BlogPage 2

Search Console

Google Just Admitted Its Own AI Report in Search Console Is Not Good Enough — 4 Data Traps to Stop Falling For

A Reddit breakdown of Search Console's generative AI performance report got a one-line confirmation from John Mueller: 'This is pretty much it.' Impressions that nobody saw, links behind the Show More expander that go uncounted, an average position that belongs to the AI Overview block rather than your link, and a filtered view you can't add to your regular report. Here's how to actually read the numbers — plus 5 things independent site owners should do now.

SEO News

This Week in SEO: Google Rewired Europe, ChatGPT Shopping Swapped Engines, And The Entry Ticket Just Changed Hands

Google handed comparison platforms and direct suppliers their own result units in Europe. ChatGPT's product picks flipped from 8% feed-sourced to 62% in a single day. Business Profile post views came back after three and a half years. Three unrelated stories pointing at one shift: whether you show up is becoming less about where you rank and more about whether you hold an entry ticket.

AI Overviews

How Much Traffic Did AI Overviews Actually Take From Wikipedia? The UW Paper Cut Its Own Estimate From 15% to 5% — Google Still Disputes It, But the Real Story Is Elsewhere

Same authors, same dataset, and the estimate shrank from 15% to 5%. A University of Washington working paper used Wikipedia's multilingual structure as a natural experiment: AI Overviews reduced English Wikipedia's search referrals by 5.45% versus German and 4.82% versus French — roughly 100 million visits a month. But the number independent site owners should care about is the one hidden inside the average: Culture articles fell 19.6%, STEM only 7.4%.

Query Fan-Out

Someone Was Doing Query Fan-Out by Hand Back in 2003: Why an Old "Russian Nesting Dolls" Keyword Trick Came Back to Life in AI Search

Google's own docs confirm that AI Overviews and AI Mode use query fan-out to split one question into many sub-queries. One dataset study showed 189 branded prompts fanning out into 1,797 sub-queries — about seven words each — with quoted exact-phrase checks climbing 25-fold from the first search in a run to the last. Your page has to hold both the short phrasing and the long phrasing. That discipline is 20 years old. It just used to be called "nesting dolls."

Google

Google DeepMind Wants One Model to Just Generate the Ranking: What Autoregressive Ranking Changes for Independent Sites

Google DeepMind, UMass Amherst and UT Austin have proposed Autoregressive Ranking (ARR) — an LLM that generates document IDs token by token and produces the ranked list itself, replacing the two-stage dual-encoder + cross-encoder pipeline. The paper includes a capacity proof (dual encoders need embedding dimensions that grow linearly with the corpus), a new SToICaL training loss, WordNet and ESCI results — and one metric that got worse.

AI Search

Google Just Split Search Intent Into 5 Verbs: Explore, Decide, Learn, Create and Do — Your Content Strategy Needs a Rewrite

Now that AI Mode has passed one billion monthly users, Google has published its first new intent taxonomy. Two fresh studies — the first empirical analysis of AI Overviews on Spanish publishers plus an eye-tracking experiment — reveal which queries trigger AI summaries, why attention and clicks have decoupled, and why 43.6% of SERPs are still a blue ocean for classic rankings.

Google SEO

Google Lost Faith in Your Site? John Mueller on Old Low-Value Pages — and Why Deleting Them Won't Bring You Back

John Mueller said something unusually blunt on Bluesky: programmatic SEO page farms can leave Google's systems having 'possibly lost faith' in your entire site — based on the old pages. That's a site-level trust problem, not a page-level one. Cleanup only stops the bleeding; proving real value is what recovery actually takes. Here's a line-by-line breakdown of Mueller's reply, where Google's scaled content abuse policy actually draws the line, and a practical cleanup playbook for site owners.

AI search

Why AI Gets Your Brand Wrong: It's Usually Not Hallucination — Old Titles, Retired Product Names and Stale PDFs Are Poisoning AI Answers

SEJ columnist Carolyn Shelby argues the biggest hidden brand risk in AI search is too much conflicting information, not too little: old CEO titles, discontinued products and outdated PDFs keep beating your current pages in retrieval. Here's how LLMs end up favoring stale vocabulary, how to write bridge content that connects old terms to new reality, and how to run a brand claim audit.

AI search

Harvard Ranked 940 Occupations: The Public Barely Objects to AI Taking SEO Jobs (2.31/7, Second From Bottom) — Competence Is the Only Moat Left

A Harvard Business School survey of 2,357 Americans across 940 occupations found the public's moral objection to automating search marketing is just 2.31 on a 1-7 scale — second from the bottom, above only file clerks. Support for full automation jumps from 30% of occupations to 58% when AI is described as stronger and cheaper, while 94% welcome AI as an augmentation tool. Combined with the HBS loan-approval experiment (human preference flips with beliefs about who is better) and a 791-person P&G study (AI-assisted teams 3x more likely to produce top-10% ideas), here's what actually protects SEO and content careers — and three practical survival strategies.

AI Shopping

AI Shopping Orders Are Up 15x — But Is Your Checkout Ready for a Machine Buyer?

Shopify's president says orders from AI search have grown 15x since January 2025. Google's UCP, OpenAI's Agentic Commerce Protocol (co-built with Stripe), and Salesforce are all live — with Etsy, Glossier, Spanx, and Vuori already selling inside ChatGPT. Yet QAwerk's real-world testing shows most stores' checkout, product data, and refund flows were never built for an AI agent placing orders at machine speed. Getting seen by AI is only half the game; getting bought by AI is the part nobody is testing. A full breakdown of agentic commerce today, plus a three-point self-audit checklist.