Why your brand is missing from ChatGPT answers

It is almost never that your product is worse. It is usually that the model has nothing to go on.

By Rohit Kumar, Founder · Updated 2026-10-03

Here are the five real reasons, in the order worth checking. The first two take an hour between them and are embarrassingly common.

1. You are blocking the crawlers

The crawlers that feed answer engines are not Googlebot. GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended are separate agents, and plenty of sites block them — sometimes deliberately years ago, sometimes because a plugin or a security product did it quietly.

Open yourdomain.com/robots.txt and read it. If those names appear under a Disallow, that is your answer and nothing else you do will matter until it changes.

While you are there: if robots.txt or sitemap.xml returns a 404, fix that too. It is an hour of work.

2. Your site does not say what it is in machine-readable terms

A model reading your homepage has to infer what you sell, who you serve, and where you operate from marketing copy. Structured data says it outright, without inference.

Check whether your pages contain any JSON-LD at all. Many sites have none. Adding an Organization block and an FAQ block is a small, well-defined job, and it costs nothing but an afternoon.

3. Everything written about you was written by you

This is the big one, and the one that takes real time.

Answer engines favour brands that other people have described. Reviews, directory entries, comparison articles, a mention in someone's roundup, a thread where a customer explained what you do. These are what a model can lean on without taking your word for it.

In our own thirty-run testing, the agencies that appeared consistently had a visible footprint of exactly this kind. We want to be careful here: that is a correlation we observed, not a proven cause. We did not run a controlled experiment adding citations and measuring the change, so we will not claim closing that gap guarantees anything. What we can say is that the pattern is consistent and that the brands with no external footprint were not the ones being named.

4. Your pages are shaped for keywords, not for answers

A model looking for something to quote wants a clear claim it can lift. The classic SEO article — circling its keyword, deferring the answer, padded for length — offers nothing extractable.

Rewriting the shape costs you nothing in ranking terms: lead with the answer, make headings the questions people actually ask, then expand underneath.

5. You are checking wrong

Worth ruling out before you spend money on the other four.

• You checked once. The answers vary between identical runs — in our testing, in four of five questions. One check is a sample of one.
• You were signed in. Personalisation is real. Results reflect your account.
• You asked a definitional question. Some phrasings reliably return a definition with no companies named at all. That is a property of the question, not a verdict on you.
• You asked about yourself. Asking "what is [your brand]" proves only that the model can read your site. Ask the question a stranger would ask.

The order to work in

1. Read your robots.txt. Today.
2. Add structured data and a sitemap. This week.
3. Re-shape your pages around real questions. This month.
4. Earn external mentions. Ongoing, and the one that actually moves the number.

Then re-run the same thirty questions and compare against your baseline. If nothing moved, you have learned that the first three were not your bottleneck — and that points you squarely at the fourth.