Why your business doesn’t appear in ChatGPT: seven technical reasons and how to fix them

Strategy · GEO

You searched for your business in ChatGPT or Gemini and it wasn’t there. Or worse: a competitor you know well was — described with data you don’t recognise, recommended for reasons that don’t reflect what you actually do. The question that follows is always the same: why them and not me?

The answer is rarely one thing. It’s a combination of technical, content and architecture factors that determine whether an AI model can find you, understand you and cite you. This article documents the seven most common failure points — with a concrete fix for each one.

First, understand this: AI doesn’t search, it selects

When you type a question into Google, the search engine returns a ranked list of pages. You choose which one to open. When you type the same question into ChatGPT or Perplexity, the model builds a direct answer from sources it has already processed or retrieves in real time. There is no list. There is one answer, with two or three cited sources if you’re lucky.

This changes the logic of visibility entirely. In Google you compete for a position among ten. In generative AI you compete to be in the answer — or you’re not there at all. And the criteria that determine who makes it into that answer are not the same criteria that determine who ranks on page one.

GEO (Generative Engine Optimization) is the discipline that works specifically on those criteria: what makes a model select your content as a reliable source when answering your potential customers’ questions. In English-speaking markets the term GEO is standard, alongside AEO (Answer Engine Optimization) and LLMO (Large Language Model Optimization). The seven reasons below are the most common failure points we find when analysing why a business doesn’t appear.

The seven reasons AI can’t find you

1. AI crawlers can’t get into your website

Before a model can cite you, its crawlers need to have been able to read your website. If your robots.txt file blocks GPTBot, ClaudeBot or PerplexityBot, the model will never have processed your content — regardless of how good it is.

To check, open your browser and go to yourdomain.com/robots.txt. Look for lines like these:

User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

If they appear, the corresponding models can’t crawl you. The fix is to remove those lines or change Disallow to Allow.

This is more common than it sounds. Research suggests fewer than 15% of websites actively optimise their crawler access for AI models — and most blocks aren’t deliberate decisions. They arrive as part of default configurations from hosting providers, CDN platforms or security plugins that bundle AI crawler restrictions without flagging them clearly. The result is a business that has invested in content and has inadvertently told every AI model not to read it.

Deslumbra IA’s own robots.txt follows the standard WordPress default: it allows all bots under User-agent: * and only blocks /wp-admin/. Simple and intentional.

2. You don’t have an llms.txt

Your robots.txt controls access. An llms.txt goes one step further: it’s a plain text file that explains directly to AI models who you are, what you do, what content you have and how to interpret it. It isn’t mandatory — models function without it — but its presence is a signal of technical intentionality that no other format transmits in the same way.

The file sits at the root of your domain (yourdomain.com/llms.txt) and follows a simple Markdown structure. The minimum useful content includes: a description of the business and its activity, services with their URLs, the sectors where you have documented authority, and a reference to the blog or content that best represents your own knowledge. The more specific and verifiable, the more useful for the models.

You can see how we’ve implemented it at Deslumbra IA directly at deslumbraia.com/llms.txt.

3. Your business doesn’t exist as an entity for the models

Language models don’t just read web pages — they build a representation of the world from what they find. For a business to exist as a recognisable entity in that representation, it needs consistent signals across multiple sources: a Schema Organisation markup on the website, a Google Business Profile, mentions in verifiable directories, and coherence between the business name, description and activity across all those places.

If your business has been operating for three years but only appears on its own website, the models treat it as effectively invisible as an entity. Not because you don’t exist, but because there aren’t enough external signals to confirm that you do.

The minimum verifiable fix: install Schema Organisation on your homepage (Google Rich Results Test confirms whether it’s correctly implemented), create or claim your Google Business Profile, and make sure the business name is consistent everywhere you appear online.

4. Your content describes your business, it doesn’t answer your customers’ questions

This is probably the most frequent reason and the hardest to see from the inside. A corporate website talks about the business: who we are, what we do, why we’re good. An AI model, when answering a purchase decision question, looks for content that answers that question with explicit criteria and comparison between options.

