Perplexity now says old Sonar requests are being quietly re-routed to its new API, so the Perplexity column can change without a single error.

The page used to read "Sonar will be supported until September 27, 2026." It now reads, verbatim: "Sonar Chat Completions support ended on September 27, 2026. Synchronous and streaming requests keep working: they are being reformulated as Agent API requests, rolling out gradually by model." Asynchronous Sonar requests "are not reformulated and are no longer supported." The page does not say which Agent API model or preset a reformulated request runs on, when each model switches, or whether the response marks it. A first-hand call at 11:52 UTC today, model `sonar` on the engine's key, returned HTTP 200, `"model":"sonar"` and 20 citations. Nothing in it says whether it was re-routed. The engine's `main` has moved to `06233ce` (28 September) but `src/adapters.ts` still names `sonar` on line 14 and calls chat completions on line 91. The adapter makes a normal synchronous call, so it is in the "keeps working" group.

Stein, verbatim: "Our Ultra and Pro subs have been using info monitoring capabilities in AI Mode in Search, and now we're rolling it out to everyone globally." A user tells AI Mode what to watch and Search keeps checking "sites, forums and social posts, our real-time data sources, and Shopping Graph." Google's examples are new restaurants nearby, local holiday activities and price drops. There is no Google blog post, and Google has not said how it picks sources or whether service businesses appear.

Still one post, the 24 September launch note. No completion message.

The brief quotes an OpenAI engineer in February 2023: "no matter how prominently we show the links, users won't click." It also says Nick Turley, head of ChatGPT, described "no good reason to click" once ChatGPT had answered. Microsoft data in the brief compares Bing Chat with Bing web search: click-through 87 to 93% lower for The Times sites, 83 to 91% for the Daily News plaintiffs, 51 to 94% for Ziff Davis. SEJ notes the Microsoft figures carry no collection dates, query volumes or method. This is a plaintiff's brief, not a finding; a ruling is not expected until 2027. The primary filing was not read.

Verbatim: "Half of the content Answer Engines cite is less than 13 weeks old." The post gives scoring thresholds for a page (live 28 days or more, at least 5 citations) but no sample size, date range or engine list behind the headline number. Profound sells the tool that measures it.