Google and Bing will both honour one shared "no AI training" line in robots.txt from early 2027, and Bing is splitting training from AI answers.

Google says it intends to support a shared machine-readable training preference from the IETF AI preferences work "in early 2027". The token is `__usage_ai-training`, written as a user-agent line in robots.txt (two underscores at the start, one in the middle; SER prints it with two in the middle, but the IETF draft and Google's page use one). A site that disallows that token and does not name Google-Extended will be treated as if Google-Extended were disallowed. Google's page still says Google-Extended does not affect inclusion or ranking in Google Search; it does cover grounding in Gemini Apps. The IETF draft (Kevin Kelley, Anthropic, version 00, 30 October) is an individual draft with "no formal standing", and one participant quoted by SER says the work is "not actually done".

Bing will read the same shared training token in robots.txt, plus page-level tags: `<meta name="Bingbot" content="ai-training:disallow">` for training and `ai-grounding:disallow` for AI answers, with a narrower `ai-grounding:disallowextended` that still allows Microsoft's own Copilot but blocks third-party grounding services. There is no site-wide robots.txt line for grounding. The new tags replace the old NoArchive and NoCache directives for these purposes. Dates: 15 December temporary grounding settings open in Bing Webmaster Tools; 1 February 2027 Bing honours the new tags as it crawls; 1 March 2027 NoArchive and NoCache stop being read for AI, and a site that has set nothing defaults to Allow for grounding. Microsoft says classic Bing rankings are unaffected, and that either grounding opt-out "will see reduced brand visibility in AI experiences".

Reddit says search referrals fell by more than 1 million a week on average in October, and its working explanation is mainly Google's AI Overviews and AI Mode. CEO Steve Huffman said AI Mode "does not drive traffic". Reddit plans to test a logged-in-only experience in unnamed markets outside the US in Q4, with some version of the site still visible before a hard login and shared links still working. US logged-out daily users fell from 30.1 million to 27.3 million between Q2 and Q3.

Method: the same questions asked daily for 20 days (AI Mode 1,996 questions, ChatGPT 3,173, AI Overviews 563, Gemini 1,276, Copilot 823), tracking every site cited on day one, about 37,500 in all. Run dates beyond "one month" are not stated; vendor data, mostly marketing and software topics.

- Gone by day two: AI Overviews 50%, AI Mode 58%, ChatGPT 59%, Gemini 81%, Copilot 83%. - Still cited all 20 days: AI Overviews 6.6%, AI Mode 6.8%, ChatGPT 4.0%, Gemini 0.2%, Copilot 0.5%. - About 75% of citations that dropped out came back within the period; citations that held 8 days or more came back 96 to 99% of the time. - Gemini and Copilot gave no citations at all in 64% and 53% of answers. - "Most cited websites appeared on only one day", against about one in five brand names. A brand's own site stayed cited all 20 days in 28 to 33% of branded factual questions on Google's tools and ChatGPT. "Best X" questions churned most: 64 to 76% of cited pages changed daily on ChatGPT and AI Mode. - On AIO and AI Mode, pages ranking in Google held a citation for a week or more about seven times as often (29% vs 4%). Correlation.

This is the fourth stability study on file (Otterly, SISTRIX and seoClarity in September). It is the first with a per-engine split that includes Copilot, and the Copilot figure is the worst of the five.

5.2 million sessions classified as agentic across 230,511 Clarity projects, 13 August to 9 September 2026, compared with human sessions on the same sites. Agentic sessions reached checkout in 25.2% vs 14.7% for humans, form submission 19.9% vs 18.6%, appointment booking 8.4% (no human comparison given), and clicked out far less (14.3% vs 27.0%). Clarity warns sessions may include human handoffs and that events do not prove a task was completed. Agents are not named beyond "OpenAI, Google, and Anthropic ecosystems".