You typed your category into ChatGPT, read the answer, and your company was not in it. Three competitors were. One of them launched after you did, and one of them is worse than you, which somehow makes it more annoying rather than less.
This is the most common result there is. Not the unusual one, the default one. Assistants build answers out of pages they can retrieve and quote, and most companies have never written the pages that make that easy. The good news is that the fix is content you were probably going to write anyway.
Before any of that, the check itself is worth getting right, because most people do it in a way that guarantees a misleading answer.
- A zero is normal, not a fault. Most sites under a few years old are absent from AI answers in their category.
- The usual cause is missing citable content, not a technical block. Assistants name brands they can quote a clear sentence about, and generic marketing copy gives them nothing to lift.
- Never check by asking about yourself. Ask the question a buyer asks, with no brand name in it, or the model will simply agree that you exist.
- The useful part of a zero is who is in your place. AI visibility tracking in ContextBolt SEO returns the leaderboard, the pages AI cites and twelve months of history on the first run, from inside Claude or Cursor, for $35 a month.
- Expect weeks, not quarters. AI answers rebuild continuously, so a page that lands this month can show up next month.
First, check it the right way
Three checks, in this order. Skipping straight to the third is how people end up confidently wrong.
The buyer question
Ask the question your buyer asks, with no brand name in it. “Best CRM for a five-person startup.” “Project management tool that works offline.” “Cheapest way to do keyword research.” If your name is in the prompt, the model will talk about you, and you have learned nothing except that it can read your prompt.
The source list
Read the sources, not just the answer. Your domain can be in the citation list while your brand is missing from the text. That is a different problem with a different fix, and it is covered in cited but not recommended. Check which of the two you actually have before you plan any work.
More than once
Run it three times, across two engines. AI answers are not deterministic. The brand list changes between identical runs, so a single absence proves very little. Three absences across a fortnight is a real signal.
Worth knowing what you cannot check. Google’s Search Console added generative AI reporting in June 2026, and the report gives impressions without clicks, so it cannot tell you whether an answer named you. Analytics will not tell you either, because a mention with no click leaves no trace at all.
The five reasons a brand is missing
No citable sentence
Nothing on your site says what you are in one quotable line. Models lift sentences. If your homepage says “the operating system for modern teams” and nowhere on the site says “a project management tool for agencies of five to fifty people”, there is no sentence to lift. Write the plain one. Put it above the fold and again on a page that answers a real question.
No comparison content
You have never written about the alternatives. Category comparison pages are the single most retrieved format in this whole channel, because a question like “best X” is answered by a page shaped like a comparison. If you have never published one, you are absent from the exact page type that gets read.
Thin category coverage
One page about a topic is not a topic. Assistants build a picture of a brand from repeated, consistent signals across pages. A single “features” page is a mention. Eight pages covering the questions your buyers ask is a position. This is the part that takes months of work rather than an afternoon.
Age and links
Nobody credible has written about you. Retrieval leans on the same authority signals search does. Ahrefs’ analysis arguing that GEO is mostly just SEO is right about this part. If you are new and unlinked, you are competing against pages that have been referenced for years, and no amount of on-page work closes that in a week.
Blocked crawlers
Your robots.txt is turning them away. This is the rarest cause and the easiest to check, so check it anyway. OpenAI documents its crawler user agents and a lot of sites blocked them in a panic during 2025 and never revisited the decision. Blocking a crawler does not protect anything if the same content is quoted from somebody else’s page about you.
What a zero actually means
A zero on its own is a dead end. A zero next to a leaderboard is a plan.
Say you sell a CRM. The check comes back with no mentions for you and this alongside it.
| Brand | Share of answers on “crm software” |
|---|---|
| competitor-a.com | 46% |
| competitor-b.com | 27% |
| competitor-c.com | 9% |
| you | 0% |
That is no longer a score. It is a reading list. Whoever holds 46% is being read by the model constantly, and the pages doing it are findable. Open them. Look at what format they are, how plainly they state what the product is, and which question each one answers. You are not copying the pages. You are reading the answer key.
