My business doesn't show up in ChatGPT: why it happens and what to do

If you ask ChatGPT about your category and your business doesn't come up, it's almost always one of five things. They're below in the order worth ruling them out: from the cheapest to check to the slowest to fix. The first one takes ten minutes to verify and is surprisingly common.

Before starting, a clarification that saves frustration: the right question isn't “does ChatGPT know me?” but “does ChatGPT name me when the customer asks about the category?”. If you type your business's name and the model describes it, that means nothing — it only says the name was in its training data. What decides a sale is the other question, the discovery one: “which one should I choose for X?”. That's where most businesses disappear, big ones included.

Cause 1: the AI companies' bots can't read your site

It's the most frequent and the easiest to fix, and many companies have it without knowing. It happens when the site's robots.txt file blocks the crawlers from OpenAI, Anthropic, Google or Perplexity — sometimes because someone turned it on deliberately two years ago to “protect the content,” and sometimes because the security plugin or the CDN provider did it on its own.

How to check, with no tools: open yoursite.com/robots.txt in the browser and look for these names.

If any of them appears with Disallow: /, that engine can't read you. And mind the nuance: blocking GPTBot stops training, but it's OAI-SearchBot that decides whether you can be cited today in an answer with web search. You can have different rules for each; what you don't want is to have them by accident. It's explained in more detail in AI crawlers.

The block can also happen without robots.txt: a firewall or an anti-bot service that returns 403 to any visitor that doesn't look like a browser. That version doesn't show up in the file, and you have to detect it by requesting the page with the bot's user agent.

Cause 2: your site exists, but doesn't answer the question the customer asks

This is the most common cause once the previous one is ruled out, and it's uncomfortable because it isn't a technical error: it's an old editorial decision.

Most corporate sites are written around what the company sells: “Solutions,” “Our value proposition,” “Services.” The customer doesn't ask that. They ask “which one is right for a thirty-person company,” “how much does this cost in Florida,” “what's the difference between A and B.” A generative engine writes its answer from pages that contain the answer in a format it can extract and cite. If yours doesn't contain it, it can't cite it even if it wanted to.

What this changes in practice:

• One question per page, and the answer in the first paragraph. Not at the end, after three paragraphs of context. Engines extract short, self-contained passages.
• Headings that are literal questions. <h2>How much does it cost?</h2> works better than <h2>Investment</h2>, and not because of keyword tricks: because it makes explicit which question that block answers.
• Concrete figures, dates and conditions. A text that says “competitive prices” can't be cited. One that says “from US$ 79 per month, billed annually” can.
• Tables for comparisons. They're the format that best survives extraction, because each cell already comes labeled by its row and column.

Cause 3: you're not in the sources the engine consults

This is the point that separates real work from cosmetic work, and the one that meets the most resistance when explained, because it means working off your own site.

When ChatGPT answers a discovery question with web search turned on, it rarely cites the candidate businesses' own pages. It cites third parties: media comparisons, Reddit threads, industry directories, reviews, press coverage. Your product page competes at a structural disadvantage with those sources, because the engine reads it as what it is — an interested party — while it treats the third party as a referee.

The direct consequence: you can have the best site in your category and still not show up, because the five pages the engine is reading to answer don't name you. The work, then, isn't writing more on your site: it's finding out which pages each engine is citing for your category and getting into them.

And there's a worse, fairly common case: being in those sources, but wrong. An outdated listing that describes you with a product you no longer sell, or a comparison that puts you in the wrong column. The engine doesn't verify; it repeats.

Cause 4: your business is too new or too small for the model

There's a part of this that doesn't get fixed this week, and it's worth saying.

Models have two ways of naming you: what stuck in their training and what they find when they search in the moment. The first has a cutoff date and a strong bias toward what was already prominent when the model was trained. If your business is eighteen months old and has three hundred mentions on the internet, it isn't in that memory, and there's nothing you can do to put it there directly.

The second way is reachable, and it's where everything is at stake for a young business: that when the engine decides to search, it finds you. That's why causes 1, 2 and 3 matter so much more for a small business than for a big one.

A detail almost nobody measures, and one that changes how you read everything: the engine doesn't always search. In some of the runs it answers from memory without consulting the web, and in those runs no change to your site could have helped you. If you don't separate the answers where it searched from the ones that came from memory, you'll credit your work with movements that aren't its doing — and the other way around. At Apoda that's reported separately, as the activation rate, for exactly this reason.

Cause 5: you're measuring with the wrong questions

Sometimes the business does show up, and the problem is the thermometer.

Three mistakes that keep repeating:

1. Asking with the business's name. “What do you think of Acme?” always returns something. It doesn't measure visibility, it measures memory. The questions that work are the ones that don't name you.
2. Trying only once. The same prompt run twice can give two different answers: models are probabilistic. A single run doesn't tell “I don't show up” apart from “I didn't show up this time.” You have to repeat each question several times and read the range, not the data point. It's covered in SEO vs. GEO.
3. Asking in the wrong language. ChatGPT answers in Spanish by consulting different pages than it uses in English, and vice versa. If your customers search in Spanish, a measurement in English gives you a portrait of another market — and in South Florida, many businesses have customers on both sides.

And one more: location moves the answer. The same question can name different businesses from Miami than from Orlando.

In what order should you tackle it?

In this one, and no other, because the cost and timeline of each step are very different:

Step 2 comes before step 3 on purpose. Without a prior measurement, any later change is indistinguishable from noise, and you'll end up defending work you can't prove. Why that's so hard — and why nobody can give you a perfect proof — is in how to know whether your GEO work actually paid off.

How long does it take to change?

The industry benchmark is three to six months to see sustained movement, and what moves that timeline most is how saturated or competitive your market is. In one where almost nobody is working on their AI presence yet, it gets shorter: at Apoda we've seen results in 68 days. And if the problem was the technical block, it can change in weeks.

Anyone offering to get you into ChatGPT in thirty days for a discovery question is selling a certainty that doesn't exist today. What can be done in thirty days is knowing exactly where you stand, with the evidence saved, and being clear about which of the five causes is yours.

If you want the starting point without hiring anything, the free diagnostic asks ChatGPT your category's questions without naming you and emails you the full answers.