Google now says it is "critical" to fact-check every AI-written page by hand before publishing, and that covers titles, descriptions and schema too.

Google's changelog entry for 1 October says it "Updated the using generative AI content guide with information from the Search Quality Raters guidelines." The page now reads, verbatim: "It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing." It gives the reason: "generative AI outputs may contain inaccuracies (also known as hallucinations)." It extends the check past the body text: "This review also applies to metadata like `<title>` elements, meta description elements, structured data, and alternate texts for images, which can appear in Search results." And it repeats the scale warning: "Using generative AI tools or other similar tools to generate many pages without adding value for users may violate Google's spam policy on scaled content abuse." Search Engine Roundtable (Barry Schwartz, 1 October) notes the page was last changed in October 2025 and that "critical" is strong wording for Google.

The page says "one of the most critical factors for assessing page quality is the quality of the main content" and lists four attributes raters look at. Effort: "The extent to which human work went into creating the content or the systems powering it." Originality: "The extent to which the content offers unique, original information or perspectives that aren't already available on other websites." Accuracy: "For informational pages, the content should be factually accurate." Search Engine Roundtable reports these two sections as new. Google's changelog does not list this page, so "new" rests on Search Engine Roundtable's reading, not on a Google statement.

No change to any of the three. Perplexity still names no model for re-routed Sonar requests. Anthropic's table still shows `claude-sonnet-4-5-20250929` deprecated with retirement on 30 November 2026. Google's incident still carries only the 24 September launch post, and Search Engine Roundtable's monthly report (1 October) describes the update as "still rolling out". The engine's `main` is still `06233ce`, with `sonar` and `claude-sonnet-4-5` on lines 14 and 15 of `src/adapters.ts`.

Otterly ran a 60-day experiment across 4 subreddits it created: 240 posts, 1,485 subscribers, 30 hours of community work, 474 AI citations. It reports that on 8 August 2026 reddit.com's share of ChatGPT citations dropped 89%, from 4.01% to 0.44%, in two steps (8 and 14 August), while total ChatGPT citations rose 23%. Otterly says its own method used bots to add engagement, which breaks Reddit's rules. Promptwatch reported the same August drop at the time (3.8% to 0.5%); this is a second vendor's measurement of it, not a new event. Otterly sells tracking, and the unit is a cited page, not a named business.

Using Ahrefs Brand Radar, the study tracked the 215 finance pages ChatGPT cited most in October 2025. By September 2026, 50% received no citations and the group kept 14% of its citation total. Wikipedia and Reddit went from 18% of the top 500 citations to 0%, and bank and card issuer pages rose from 2% to 22%. The author states the limits: correlation only, one vendor's tracking, US finance content, and the measured factors explain 9% to 18% of the variation.