The same goes for the citation half. If the check tells you which URLs feed answers in your category, and none of them are yours, you know the gap is content rather than technical. If some of them are yours, the gap is how you are named on your own pages instead.
A 30-day plan
Nothing exotic here. This is the sequence that moves the number for a small site.
Week 1. Write the plain sentence and put it everywhere. Category, who it is for, one differentiator. Homepage, about page, and the top of your main product page. Same sentence, not three variants. Consistency across pages is itself a signal.
Week 2. Publish one honest comparison. Your category, the real alternatives, and your own product on the list with the same depth as everyone else, limitation included. This one page does more than the next five combined, because it is the format the question resolves to.
Week 3. Answer the five questions your buyers ask before they buy. Pricing, migration, the objection you hear most, the integration everyone asks about, and the “is it right for me” question. One page each, question-shaped headings, short paragraphs a model can lift whole. Our guide to how to rank in ChatGPT goes through the mechanics of that in more detail.
Week 4. Fix the machine-readable half. One H1 per page, FAQPage schema where you have real questions, titles that describe rather than tease, and a page that renders without JavaScript. None of this gets you named on its own. All of it makes the pages above easier to read. An llms.txt file is cheap to add here too, with the honest caveat that adoption is still limited.
Then measure the same questions again. Same wording, same engines, three runs. Compare to what you wrote down in week one. If nothing moved, it is almost always because the comparison page is still missing or still leaves you off its own list.
ChatGPT is not the only engine, and they disagree
Worth checking more than one, because being absent everywhere and being absent from one engine are different problems.
ChatGPT and Google AI answers can be counted at scale. There is enough of an answer log behind both that a tool can measure how often a brand appears across many answers, month over month. That is where a trend line comes from.
Perplexity, Claude and Gemini cannot. None of them publish an answer log, so nobody indexes them. Every tool claiming a monthly Claude mention count is modeling it. What you can genuinely do is ask each one a specific question and read the reply, which is a real check on a real engine, just one question at a time.
They give different answers to the same question. In our own testing on a single category question, one engine cited our site and named a rival, another named a rival and did not cite us at all, and a third had never heard of us. Same question, same day. If you only ever check ChatGPT you are optimizing against one quarter of the picture.
Start with the two you can measure. Get counted mentions moving on ChatGPT and Google AI, because the work that moves those is the same work that moves the rest. Then spot-check the other three on your top five questions once a month. Our post on how to rank in ChatGPT covers the ChatGPT half in detail, and the same content work carries over.
What will not work
Asking the model to remember you. There is no submission form and no index you can add yourself to. Every “get listed in ChatGPT” service is either publishing content, building links, or selling you nothing.
Keyword stuffing your brand name. Repetition is not a retrieval signal. It makes the page worse to read, which makes it worse to quote.
Buying a dashboard and waiting. The category’s own users complain that these tools benchmark and then nothing changes. A weekly score with no page attached to it is a subscription, not a plan. The number is only useful if it points at a file you can open.
Treating one run as the truth. The variance between identical runs is larger than most of the changes you are considering. Three runs, two engines, then decide.
Where to start
Open your category question in ChatGPT right now, with no brand name in it, and write down every company in the answer. That list is your real competitive set for the next two years, and it is probably not the list in your pitch deck.
Then write the comparison page that puts you on it, properly, with a limitation you are slightly uncomfortable publishing.
If you want the measurement to keep running rather than becoming a thing you did once, AI visibility tracking inside ContextBolt SEO gives you twelve months of history on the first run, the leaderboard, the pages AI cites and the questions with their full answers. It runs from Claude, Cursor or Codex for $35 a month alongside keyword, SERP and backlink data, so the report and the rewrite happen in the same chat. There is also a free AI visibility checker if you want a number before you commit to anything.