Google is still feeding altered results to automated searches, and the bug report at the service that collects our Google answers has been open since May, so the Google column needs a hand check before it is trusted.

This is new since our 12 and 14 October items on "poisoned" results, and it is a different fault. Nozzle judged 253 million result pages in 47 countries from 9 March to 21 October 2026. Since 16 September, 13.9% of pages ignored part of the query (0.06% before), peaking at 39.9% on 28 September; on 66.2% of them only one word was dropped, while the page title still showed the full query 99.8% of the time. AI Overview source panels on those pages kept 52.4% of their usual sites, against 72.0% normally. Ahrefs said on 22 October it sees "between 5% and 20% of search results affected" depending on the data provider and is now filtering and replacing bad results (https://ahrefs.com/blog/serp-data-update/). SerpApi's public bug "[Google Search API] Searches returning irrelevant/poisoned results" was opened 4 May 2026, is still open and labelled urgent, and its 14 October update says a fix to reduce it was deployed but "We're still working on fully addressing the issue." (https://github.com/serpapi/public-roadmap/issues/3781). Google has not commented.

ChatGPT's newest entry is still locked chats (20 October). Google's changelog adds only a 23 October note on hyphens versus underscores in URLs, and the status dashboard shows no October incident. Perplexity's changelog is unchanged and marks no current Sonar model as deprecated. Anthropic's newest posts are still 22 October; the deprecations table is unchanged (Sonnet 4.5 retires 30 November). Anthropic has also published a new API version, 2026-10-21, which removes the temperature, top_p and top_k settings and rejects citations sent inside user messages; it applies only when a request asks for it. Bing's webmaster blog still ends at 13 October, and Cloudflare's newest posts are not about crawlers.

Method in one clause: 48,527 pages cited by ChatGPT, AI Mode or AI Overviews in German-speaking markets in Q2 2026, each paired with an uncited page from the same domain (96,418 URLs, 17,380 domains, 43 features), conditional logistic regression with correction for multiple tests. Odds of citation went with a table of contents (1.7, though only 14% of pages have one), Article schema (1.6), genuine FAQPage schema (1.5), an author box with photo (1.4), and the current year in the title (2.7, which the authors call "an initial indication"). Well-optimised pages were cited 2.1 times as often as barely optimised ones. llms.txt showed no measurable effect (11.7% of cited URLs came from domains with the file against 14.0% overall), and readability scores did not matter. AI Overviews favoured Article schema (2.0) and disfavoured product pages (0.5); ChatGPT favoured question-style titles (1.9). The authors: "it shows correlations, not proven causes, and is a snapshot of the DACH region." Perplexity, Claude and Gemini chat were not tested.

Method in one clause: 14,000 API calls (10 industries, 7 question types, ChatGPT, Perplexity, Claude and Gemini, 50 runs per cell, 5 simulated personas), run dates not stated. Depending on the model, 45% to 72% of answers to educational questions named no brand, and about 20% of pricing answers named none. Comparison questions were the most stable ("The same brand leads the response 91% of the time."), and purchase questions the least, with 40% brand overlap. The industries are large sectors such as insurance, banking and cloud computing; no local, service or coaching businesses.

Fishkin's whiteboard summarises the three traffic studies the radar covered on 22 October and adds a Go Fish Digital case study that gives no numbers. His conclusion is new: "Traffic from AI assistants (ChatGPT, Claude, CoPilot, Deepseek, Perplexity, etc.) is nearly non-existent", yet AI is "probably, quite influential on some buyers in some sectors" and should be funded and measured as a brand channel, like a billboard. No new data.

On its own website, Brainlabs counted about 27,190 AI bot requests (13.15% of measured traffic) and an average of 90 bot requests per referral visit: OpenAI 34 to 1 (76 referrals), Google 117 to 1 (25), Anthropic 776 to 1 (3). Clarity's Topic Insights caps a topic at 15 prompts and does not write them. One site, an agency's own, a few weeks.