Why your brand does not appear in ChatGPT, Gemini, or Claude

Why is your brand not showing in ChatGPT even though the site ranks in Google? You ask a category question and see competitors in the answer but not your company. The same pattern may repeat in Gemini, Claude, Perplexity, or Google AI Overviews. Teams often jump to one of two conclusions: "AI is wrong and nothing can be done" or "we need to rebuild the entire site."

Both reactions only burn time. In most cases, the problem is not one, but consists of several signals at once. Here it is more important not to panic, but to go through the check in the correct order: from the request itself and the availability of crawlers to the content, external mentions and brand positioning.

Start with an AI visibility audit

Before looking for the cause, you should make sure that the problem really exists. Realistic minimum to talk about stable invisibility:

3-5 repetitions of the same query on different days;
verification in several models (ChatGPT, Gemini, Claude, Perplexity);
a fixed mode, with or without web search;
the date, wording, and model recorded for each measurement;
comparison across at least two query types, such as category-level and comparison prompts.

If after that the brand does not appear, then we check each possible block in turn.

Kozak compares three repeated checks across four AI models instead of drawing a conclusion from one test

Diagnostic Tree

Before diving into the causes, it is useful to see the general logic. Here's a simplified tree:

What do we see in the answer	The most likely cause	Where to go next
The model does not understand the query, responds in a generalized way	Noise query - it does not intend to choose	Chapter 1
AI Cites Third-Party Sources, But Your Brand Is On The Site	Lack Of Content For Selection Scenario	Chapter 3
AI Cites Sources Where Your Brand Doesn't	Weak External Presence	Chapter 4
AI Cites Your Site But Doesn't "Need" URLs	Structural/Technical Page Problem	Chapter 5
One model has a brand, others do not	The logic of each model is different	Chapter 6
In the answer without web search, we are not at all	Weak trace in training data	Chapter 7
Brand is mentioned vaguely, confused with another	Name conflict or lack of positioning	Chapter 8
The site is not really accessible to AI crawlers	Technical blocking	Chapter 2

Then — one by one.

Reason 0. A query alone does not lead to a brand

The first thing to check is whether this request should give the brand name in the answer at all. If you ask "what is CRM", the model answers explanatory, without brands — and that's okay. Questions that you need to wait for a mention look different:

category-level with a decision intent ("which CRM should a sales team of up to 20 people choose?");
comparative ("HubSpot vs Pipedrive for SaaS");
problematic ("how to automate customer accounting in services");
clarifying with budget, region, segment.

If your monitoring is based mainly on information requests, this is not "no brand", this is a noise pool. For more details, see How to Choose the Right Queries for AI Visibility Monitoring.

Reason 1. Technical accessibility of the site for AI crawlers

A common cause is that search crawlers cannot access the site or a CDN blocks them before robots.txt is evaluated. Separate crawlers used for AI search, model training, and user-initiated retrieval.

User-Agent	What is it
`OAI-SearchBot`	OpenAI search crawler used to surface sites in ChatGPT Search
`GPTBot`	Separate OpenAI crawler for content that may be used to train models
`Claude-SearchBot` / `ClaudeBot`	Separate Anthropic crawlers for search and model training
`PerplexityBot`	Perplexity search crawler
`Googlebot`	Google Search crawler; pages must be indexable to appear as supporting links in AI Overviews and AI Mode
`Google-Extended`	A control token for some generative uses by Google; it does not control Google Search indexing

Specific recommendations for robots.txt for most of them are public: OpenAI documentation (OpenAI Bot docs), Bing Webmaster Guidelines (Bing), Google "Overview of Google crawlers" (Google).

Blocking one crawler does not disable every AI surface. For example, blocking GPTBot is not the same as blocking OAI-SearchBot, and Google AI Overviews rely on Google Search controls rather than Google-Extended. Test the relevant user-agent separately and review server logs because a CDN or WAF can block requests regardless of robots.txt.

Reason 2. No content for the scenario of choice

Many sites sell well but explain poorly. They have a lead form, benefits, pricing, and case studies but no pages that answer the buyer's specific questions before a decision.

It is easier for AI models to use pages that clearly explain:

for whom this decision is made;
what task it closes;
how it differs from alternatives;
in which cases it is suitable and in which it is not;
what selection criteria are important.

Signals in monitoring:

AI cites competitors' explanatory pages ("what is...", "how to choose...", "X vs Y");
the brand appears only in narrow brand queries;
the model "does not know" in what context to recommend you.

Working formats: explanatory articles, guides, FAQs, comparisons, glossary of terms, cases with numbers. For details, see Which pages of the site are most often included in AI responses.

Reason 3. The brand is poorly represented in external sources

AI responses in your niche usually rely on more than just the brand's website. If the answers constantly quote media, ratings, catalogs, and selections, and your brand is not there or it is poorly represented, the model does not "see" you in the context of the niche.

It's not "you have to buy more links". It is more useful to think about thematic presence:

in which materials of industry media competitors are already mentioned;
which domains are repeated in AI responses;
where your brand should be in the natural order of things (Clutch, G2, Capterra, profile ratings, annual selections of industry media);
whether your profiles in the catalogs are relevant and whether there are cases/reviews there.

