Apple has now written down that its AI opt-out cannot cost you search ranking, eight days before Cloudflare's new default makes it easy to block the wrong thing entirely.

Apple's Applebot help page now carries the line, verbatim: "Site rules for Applebot-Extended are not considered in ranking for Search." It sits alongside two sentences that were already there, verbatim: "Applebot-Extended does not crawl webpages. Webpages that disallow Applebot-Extended can still be included in search results", and a description of what the opt-out actually covers, verbatim: "Web publishers can choose to opt out of their website content being used to train Apple's general purpose foundation models powering generative AI features."

This is one added sentence, not a policy change. What makes it worth carrying is the shape of the mistake it rules out. Applebot-Extended is a directive-only user agent that never fetches anything, so a rule aimed at it is a training opt-out and nothing else. Applebot, without the suffix, is the crawler that feeds Spotlight, Siri and Safari search. The two get conflated constantly, and Apple has now put its own sentence behind the distinction.

Crawling, indexing and serving are all clear, with the newest incident still the 18 August spam update. The Google Search Central blog still stops at the 28 August site reputation policy update. Cloudflare has published nothing on AI crawlers since 3 September. Search Engine Land's newest article is still dated 4 September. Perplexity's two September posts are about on-device routing and serving infrastructure, with nothing touching citation or retrieval mechanics. Anthropic published nothing on Claude web search or citations. Microsoft and Bing have nothing after the AI Performance report expansion already carried. Of Search Engine Roundtable's six items this morning, four are advertising or shopping stories and one is a widget note below.

Older material, never carried here, and the largest properly documented measurement of what an AI Overview citation is worth in clicks. Method stated on the page: 53 brands, 5,471,127 distinct queries, 2.43 billion organic impressions and 296.9 million paid impressions, January 2025 through February 2026 as actuals with March 2026 projected, restricted to search terms with complete monthly data across the whole period.

The figures, for informational queries, per million impressions: no AI Overview present, roughly 33,500 clicks; present and cited inside the AI Overview, roughly 20,743; present on the page and not cited, roughly 9,445. Seer's own summary line, verbatim: "Being cited in the AIO delivers +120% more organic clicks per impression versus when you are not cited." Two directions in one dataset, and both matter: a citation roughly doubles clicks against not having one, and it still lands about 38% below what the same query returned with no AI Overview at all.

Their stated limits, in their words, are that they "cannot claim causation" and cannot separate Google's behaviour from a brand's own SEO performance without account-level investigation, and that their linear model "underestimates in recovery environments." A third limit is not stated and should be: AI Overview status was assigned from the most recent snapshot and applied backwards across a fourteen-month window, so queries that only recently gained an AI Overview are labelled as having had one throughout.

The page states, verbatim: "Only 30% of brands stay visible from one answer to the next, and just 20% remain present across five consecutive runs." It attributes the work to AirOps and Kevin Indig and gives no sample size, no prompt count, no date range and no list of engines. Three further figures on the same page have the same problem, all verbatim: "About 85% of brand mentions originate from third-party pages rather than owned domains", "About 48% of citations come from community platforms like Reddit and YouTube", and "Roughly 60% of AI Overview citations come from URLs not ranking in the top 20 organic results."

The persistence claim matters more than the other three, because it is the one measurement nobody credible has published. The Wellows study carried on 5 September covered 1,146,483 questions across five engines and stated plainly that each question was collected once per engine, so run-to-run variance was explicitly outside its scope. seoClarity's citation volatility work measures whether sources were cited at all, not whether a named business persists.