One site that switched off Google's AI answers lost ordinary clicks too, and Bing is reportedly putting ads inside its own AI answers.

A screenshot shared on X shows "Sponsored Results within AI Overviews" on Bing, the Copilot-powered answer box Bing renamed AI Overviews a few months ago. Schwartz says he does not recall seeing ads in Bing's AI answers before and could not trigger one himself. Microsoft has not commented. Treat it as a single sighting until it is confirmed.

Gemini began adding `utm_source=gemini` to cited links in early September, and the tags disappeared around 28 September, according to a chart Natzir shared. Desktop Gemini still passes a gemini.google.com referrer, so that traffic stays visible. Gemini's iOS app opens links in its own browser and sends no referrer, so those visits land in GA4 as Direct, with only a "GeminiiOS" user agent in server logs. Google has made no statement.

Asked why more bots pretend to be Googlebot, Illyes said: "I think it is AI." Scrapers gathering training data claim to be Googlebot to get past blocks; his honeypots saw requests calling themselves Googlebot from unrelated IP addresses. He points site owners to reverse DNS and Google's published IP ranges to verify a real crawler, calls robots.txt a suggestion rather than an enforcement tool, and says firewalls and CDNs have got better at spotting fakes.

ChatGPT's newest entries are higher usage limits for Pro (29 October) and texting in beta for US Pro users (28 October), nothing on search or citations. Google's changelog newest entry is still the 26 October spam guide, and the status dashboard shows no October incident. Perplexity's newest changelog entry adds an Agent API effort level, with no current Sonar model deprecated. Anthropic's newest post is still 28 October (AI for Science). Bing's newest webmaster post is still 13 October, and Cloudflare's newest is a Q3 DDoS report.

Method in one clause: one site, opted out of AI Overviews and AI Mode with Google's Search generative AI control, comparing Search Console impressions and clicks with the site in and out of AI features; no dates, query count or site type given. AI impressions fell to zero after the opt-out and had been about 8% of total impressions. Shepard estimates the AI features' click-through rate at about 0.9%, worked out from totals rather than measured directly. His surprise: "regular organic clicks ALSO DROPPED when AI features were disabled." On the pages with the most AI visibility, average position fell during the opt-out and recovered afterwards.

Shepard calls it a "small case study" and adds "Results not guaranteed. Always test your own data." Schwartz calls it "a super limited study based on a small sample size" and notes that AI Overviews still appeared on the results pages throughout; only this site was removed from them. Shepard's explanation, which he calls a "Search SERP Halo Effect" (showing up in more places makes a click anywhere more likely), is a hypothesis.

Method in one clause: 2,240 answers from ChatGPT Instant and Meta's Muse Spark with web search on, 14 consumer-electronics shopping prompts, seven synthetic user memory profiles plus one empty control, 10 runs each, collected on 19 October 2026. Repeat answers for the same user overlapped about 42% on named products for ChatGPT and 18% for Muse Spark; between a user with relevant memory and one with none, overlap was roughly half that, for both models. Relevant memory showed up in 19% of ChatGPT's search queries and in 88% of its answers. ChatGPT's number of options and cited domains barely moved (2.5 against 2.7 domains per answer), while Muse Spark narrowed to fewer options and fewer sources.

Profound sells an AI visibility tracker; the users are synthetic, the products are electronics, and it is one collection date. It does not say whether the runs were through the app or the API.