OpenAI joins Google and Microsoft in honouring one shared "no AI training" line, and it still leaves ChatGPT search on its own switch.

OpenAI says it plans to honour the same robots.txt line for AI training preferences that Google and Microsoft announced on 30 October, and that it "aims to honor the signal in the next few months". Its current robots.txt rules stay in force until then. SEJ notes that none of the three companies supports the line yet, and that neither OpenAI's post nor its crawler documentation says what happens when the shared line and a GPTBot rule disagree. OpenAI's crawler page (read first-hand today) does not yet mention the shared token at all. It still says GPTBot covers training and OAI-SearchBot decides whether a site appears in ChatGPT search answers.

Method: Sistrix asks ChatGPT a random sample of questions daily while logged out, 15 September to 28 October 2026, in Germany, Austria, Switzerland, the UK, the US and Spain. Sample size is not stated.

- The new answer layout switched on overnight: none at 2pm on 8 October, almost all from 3am on 9 October, 97 to 100% from 23 October. - Answers with a source list at the end fell from about 72% to 1 to 19% between 9 and 22 October, then came back to 70 to 84% from 23 October. Sources moved into tiles beside the paragraphs: "whoever was cited remained visible, just in a different place." - Answers with at least one citation in the text rose from 72.4% to 76.2%. - Linked pages per answer rose from 3.3 to 4.1, and distinct websites per answer from 2.3 to 2.9, in two steps around 9 and 22 October. Beus says the model change may not be the cause. - Links tagged utm_source=chatgpt.com fell to 1 to 24% from 9 to 15 October and were back at 99 to 100% from 17 October. This confirms, with Sistrix's own data, the fix first reported on 17 October. - Manufacturers and institutions gained 2 to 4 points of citation share; publishers lost about the same.

Method: 16,310 US desktop users matched across two Datos panels (Google clickstream, and an AI panel of Perplexity, Copilot and you.com), November 2024. ChatGPT is only 0.1% of the AI panel, and the panel skews 35 to 44 and male.

- Informational questions were 26.8% of AI prompts against 8.1% of Google searches. - AI prompts averaged 8.4 words against 4.2 on Google. - 8.0% of AI prompts revealed the asker's role, situation or purpose, against 1.2% on Google. - AI was 2.7% of searches among people who used both, and 96% of AI prompts had no matching Google search within 30 minutes.

The data is two years old and nearly ChatGPT-free, so it describes the shape of AI questions, not today's volume.