GA4 now has a native AI Assistant channel, and it undercounts. Here is how to set up the reporting properly, why a large share of assistant clicks land in Direct no matter what you configure, and how to sanity-check the real number.
The stat doing the rounds — that most AI-cited pages carry structured data — is real and proves much less than it is used to prove. What schema demonstrably does, what the evidence cannot support, and why it is still worth shipping.
From 15 September 2026 Cloudflare blocks 'mixed-use' AI crawlers by default on ad-bearing pages — for new customers, new sites, and every free plan. What that changes, who it hits, and the settings to check before the date.
One is a summary above the blue links; the other is a separate conversational surface that returns no blue links at all. They pick sources differently, they overlap less than anyone expected, and Google is now merging them.
A book is four jobs, not one: plan it, draft it, finish it, then make it good enough to send. Most novel writing software solves the second and leaves you alone for the rest. Here are the ten tools worth paying for in September 2026, ranked by how much of the book they get across the line.
Since June 2026 there is a real Search Console control that removes a site from AI Overviews and AI Mode without touching its rankings. Here is exactly what it covers, what it costs, and why the decision is impossible to test.
GEO is sold as a new discipline with new deliverables. The controlled research points somewhere less exciting: specific, verifiable, well-referenced writing. Here's what holds up and what doesn't.
llms.txt is pitched as the robots.txt of the AI era. Google says it ignores the file, and a 137,000-site study found 97% of them are never requested. Here's what the evidence shows and when it is still worth shipping.
AI Overviews are the AI-written summaries at the top of Google results. Here's the retrieval process behind them, what the click-through data actually shows, and the controls you really have.
AI bots split into three jobs — training, answer indexing, and fetching a page a user asked about — and blocking the wrong one costs you citations. A reference table and a robots.txt policy you can defend.
The gap between a mediocre AI answer and a great one is usually the prompt, not the model. Here's a practical framework — task, context, examples, format — with a before-and-after you can copy.