We documented this in the UK private health insurance experiment: not one corporate website for Bupa, AXA Health or Vitality appeared as a cited source across the three incognito platforms we tested. The sources controlling the answers were WeCovr, Drewberry, MoneySavingExpert and Which? — brokers, comparison sites and consumer organisations that publish exactly the kind of content models need to answer decision questions. The insurer with the best product didn’t win visibility by having a better corporate website. Visibility went to whoever answered the user’s question most directly.

The fix isn’t to remove corporate content. It’s to add content that answers the real questions your customers ask before they hire you: comparisons, decision guides, FAQs with concrete answers, documented cases with verifiable data.

5. There aren’t enough verifiable external mentions

Models build their knowledge about a business from what they find across multiple sources, not just the company’s own website. Google reviews, mentions in sector publications, listings in relevant directories, appearances in specialist forums — all of this contributes to a model having enough context to include you in a response.

An important nuance that the gestoría experiment made clear: reviews help but they aren’t sufficient on their own. Abaq had 72 Google reviews with a 4.9 out of 5 rating and was invisible across all four AI platforms we tested. Reviews are a trust signal, not a direct citation mechanism. What models also need is external content that talks about the business in the context of the questions its customers are actually asking. See the full case in A gestoría with a 4.9 rating and a genuine edge on international billing. AI has never heard of it.

The lever sits in the sources models already cite in your sector. Identify which ones they are — search for yourself in incognito mode on ChatGPT and Perplexity and look at which sources appear — and work on your presence in those sources before creating new content on your own website.

6. Your structured data is wrong or missing

Schema markup is the technical layer that tells models — and Google — what type of content each page contains and how to interpret it. Organisation on the homepage, Article on blog posts, FAQPage on frequently asked questions, Product or Service on service pages.

Check your Schema status in Google Rich Results Test (search.google.com/test/rich-results). If you have nothing implemented, the starting point is Organisation on the homepage and FAQPage on your frequently asked questions section.

7. AI knows you, but only through what others say about you

This reason is different from the others because it has no direct technical fix. If the content about your business online is produced primarily by comparison sites, directories or aggregators, the narrative models build about you isn’t yours — it belongs to those sources, with their editorial criteria, their data (which may be out of date) and their commercial interests.

We saw this clearly in the UK private health insurance experiment: four AI platforms answered questions about Bupa, AXA Health and Vitality citing exclusively WeCovr, Drewberry, MoneySavingExpert and Which?. No corporate website for any of the three insurers appeared as a source. The result was that different platforms produced different assessments of Vitality based on how each one weighted those same external sources — without Vitality having participated in either response.

The underlying fix is to influence the sources models already cite, not just improve your own website. That means understanding what each source says about you, what data it works from, and what concrete actions can update that narrative. It’s distribution and relationship work, not just content work.

How to find out which of these seven reasons applies to your business

The basic diagnostic can be done in under half an hour, with no paid tools.

1. 1Check your robots.txt: go to yourdomain.com/robots.txt and look for whether any AI crawler is blocked. If there are no Disallow lines specifically targeting GPTBot, ClaudeBot or PerplexityBot, you’re fine on this point.
2. 2Search for yourself across three platforms in incognito mode: open ChatGPT, Gemini and Perplexity without an identified session and ask directly about your sector and type of service — not your business name. Note whether you appear, in what position, and which sources the models cite when they mention you or your competitors.
3. 3Check your Schema in Google Rich Results Test: enter your homepage URL and your service page URLs. The test tells you which Schema types it detects and whether they have errors.
4. 4Check what external sources say about you: search your business name in the publications, comparison sites or directories the models cited in step two. If you don’t appear in any of those sources, you have a clear indication of where the problem sits.

These four steps don’t give you a complete diagnosis — some factors require deeper analysis by sector and by platform — but they tell you where in the problem you are.

There’s a trap in this process that deserves its own article: asking AI to evaluate your AI visibility. The model will tell you your content is well structured, your strategy makes sense, you just need minor adjustments. We documented this with data from a study published in Science: models have a structural bias towards affirmation that makes them particularly unreliable as auditors of their own citation criteria. If you want to understand why that diagnosis always comes back positive, it’s explained in detail in I asked ChatGPT whether my GEO strategy was working. It said yes.