Blocking GPTBot did not cost sites their name in ChatGPT

What did we ask ChatGPT?

At the end of October 2026 we asked ChatGPT ordinary Japanese questions, the kind a person asks when looking for a service in some industry. No question named a site. Then we checked whether ChatGPT mentioned the site we were watching. Half of those sites block GPTBot; the other half do not. Both halves were named almost exactly as often.

Why do people worry?

GPTBot is the bot OpenAI sends to read web pages for training its AI. Many sites block it in robots.txt (the public file that tells bots what they may read) because they do not want their text used as training data. But ChatGPT is OpenAI's too. So the worry is natural: if I block the training bot, will ChatGPT stop naming me when a customer asks?

How did we test it?

We matched 49 pairs of large Japanese sites. In each pair, one site blocks GPTBot and one does not, in the same industry and of a similar size. We asked ChatGPT about each site's industry, with a free account, from Japan. Before asking, we wrote down what would count as a gap.

1. Pick a pairTwo large .jp sites, same industry, similar size: one blocks GPTBot, one does not
2. Ask ChatGPTAn ordinary Japanese question about the industry, twice per site, no site name
3. Check the answerIs the site named? Is there a link to it?
4. CompareBlocking sites against their open partners, overall and pair by pair

Source: our test, 49 pairs of large .jp sites, ChatGPT free account from Japan, 28 to 31 October 2026. What counts as a gap was written down on 28 October, before the first question.

The result: no gap

ChatGPT named the sites that block GPTBot in 94% of its answers. The open sites: also 94%.

Named: sites blocking GPTBot		90 of 96
Named: open partner sites		91 of 97
Linked: sites blocking GPTBot		87 of 96
Linked: open partner sites		91 of 97

Source: our test, 49 pairs of large .jp sites, ChatGPT free account from Japan, 28 to 31 October 2026. 193 scored answers. The small gap in links comes from two pairs whose company site sits on another domain. Both groups are near the maximum; this does not show that blocking is harmless.

Pair by pair, the picture is the same. In 4 of every 5 pairs the two sites tied. The blocking site lost in exactly as many pairs as it won.

Tied		39 pairs
Blocking site named less		5 pairs
Blocking site named more		5 pairs

Source: our test, 49 pairs of large .jp sites, ChatGPT free account from Japan, 28 to 31 October 2026. Whether the site was named, compared inside each of the 49 pairs.

What surprised us most

Some sites block both GPTBot and OAI-SearchBot. OpenAI says OAI-SearchBot is the bot that decides whether a site appears in ChatGPT search. We expected this group to be left out. Yet ChatGPT named them almost as often as their open partners, in about 9 of 10 answers.

Only GPTBot blocked		52 of 54
Their open partners		52 of 54
GPTBot and OAI-SearchBot blocked		38 of 42
Their open partners		39 of 43

Source: our test, 49 pairs of large .jp sites, ChatGPT free account from Japan, 28 to 31 October 2026. 27 pairs block only GPTBot, 22 pairs block GPTBot and OAI-SearchBot. Small groups: a difference of one or two answers would not mean much.

Two bots, two decisions

OpenAI's documentation keeps the two bots apart. GPTBot is for training models. OAI-SearchBot is for showing sites in ChatGPT's search results. Each setting is independent: you can block GPTBot and still let OAI-SearchBot in. Our result fits that reading.

• GPTBot

  What it does

  Collects pages to make OpenAI models more useful and safe (training)

  If you block it

  You ask OpenAI not to train on your pages

• OAI-SearchBot

  What it does

  Shows websites in ChatGPT search results

  If you block it

  Not shown in ChatGPT search answers, though the site can still appear as navigational links

Source: OpenAI bot documentation, developers.openai.com/api/docs/bots, read 13 November 2026. Our test covered blocking GPTBot; it did not measure the documented effect of blocking OAI-SearchBot directly.

How many Japanese sites block them?

In our scan of large .jp sites on 2 November 2026, about 5% blocked GPTBot but let OAI-SearchBot in. About as many blocked both. Very few did the opposite.

Block GPTBot, allow OAI-SearchBot		38 of 719
Block both		38 of 719
Block only OAI-SearchBot		fewer than 5

Source: our weekly scan of large .jp sites (Tranco top list), 2 November 2026, 719 sites with a readable robots.txt. Counts below 5 are hidden.

So what should you do?

If you block GPTBot because you do not want your text used for AI training, this test found no sign that it costs you your place in ChatGPT. Make two separate decisions:

1. GPTBot: decide on training grounds. Do not block it, or unblock it, out of fear for your place in ChatGPT.
2. OAI-SearchBot: if you want to appear in ChatGPT search, let it in.
3. See what your robots.txt tells each AI bot with the free check on pw (our research site): https://pw.migurei.ai/en/free-scan

What we do not know yet

There is one sign we do not count as a result. Sites that block both bots were still named, but they showed up less often as a source card (the small link card ChatGPT attaches to an answer) than their open partners. That fits OpenAI's note that such sites can still appear as navigational links. We only noticed it after looking at the data, so it needs its own test before we can say more.

This post is written by Parallel World (migurei.ai) about measurements from our own lab. The free check above is our own service.

The full numbers, dates and how we counted: ChatGPT search and GPTBot: 49 pairs of .jp sites (https://pw.migurei.ai/en/research/chatgpt-search-gptbot-blocked-vs-open-49-pairs)