Google routes its own citations inward as agent models get cheaper

AI Overviews now appear in 39.4% of US desktop searches

Why it matters for your score

Comscore's Q2 2026 AI Intelligence Report puts AI Overviews on 39.4% of US desktop searches in June, up from 25.8% in July 2025, with Bing's Copilot Search at 17.3%. On 4 searches in 10, the first thing a buyer reads about your category is a summary you did not write. Your job is to be quoted inside it.

Google is testing AI Overview links that lead back into AI Mode

Why it matters for your score

Links at the bottom of an AI Overview look like ordinary citations but open a follow-up question in AI Mode instead of a website. If Google keeps the user inside its own answer layer, the click you used to measure stops existing and being named in the answer becomes the only visible win.

Google documents hard thresholds for store ratings

Why it matters for your score

Merchant Center now states the bar plainly: roughly 100 eligible unique reviews, an average of at least 3.5 out of 5 for paid text and Shopping ads, a rolling 24-month window, and a separate rating for every country. These are pass or fail conditions on whether a rating shows at all, and AI shopping surfaces read the same merchant data.

OpenAI cut API prices by half with GPT-6 Sol and Luna

Why it matters for your score

Sol drops from $4 to $2 per million input tokens and Luna from $0.20 to $0.10, with 90% discounts on cached reads. Cheaper reasoning means agents can afford to read more of your site, compare more competitors, and do it more often. Your crawl budget problem is about to become someone else's research budget.

Anthropic shipped Claude Opus 5.5 at 40% lower cost to run

Why it matters for your score

Opus 5.5 leads on agentic coding and computer use benchmarks while costing 40% less than Opus 5 on typical workloads, and it is live on AWS, Google Cloud and Azure. Computer use is the part that matters for you: models that operate interfaces will try to complete tasks on your site, not just read it.

Spammers got roughly 12,000 Gemini Notebook pages indexed

Why it matters for your score

Public Gemini Notebook pages are indexable, and about 12,000 spam pages are now sitting in Google's index with no cleanup yet. Every new AI surface ships with an open door, and whoever walks through it first gets to be the answer in your category while you wait for the platform to notice.

Google keeps shortening the path from one of its answers to the next one.

Each new layer is another place where your name either appears or does not.

Agent inference got cheaper twice in a single day.

Falling token prices show up on your side as more crawls, deeper comparisons, and more pages read by something that will never click.

The conditions that decide whether you appear are increasingly written down.

Review counts, star floors and per-country segmentation are thresholds, not ranking factors, and you either clear them or you are absent.

Parasite content reaches new AI surfaces faster than the cleanup does.

Owning the plain answer to your own category question is the only defence that does not depend on a platform noticing.