excel to CRM?	separate instructions or section on migration with stages and risks
Who developed a similar system?	case with task, architecture and result
Which company implements CRM for real estate?	industry landing page with relevant cases
How to check an SEO contractor?	expert article with verification criteria

If the question is answered only by the paragraph “we provide a full range of services,” there is nothing to select or quote there. Specific conditions, restrictions, steps and confirmations are needed.

Quick page test

Open the URL and try to find answers to five questions in 20 seconds:

what exactly is being offered;
who is it suitable for?
how the work goes;
how long does it take and what does the price depend on;
what experience has confirmed.

If the answers are scattered across the page, hidden behind generic language, or missing, first correct the page itself. Special “optimization for neural networks” will not help here.

Step 3: Give a direct answer and then an explanation.

The page does not have to consist of short fragments, but each important section should quickly answer the stated question. The working structure looks like this:

Вопрос или проблема
Короткий ответ в 2-4 предложениях
Условия и ограничения
Пошаговое объяснение
Пример или доказательство
Следующее действие

For example, instead of the phrase “the cost is calculated individually,” it is more useful to write:

The cost of CRM depends on the number of roles, integrations, volume of migration and analytics requirements. A small first release is usually assessed after a description of the funnel and a list of systems with which the exchange is needed. Before evaluating, it is important to separate the must-have startup functions from the second-line automations.

Such a block does not promise a fictitious fixed price, but gives a person decision criteria. It is our own expertise, examples and verifiable details that distinguish a useful page from a retelling of general advice.

Step 4. Remove contradictions in the facts about the company

ChatGPT can find information on more than just one landing page. If the website, company cards, catalogs and publications name different addresses, services, positions or dates, the source becomes less clear.

Check the consistency of the following data:

brand name and legal name;
main domain and contacts;
cities and geography of work;
list of key services;
names and roles of specialists;
dates of publication and updating of materials;
prices, terms and conditions, if specified;
numbers in cases and periods confirming them.

The site should have related pages for the company, team, services, cases and authors. Organization, Person, Article, Service and BreadcrumbList are useful as a machine-readable description of already visible data, but markup does not correct weak or inconsistent content.

Step 5: Add Evidence Instead of Presentation

The phrases "market leader", "expert team" and "customized approach" cannot be verified. For a commercial topic, observable facts work more strongly:

a case with the original task, the work performed and the result;
the specialist's name, role and related publications;
screenshots of the interface or process, if applicable;
calculation method, checklist or fragment of a technical solution;
date of information update;
link to primary documentation next to the technical statement;
An honest description of the limitations and cases where the service is not suitable.

External references also matter, but buying chaotic “AI signals” is not worth it. It is more useful to receive specialized publications, expert comments, reviews with clear context, and company cards with up-to-date data. This not only helps ChatGPT but also the user who checks the contractor before bidding.

Step 6: Check if pages are cannibalizing each other

If there are five articles on a site with the same promise of “how to get into ChatGPT,” it is difficult for the search engine and internal editors to choose the main one. Each page needs a separate question.

For this cluster, the logical division is as follows:

this article - why ChatGPT does not see the site and how to find the reason;
AEO, GEO and AI SEO - differences in terms and choice of strategy;
citation-ready content - how to rewrite individual material for citation;
Google AI results - AI Overviews and AI Mode requirements;
AI crawler research - analysis of site crawling data;
neuropromotion in Yandex - a separate commercial task for Yandex.

There is no need to re-explain the entire AI search in every article. It is enough to give a short context and direct the reader to a page that goes deeper into the next question.

Step 7. Test the site with specific queries

One request “what do you know about company N?” shows almost nothing. We need a small test set tied to the client’s funnel.

Collect four types of requests:

Branded: “what NBM-IT does”, “what services NBM-IT provides”.
Categorical: “who develops a CRM for a real estate agency.”
Problematic: “how to automate the processing of applications from the website and telephony.”
Comparative: “custom CRM or Bitrix24 for a complex sales department.”

For each request, record:

whether the brand is mentioned;
is there a link to the site;
which page is cited;
what competitors and sources were used;
Are the facts accurate?
Does the answer change if the question is formulated differently?

It is better to repeat the check using the same list, rather than selecting a new successful prompt every time. Answers may vary, so a single mention cannot be considered a consistent result.

Step 8: Set up conversion measurement from ChatGPT

OpenAI adds parameter to referral URLs utm_source=chatgpt.com. Create a separate segment in your analytics for such transitions and do not mix it with all referral traffic.

