Three of today’s stories describe the same thing from different angles: work that gets used without being named, counted, or linked.
The Citation Tap Closed: Publisher Links Are Going Missing On Two Fronts
One is a model swap Google says it will fix. The other has no explanation at all.
- AI Mode’s New Gemini Model Is Dropping Source Links Testers found Gemini 3.8 Flash answering top-of-funnel queries in AI Mode with no citations whatsoever, and Google has confirmed the behaviour is a bug rather than a design decision. The distinction matters less than the timing: any publisher reading a drop in AI-surface referrals this week now has two candidate causes it cannot separate, the model swap and its own pages.
- Google News Tab Is Dropping Indexed Stories, Then Restoring Them Articles ranking normally in web search are falling out of the News tab and returning hours later, with no pattern identified and no comment from Google. For a publisher whose traffic peaks in the first hours after publication, an intermittent absence during exactly that window costs more than a permanent one that can at least be diagnosed.
Priced By The Model: What AI Mode Actually Shows Shoppers
The generative shopping layer is not a reskin of the product carousel. New data says it ranks on different terms, and Google is finally giving merchants a way to watch.
- AI Mode’s Lead Product Price Runs 21.6% Above Standard Search Productrise measured more than two million listings and found that when the same product appears on both surfaces, AI Mode carries the higher price nearly seven times out of ten. The finding points at ranking rather than pricing: the cheapest offer is a strong signal in the traditional carousel and apparently not in AI Mode, which changes which merchants are worth optimising for which surface.
- Merchant Center’s AI Shopping Report Adds Four Countries Australia, Canada, India, and New Zealand can now see how their catalogues surface inside conversational shopping queries, months after the feature landed in the United States. It arrives in the same week as data suggesting AI Mode ranks products on unfamiliar terms, which makes the report less of a nice-to-have and more of the only instrument merchants have on that surface.
Rewriting The Rules Underneath: Remedies, Audiences, And Measurement
A courtroom, an ads platform, and a measurement vendor all moved on the same question: who gets to set the terms of the auction.
- Google Keeps Its Ad Exchange, Loses Its Auction Shortcuts A federal judge declined to order a breakup after finding monopoly power, choosing instead to strip out the bidding advantages that built Google’s position in publisher ad tech. Behavioural remedies leave the asset intact and depend on sustained supervision, which is a slower lever than divestiture in a market that reorganises faster than a compliance schedule.
- OpenAI Lets Small Audiences Exclude, Not Just Target, In ChatGPT Ads Dropping the minimum audience size for exclusion lists opens ChatGPT advertising to smaller brands that could never assemble a list large enough to qualify. Suppression is usually the first thing a disciplined advertiser wants and the last thing a young ad platform ships, so its arrival is a reasonable marker of where the product is in its build-out.
- NIQ Pairs With Similarweb To Measure AI Shopping, Q4 Debut A retail measurement giant is buying its way into AI-referred commerce data, with availability planned for the fourth quarter. When the incumbent panel vendors start selling AI-referral metrics, the question at the next budget review stops being whether to track generative visibility and starts being which number to defend.
Today’s Quick Hits
- Google’s Ad Tag Gateway Leaves Beta On Akamai, Fastly First-party tagging through the CDN edge is now generally available, giving advertisers a durable measurement path as third-party signal keeps eroding.
- Google Ads Clarifies Phone Call And Offline Conversion Rules Refreshed documentation pins down click-time attribution for calls and the 90 day click ID window for offline uploads, two settings that quietly distort reporting when they are wrong.