Tuesday’s thread was control over the destination. Google tested routing Shopping clicks straight to merchant sites, a rare move that sends traffic back to the open web, while the AI layer kept re-engineering where sources sit: Microsoft turned Bing into a grounding layer for agents, and Google tried relocating AI Mode citations below the answer.
A Click Points Back at the Open Web
A rare test sends Shopping traffic outward instead of holding it inside Google’s interface.
- Google Tests Direct-to-Retailer Links in Shopping Results Google is testing product links that drop shoppers straight onto the retailer’s page instead of the in-Google overlay panel. If it ships, the merchant gets the session, the analytics, the retargeting pixel, and the full path to checkout back. It also reverses years of Google keeping the click inside its own walls, which makes product page speed the first impression again rather than something the overlay quietly absorbed.
The Grounding Layer: Where AI Decides What to Cite
Two signals that source attribution is being re-engineered at the passage level, not the page.
-
Microsoft Web IQ Routes AI Agents Through Bing at the Passage Level Microsoft’s new grounding API feeds AI agents discrete passages from the Bing index instead of whole pages. The optimization target it implies is not a URL, it is a paragraph: a page can rank in position three and still own no passage worth citing, while a page that never ranks can hold the one block an agent quotes repeatedly. The numbers are Microsoft’s own, but the shift in what counts as visibility is the story.
-
Google Tests Moving Source Attribution Below AI Mode Answer Cards Google is testing a layout that pushes the source count and favicons below the citation cards in AI Mode rather than above them. In a zero-click result, attribution that sits after the answer earns less eye contact than attribution above it. One layout test is weak signal, but small changes to where a source sits compound across billions of queries, so the teams already segmenting their AI Mode referrals will see the effect first.
Sharper Tools for Paid Search
A pre-flight confidence check arrives for the experiments most accounts run badly.
- Google Ads Adds Experiment Power Score to Flag Weak A/B Tests Campaign Guidance now scores an experiment’s likelihood of reaching statistical significance before it launches, banding it low, medium, or high across Performance Max and Broad Match tests. It is a useful fix for the most common testing mistake, an underpowered run treated as a verdict. The catch Google underplays: a high score is bought with a 50/50 split and a longer runtime, so more budget is committed before the result can be trusted.
Today’s Quick Hits
- Google’s Mueller: Hyphenated Domains Carry No SEO Penalty John Mueller confirmed on Bluesky that hyphens in a domain carry no ranking penalty, and that the practical ceiling is apparently 61 of them. The spammy reputation came from the low quality of the sites that leaned on keyword domains, not the dash itself. Choose a domain on brand and usability, not on a penalty that does not exist.