Thursday’s stories share a diagnostic problem: a referral log, a Search Console report and a click count all show less than what actually happened. A missing signal is not proof of a missing effect, whether that is a chat that satisfied a reader before a click, an edge rule that blocked a crawler, or a citation moved one menu deeper.
The Answer That Leaves No Trace: A Referral Share Too Small To Prove Anything
A Scrunch study puts AI referrers at 1.1 percent of publisher visits, and that number is exactly what substitution would look like, not evidence against it.
- AI Referral Share Is Tiny. That Isn’t Proof Publishers Are Safe. A Scrunch study puts AI referrers at 1.1 percent of publisher visits in the week after a news related AI conversation, against roughly 75 percent direct and about 9 percent search. A referral log cannot register an answer that satisfied the reader before a click ever happened, so a low share is what substitution looks like, not evidence against it. The study is a vendor’s own analysis, not independently replicated. AI Overviews also showed up disproportionately on queries that already had lower click rates: once the query itself was held constant, the click gap narrowed to about two points.
Same Blind Spot, Two Instruments: A Crawl Ratio Built In, A Block Search Console Can’t See
One dashboard now does by default what a site owner used to reconstruct from raw logs, while two documented cases show a single Cloudflare setting can erase Googlebot without leaving a mark anywhere Google shows you.
- Microsoft Clarity Adds A Scrape-to-Referral Ratio Card To Bot Analytics Microsoft added an AI Scrape-to-Referral Ratio card to Clarity’s Bot Analytics on August 13, ranking AI operators by how much they crawl against how much traffic they send back. It is the same scrape-to-referral arithmetic one site owner worked out from raw server logs, covered here August 10, now shipped as a built-in report. Microsoft’s own illustration shows a ratio of 6,000 to 1 against 41 total referrals. Microsoft runs its own crawler and its own AI answer surface, so it is not a neutral party in what the card measures.
- Two Documented Cases Show Cloudflare Settings Blocking Googlebot Two named practitioners, Jonathan Bird and Brodie Clark, documented sites losing rankings and traffic after Cloudflare bot and crawl-control settings blocked Googlebot. It is the follow-up to our August 5 report that Cloudflare’s AI bot toggle might be blocking Googlebot: a suspected failure mode now has documented cases attached, a general crawl-control setting in one and firewall-level bot rules in the other. Nothing in Google Search Console named an edge rule as the cause in either case.
The Manual Control Retires: Google Ads Hands Language Matching To Its Own Signals
From late September, a campaign-level language setting disappears from Search campaigns, and the ad creative, the landing page and Google’s own signals decide instead.
- Google Ads Retires the Language Setting, Leans on Signals Starting in late September, Google Ads removes campaign-level language targeting from Search campaigns and from the Search inventory inside Performance Max, moving language matching onto the ad creative, the landing page and Google’s own signals about the searcher. It is another manual control retired in favor of automation, and Google has not announced new reporting to show which language actually served.
Today’s Quick Hits
- ChatGPT Buries Its Sources Link Behind an Extra Click Glenn Gabe spotted ChatGPT moving its Sources button into a More actions overflow menu on some replies. One practitioner’s observation of what may be a test; OpenAI has not commented.
- Google Cuts Follower Bar For Search Profiles, Not On TikTok Google lowered the threshold to claim a Google Search Profile from 100,000 followers to 35,000 on YouTube, Instagram and X. TikTok stays at 100,000, confirmed by Google’s Robby Stein.
- Spammy Canonical In Search Console Traces To Domain History John Mueller said a spammy Google-selected canonical on a new, unindexed site can reflect what the domain previously served, and advised waiting a few weeks.