There is another layer that is easy to underestimate - large content partnerships that directly affect citations:

Google and Reddit have a content access agreement (Reddit, 2024) — Reddit discussions often end up in AI Overviews.
OpenAI has licensing agreements with News Corp, Axel Springer, Le Monde, Vox Media, The Atlantic, Time, Financial Times, and others (OpenAI News Partnerships).

This means that appearances in relevant publications and discussions have an indirect but stable effect on citations.

Kozak builds a bridge from the website through media, reviews, and directories to an AI answer

Reason 4. AI cites the site, but not the intended URLs

Sometimes the site is in the sources — but the model cites a secondary page, an old post, or template material instead of your strong commercial page. This is a structural problem.

What to check on your landing page:

H1 exactly matches the real request of the user;
on the first screen — 2-4 sentences of direct answer;
there is an H2/H3 structure, lists, tables, FAQ block;
there are schema.org markings (FAQPage, Article, Service, Product, BreadcrumbList);
the page is updated and has a fresh date;
there are internal links from 5-8 relevant pages;
there are no conflicts of cannonicals and doubles.

It is often enough to "finish" the landing page rather than create a new one.

Reason 5. Different model logic

Just because you're in ChatGPT doesn't mean you'll be seen by Gemini, Claude, or Perplexity. The models have different training bases, different citation logic, different depth of web search, different dates of the "cross-section of knowledge".

Common scenarios:

ChatGPT Search relies heavily on the current results of Bing and its sources — it cites fresh pages well;
Gemini in Search relies on the Google index — correlates with classic SEO more strongly than others;
Perplexity is very dependent on the current SERP and often quotes Reddit, Wikipedia, GitHub, specialized media;
Claude (web search) shows fewer sources, but citations are usually of high quality;
Claude and ChatGPT in non-grounding mode are much more dependent on training data.

Conclusion: we always check on at least two models. Otherwise, there may simply be no "brand failure" — it is local for one provider.

Kozak tunes and compares signals from ChatGPT, Gemini, Claude, and Perplexity

Reason 6. Weak trace in training data

If the brand does not appear even in answers without web search, this is a signal of the training layer. ChatGPT, Gemini, and Claude training data are generated from open sources as of a certain date (cutoff). If your brand has had little representation on the web in the last 2-3 years, the model simply hasn't gathered a stable context about you.

What works for this layer:

stable presence on the open Internet (media, specialized platforms, forums);
a Wikipedia page, if the brand meets the notability criteria;
the same brand description on different resources (one name, one context, the same wording of services);
stable external mentions for 6-12 months.

This layer changes more slowly than web-grounded answers; the realistic horizon is months, not weeks.

Reason 7. Name conflict or fuzzy positioning

A special case is that the model confuses a brand with a homonym. The name coincides with a generic term, other business or media object. In the answer, then a more "understandable" player appears.

What to do about it:

everywhere in external brand descriptions we use the same formula "name + category + country/segment";
on the main page, one sentence about who you are for and what your role is in the category;
in meta-description and H1, use the same formula;
in Wikipedia/directories/Crunchbase, we record the same descriptions.

If the positioning is "blurred", this is not a task for SEO, but for marketing.

Self-diagnosis checklist

Quick passage through the main points:

If the answer is "no" to four or more items, you have a concrete backlog rather than a vague goal to "improve AI visibility."

How to build an action plan

The sequence that usually gives the result is:

Check the technical availability of the site (1 day).
Make a full-fledged monitoring diagnosis: 40-80 requests × 4 models × 3 repetitions (1-2 weeks).
Classify the found problems by type: content, structural, reputational, positional.
Make a plan for the quarter: separately for SEO/site, content and PR.
Run a follow-up measurement after 4-6 weeks and adjust the plan.

For more information on converting monitoring into a plan, see How to turn an AI visibility report into a plan for SEO, content, and PR.

Kozak repairs the weakest link between access, content, authority, and follow-up measurement

Frequently Asked Questions

How do I get my company listed in ChatGPT? There is no submission directory that guarantees inclusion. Make the site accessible to search, describe the company and its services clearly, keep external profiles consistent, and test the result with a stable query set.

How quickly can a brand appear in answers? There is no fixed timeline. Search-backed answers can change after recrawling and source updates; answers without live web search may take much longer to reflect new information.

Is it possible to "ask" a model to start recommending a brand? Not directly. Models rely on public signals, so the work focuses on improving those signals.

Should I block AI crawlers? Do not block all of them by default. Decide by role: search crawlers affect search-backed visibility, while training crawlers control a different use of public content. Protect private pages with access controls rather than robots.txt.

Can the problem be solved by SEO means alone? Sometimes it is - if the reason is in the structure of the pages or in technical accessibility. But if the model does not have your external trace, you cannot do without PR.

What are the first three steps if there is no time for a full diagnosis? Check access for the relevant search crawlers, improve three to five priority pages with clear answers and accurate structured data where appropriate, and update profiles in the main industry directories for your niche.

What else to read

How we do it in VYDAI

You can run this diagnosis manually, but the workload grows quickly: 80 prompts, four models, three repetitions, recorded dates, citations, and competitors for the first measurement alone.

VYDAI automates this part. For each prompt, it shows which models mention the brand, which URLs they cite, which competitors appear beside it, how the brand is described, and whether the mention is an actual recommendation. That evidence makes it easier to identify causes and build an action plan.

To see where your brand drops out, create an account or view the demo.