Why ChatGPT never cites my site

Updated on

You ask ChatGPT a question your site answers. It cites three competitors and never you. Your content is better, more current, better written. You have no idea what decides this invisible ranking, or what you could possibly change about it from where you are.

ChatGPT does not send one crawler but several, and they do different jobs. The one deciding whether your site appears in an answer is not the one used to train models. Many sites block the wrong one, believe they are refusing training, and make themselves invisible in answers.

The misunderstanding that costs the most

When a ChatGPT answer cites sources, people assume it drew on “what it learned”. That is wrong, and the mistake leads to a far more annoying one.

Answering by citing pages and learning from pages are two separate operations, run by two different crawlers, with two independent settings.

OpenAI documents this plainly. GPTBot is used to train the models. OAI-SearchBot is used to surface sites in ChatGPT’s search results. And the publisher spells out what happens if you turn the second one away: sites opted out of it are not shown in ChatGPT search answers.

There is a third case, often mistaken for the other two: ChatGPT-User, which visits a page because a human just asked a question. That one does not crawl the web on its own.

How so many sites switched themselves off

Over the past few years, plenty of sites decided they did not want their content used to train AI models. A perfectly legitimate decision.

Except a great many applied it by blocking anything that looked like an AI crawler, without separating the roles. The result does not match the intent: the content trains nothing, and the site appears nowhere either.

That is the exact opposite of what someone selling online wants. You want to be cited. Refusing training and appearing in answers are two distinct switches — and they are set separately, precisely for that reason.

Being on Google does not put you in ChatGPT

That is the second surprise, and it is structural.

They are two separate circuits with their own access conditions. To appear in ChatGPT’s answers, its search crawler has to be able to read your site, and your host or delivery service must not block the addresses it works from.

That last point is treacherous, because it is not decided inside your site. Anti-bot protection switched on at your host can refuse access without anything in your pages showing it.

What your answer needs to be quotable

Then there is the most common case of all: your site can be read, and it still never gets cited.

An AI answer cites what it can extract cleanly. It does not read your page like a visitor who wanders, hesitates, backtracks and eventually gets it. It looks for a statement that stands up on its own.

A text that answers its own question early and clearly can be picked up. A text that sets the mood for three paragraphs, spreads the information across the whole page and never quite concludes gives nothing to quote — however good, however right.

It is also why a figure with no dated source rarely makes it through: it cannot be verified, so it does not get reused.

The three free checks

Your robots.txt file. Add /robots.txt to your address and read it. It is public, it runs to a few lines, and it says exactly who you allow. That is where the confusion between the two crawlers usually sits.

Your page without its styling. If the substance of your point only appears after the visitor does something — a tab to open, a “read more” button, a form to fill in — that is not reading, it is an obstacle.

The question, asked directly. Go into ChatGPT and ask the question your page answers. Look at who gets cited. It is not necessarily the big sites: it is the ones whose answer was extractable.

What not to do

Block every AI crawler “just in case”. That is the move producing exactly the outcome you were trying to avoid. The settings are separate so that you never have to choose between the two.

Repeat your keyword through the whole page. It stopped working on classic search engines, and it actively loses you visibility with the engines that answer using an AI.

Pay to “get listed in the AIs”. Nobody sells access to ChatGPT’s answers. What exists is crawlers not to block and pages that can be read — and both are free.

Frequently asked questions

Does blocking GPTBot keep me out of ChatGPT?
No. GPTBot is about training models. The one deciding whether you appear in search answers is a separate crawler, and the two settings are independent of each other.
Is ranking well on Google enough?
No. They are two separate circuits with their own access conditions. A site ranking superbly on Google may never be consulted by ChatGPT if it turns away its search crawler.
Can I refuse training and stay visible?
Yes, that is exactly why the crawlers are separate. Each setting is independent, so you can allow one and refuse the other.
How do I know whether my site can be read?
Your site's robots.txt file, visible by adding /robots.txt to your address, lists what you allow and refuse. It is the first place to look, and it is public.

Sources

  1. OpenAI documentation — overview of OpenAI crawlers — accessed
  2. OpenAI Help Center — ChatGPT search — accessed
  3. OpenAI Help Center — publishers and developers FAQ — accessed