Research
Original analysis and data on AI search visibility. What moves the needle, and what's just noise.
Every piece is anchored in proprietary data, a documented test, or a timely, evidence-backed point of view. No hot takes.
Researched, written and published by Ronnie Huss, founder of SearchScore.
-
AI crawler policy, re-measured after a parser fix (October 2026)
Of 120,190 websites whose robots.txt has a rule that reaches AI crawlers, GPTBot was blocked on 10.39% and OAI-SearchBot on 5.48%. Post-fix audits only, from our own scanning campaigns, so not a random sample, and not a trend against the August study.
-
19,998 Sites Block GPTBot While Allowing OpenAI Search
Across 347,988 websites audited in one 90-day window, GPTBot was blocked on 13.50% while OAI-SearchBot was blocked on 7.83%. Of GPTBot blockers, 42.6% allow OpenAI's search crawler; only 270 sites show the reverse configuration.
-
Google is testing a slot for the official site in AI answers. Your site has to say why it is you.
Google is testing a section at the top of AI Mode and AI Overviews responses for the "one or two official places on the web" to complete a task, with context on why each link is useful. What Google said, what it has not said, and what a small business site should state plainly.
-
Four Teams Counted the AI Searches in Search Console. The Answers Are 0.34%, 2.32% and 18.9%.
93,448 Search Console rows across two properties, read against the three other published studies. Everyone finds the same query families and the same zero clicks, and the headline shares are fifty-five times apart. Borrowing three detectors from those papers raised our own figure by half in an afternoon, which is the point: the share is the instrument, not the web.
-
ChatGPT Is Now Europe’s Second-Largest Designated Search Engine.
The European Commission designated ChatGPT a Very Large Online Search Engine. On the Commission’s own register its 159.1 million monthly EU users sit above Bing’s 119 million. The researcher-access clause matters more than the label.
-
Most Sites Blocking AI Training Crawlers Are Blocking the Search Ones Too.
Training and search are separate crawlers with separate rules. 56.2% of sites blocking GPTBot also block OAI-SearchBot, removing themselves from ChatGPT’s answers. Fewer than one in 50,000 block Claude’s search crawler at all.
-
Google Documents an Up-Rank in Top Stories and a Badge in AI Answers. Those Are Not the Same Promise.
Preferred sources uses different wording for different surfaces: content is “more likely to appear” in Top Stories, and “can be highlighted with a ‘preferred’ badge” in AI Mode and AI Overviews. Read from the documentation rather than the commentary.
-
We Crawled 4,000 llms.txt Files. Four in Five Pass Chrome’s New Test, and Two in Five Were Never Files at All.
Of files that genuinely exist, 79.4% satisfy the Lighthouse definition, and the failures are almost all bare URLs. The larger finding is that 41.8% of sites that appear to have one answer that path with a web page, so adoption is being over-counted.
-
Chrome Now Grades Your Site for AI Agents. Three of Its Six Checks Are for a Standard Almost Nobody Uses.
Lighthouse 13.3 added an Agentic Browsing category. Half of it is WebMCP, every agent check returns Not Applicable, and the category as shipped cannot cost any site a point. Read from a run rather than the announcement.
-
176 of the 178 AI-Ready Websites Publish Organisation Schema. The Tag Is Worth 1.45 Points.
A pre-registered result on 102,873 domains. Sites with Organisation schema score 15.83 higher on the entity sub-score, but the tag itself accounts for 8% of the gap. Why it is a marker of a well-built site rather than a lever, and the one category where its advantage nearly disappears.
-
We Asked Whether llms.txt Works. It Is Worth 1.37 Points.
Sites with an llms.txt score 5.36 points higher and a quarter of that is the scorer counting the file. 22% of the least ready websites on the index already publish one, against 5.6% for Organisation schema. A pre-registered result, published as it came out.
-
One AI Engine Shows Its Sources Ten Times More Than the Rest
Across 10,201 source URLs from six AI engines, 73.8% came from Perplexity alone. ChatGPT returned 4.2%, Gemini 3.8%. Why that is a disclosure gap rather than a research gap, and the redirect bug that made Gemini look like a single publisher.
-
The Ceiling: A Million Websites Scored, and Nobody Has Cleared It
216 of 1,000,000+ websites were AI-Ready and none were Verified. The highest score in the Q3 2026 census was 86.6. Only 13.5% of websites block GPTBot, but 55.7% publish no structured data at all (corrected 18 August 2026).
-
ChatGPT Started Linking Out. The Biggest Winner Was Your Homepage.
Referral traffic nearly doubled after the May 2026 links update, but homepages grew 1,316% while blog sessions fell 28%. Why AI search rewards being nameable rather than comprehensive, and why the headline number is the wrong one to quote.
-
Microsoft's AI Shopping Guide, Translated for Service Businesses
Microsoft told retailers how AI picks what to recommend. Here's the part that applies if you don't sell products.
-
The JavaScript Trap: Why AI Cannot See Most UK Law Firms
We scored 9,487 UK law firm websites. Just three are AI-Ready. Firms built as JavaScript apps are 76 times less likely to be visible, and almost half of every invisible firm in Britain is one of them.
-
Britain's Best-Rated Care Homes Are the Ones AI Cannot Find
We scored 2,307 UK care homes for AI search visibility. Not one is AI-Ready, and a CQC Outstanding rating barely moves the needle. Why the badge families trust is invisible to the tools they now ask.
-
Google Search Console Now Tracks Your Social Profiles: What Changed
New platform properties let you measure Instagram, TikTok, X and YouTube performance in Google Search. We separate the three launches everyone is conflating, and explain what to actually do.
-
Almost Nothing Is Ready for AI Search. We Scored 1,000,000 Sites to Be Sure.
From the whole web to a single profession, the AI-Ready rate lands between 0% and 0.2% at every scale. The cross-edition SAVI benchmark, and why the gap is universal.
-
One Question, A Dozen Searches: What Google's Query Fan-Out Means for Your Business
Google researches the whole question, not one keyword, and that changes how you need to cover a buying decision.
-
Google Shipped Another AI File Format (OKF). Why We're Not Rushing.
The third "feed your site to AI" format in eighteen months. Well designed, explicitly v0.1, and not read by any consumer AI engine today. Here's what OKF actually solves, and why the fundamentals haven't moved.