A draft field experiment found AI Mode assignment cut outside clicks and trust in Search, a vendor study logged ChatGPT naming fewer B2B brands than it did in March, spam turned up inside Google’s own NotebookLM domain, and job board operators say Indexing API approvals have gone quiet.
Routed, Not Chosen: AI Mode Absorbs the Click
A field experiment and a new anchor-text test both point the same way, AI Mode keeping the click for itself instead of sending it to a publisher.
- Draft Study: Users Assigned AI Mode Clicked Out Less and Trusted It Less A Penn and Northeastern field experiment enrolled 1,444 Google users and randomly assigned them to AI Mode, no-AI, and normal-search groups and found the AI Mode group clicked outside links less, trusted Search less, and was more likely to say it intended to switch to Bing. The paper is still a draft, and its control arm, where AI features were stripped out entirely, found no evidence that doing so hurt anyone’s experience.
- Google tests AI Overview links that open AI Mode, not pages Google is testing an AI Overview anchor text that opens AI Mode with the original query carried forward instead of sending the click to a publisher’s page, a test search watcher Gagan Ghotra documented on X. It follows a separate test placing an AI Mode button directly inside the results-page search bar.
Borrowed Trust, Thinning Shortlist: Cracks in What AI Names and Cites
One report finds spam hosted on Google’s own AI tool, the other finds ChatGPT naming fewer brands than it did six months ago.
- Parasite SEO Finds a Host Inside Google’s Own NotebookLM Public Gemini Notebook reports, built on Google’s own notebook.google.com domain, are turning up in Google’s index promoting coupon codes, crypto refund offers, and paid social-engagement services. One practitioner’s claim that his report became an AI Overview citation within 24 hours could not be independently reproduced by Search Engine Watch.
- BrightEdge: ChatGPT Names Fewer B2B Tech Brands Per Answer BrightEdge ran a fixed panel of non-branded B2B tech prompts through ChatGPT on the same dates in March and September and found the average count of distinct brands named per answer fell 27.9 percent, with the steepest drop at the informational, top-of-funnel stage. The study covers ChatGPT alone and does not measure whether Google’s AI Overviews or AI Mode show the same pattern.
A Renamed File, a Silent Form: Technical Signals Nobody Can Verify
A new Lighthouse audit checks a manifest path the spec it validates has already replaced, and job boards say Google has gone quiet on an approval form.
- Lighthouse 13.5 Adds an Experimental Audit for AI Agent Discovery Lighthouse 13.5 shipped an experimental audit that checks whether AI agents can find a site’s callable services, but the audit still searches for the older ai-catalog.json filename even though the Agentic Resource Discovery spec it validates moved to ard.json on August 26. Google’s release notes never connect the audit to Search rankings, and it carries no signal about traffic or visibility.
- Job Boards Say Google Stopped Responding On Indexing API Access SEO consultants tracking data from more than 100 job boards say they suspect Google hasn’t approved any expanded Indexing API access requests in 2026, though they stop short of confirming the number is zero. Search Engine Journal put direct questions to Google about the delays and had not received a response by publication.
Quick Hits
- Microsoft Advertising tests Copilot Audience Generation A closed beta in the US, Canada, UK, and Australia lets advertisers describe a target customer in plain language and get AI-built audience segments back, though the tool works only within Audience campaigns and can take up to 72 hours to activate.
- Merchant Center drops impressum rule for German, Austrian local listings Merchants in Austria and Germany no longer need an impressum URL to run local inventory ads or appear in free local listings, while Switzerland keeps the requirement across all local ad formats.
- xAI moves X Search API to per-post billing with Grok 4.7 Alongside the Grok 4.7 launch, xAI replaced its per-call pricing for the X Search API with charges of $5 per 1,000 posts fetched and $10 per 1,000 profiles fetched, a shift that hits high-volume monitoring tools hardest.