The minimum report must contain:

Metrica	What does it show
Sessions with `utm_source=chatgpt.com`	were clicks from ChatGPT search responses
Landing page	what material or service brought the user
Conversion to a form, call or request	does the channel bring business results?
Brand requests	does interest in the company grow after mentions?
Share of mentions in the test set	does visibility become more stable?
Share of answers with correct facts	Is outdated information being disseminated?

No clicks don't always mean no visibility: the user can get a response without clicking. Therefore, combine web analytics with regular manual query reviews and CRM data.

Patch priority

Correct the reasons in this order:

Access: OAI-SearchBot, IP ranges, 200 OK, canonical, absence noindex.
Intent page: one clear question - one best URL.
Answer: a summary, terms, steps and example instead of general SEO text.
Evidence: cases, authors, sources, dates and verifiable facts.
Contacts: the service is related to articles and cases, the materials do not duplicate the intentions.
Reputation trail: current independent references to the company and specialists.
Dimension: test set of requests, analytics of transitions and conversions.

If the first step is not completed, content edits may remain invisible. If technically everything is available, but there is no good page in question, the allowed robot will simply see a weak source.

What will not solve the problem on its own?

File llms.txt

llms.txt is not an application to add a site to ChatGPT and does not replace access OAI-SearchBot. Google specifically states that it does not use such AI files for ranking in organic or generative searches. You can support a file for specific systems that can read it, but you shouldn’t expect an automatic increase in visibility.

FAQ markup without useful FAQ

Markup helps describe existing content, but doesn't turn general questions into strong answers. First write a useful block for the person, then check that the schema matches the visible text.

Mention the keyword in each heading

ChatGPT does not select a source based on the density of the phrase “how to get into ChatGPT”. Repetition makes reading worse and creates cannibalization between articles. What is more important is exact compliance with the task and completeness of the answer.

Mass generation of similar articles

Dozens of pages with rearranged words do not create expertise. For a complex topic, it is more useful to have one diagnostic material, separate instructions, a strong service page and several supporting cases.

Purchase “100 mentions for neural networks”

Without a real audience, context, and verifiable information, these placements become noise. Reputation signals should help a person verify a company, rather than imitate fame.

Test plan for 7 days

Day 1: technical access

Check robots.txt, meta tags, HTTP responses, canonical, CDN and server logs for OAI-SearchBot.

Day 2: request map

Collect 15-20 customer questions and assign a primary URL to each. Mark the questions that the site does not currently answer.

Day 3-4: two priority pages

Rewrite one commercial landing page and one expert article: add a direct answer, conditions, evidence, and the next step.

Day 5: Entities and Facts

Check the company name, services, team, contacts, dates and numbers in the cases. Correct inconsistencies on the website and key external platforms.

Day 6: Query Test

Run a single set of brand, category, issue, and comparison queries. Save sources and errors in answers.

Day 7: analytics and backlog

Set up a segment utm_source=chatgpt.com, prioritize tasks and set a date for the next measurement.

FAQ

Can I manually add a site to ChatGPT Search?

OpenAI does not provide a separate form for adding a website. To allow public pages to appear in search responses, please allow OAI-SearchBot, check access from published IP ranges and make content relevant to user queries.

Why does ChatGPT know the name of the site, but does not provide a link?

The title may have appeared from an external source or navigation data, while the content of the page is not accessible to the robot or is not suitable as a source of the answer. Check OAI-SearchBot, the actual HTTP response and the page that should close the request.

Do I need to enable GPTBot to get into ChatGPT search?

No. OpenAI separates the control of search and possible use of content for learning. For search answers OAI-SearchBot; access GPTBot can be configured independently.

Why does ChatGPT show a competitor even though our site is higher in Google?

Position in classic search results does not guarantee selection in a particular answer. A competitor's page may answer part of the question more accurately, provide better evidence, or be supported by other sources. Compare pages and citations for a specific query.

How long to wait after changing robots.txt?

OpenAI writes that it may take about 24 hours for search engines to take into account the new rule. Updating a specific page and changing answers may take longer; There is no guaranteed period for revaluation.

Is it possible to guarantee a site mention in ChatGPT?

No. OpenAI explicitly states that top positions cannot be guaranteed. You can eliminate technical blockages, improve the relevance and credibility of pages, and then measure dynamics on a stable set of queries.

Conclusion

When ChatGPT doesn't see a site, don't start with a trendy term or mass content. Find a specific gap: the robot does not receive the page, the site does not have a URL for the question, the answer is too general, the facts contradict each other, or the experience is not confirmed by anything.

The correct sequence is to open access OAI-SearchBot, select priority queries, enhance the best pages, link them to services and cases, then measure mentions and conversions. If you need external diagnostics of this system, start with site audit or solve the problem on the page AI SEO promotion. The practical result of this approach is shown in case AI2SEO.