SEO News
This was the week the shortcuts died. Google’s Milan event told SEOs to stop chunking for AI, confirmed llms.txt and Markdown do nothing for rankings, and Ahrefs found 97% of llms.txt files are never read. The flip side: Bing shipped the first genuinely useful first-party AI visibility data, and the real lever is still crawlable, useful HTML.
📰 Search Engine Updates
Lead newsBing Webmaster Tools adds AI visibility insights: Intents, Topics, Citation Share, Compare
Bing is expanding Webmaster Tools with views into why you are cited, which intents and topics you show up for, your citation share, and how you compare with other sources. Why it matters: not full AI tracking, but one of the most useful first-party AI visibility layers yet. Pair it with Google’s new Search Console AI reports for a two-engine baseline.
Google on chunking, site signals, paywalls, subscriptions and AI clicks
At Search Central Live Milan, Google said forcing paragraph “chunks” for AI is pointless, there is no reward for HTML validation, and 15% of daily searches are completely net-new. It also pushed subscription linking via Reader Revenue Manager, citing a +34% engagement lift in internal tests. Why it matters: the takeaway is to skip AI shortcuts and invest in useful, unique, crawlable content backed by site-level quality.
Google Cloud announces the Open Knowledge Format
An open specification for organizing and exchanging knowledge that AI systems can use: datasets, metrics, APIs, tables and runbooks. It signals that structured, machine-legible knowledge is becoming its own layer beneath AI answers.
Google: llms.txt and Markdown files do not help or hurt rankings
Google clarified that Search does not use llms.txt or Markdown mirrors, so they neither help nor hurt. On the Search Off the Record podcast, Mueller and Splitt made the same point: engines already understand HTML, which carries context through links, layout and structure that Markdown strips out. Why it matters: make your HTML easy to extract instead of maintaining parallel AI files.
◆ Experiments of the Week
Two tests, one lessonDoes anything read your llms.txt?
97% never readAhrefs checked 137,210 domains: about 28% publish an llms.txt (an upper bound, since their users skew technical), but 97% of those files got zero requests in May. Of the few that were fetched, 96% were bots and only 19.5% were named AI tools like GPTBot and Claude-Code. They also flag llms.txt as a possible prompt-injection surface. Verdict: treat it as hygiene and future agent-readiness, not a visibility lever.
Do AI assistants render your JavaScript?
12 of 12: noAndre Alpar tested 12 AI assistants and found the major US assistants read raw HTML rather than rendered JavaScript when grounding answers. Verdict: server-side render or pre-render critical content, put key facts high in the HTML, and validate with server logs rather than trusting what a chatbot claims to see. It is the same lesson as the llms.txt result, AI consumes your raw markup.
🤖 AI Search Insights
Data and readsWhy calling yourself “the best” can help your competitors win in AI search
Ray’s research shows self-promotional “best [category]” listicles can get cited by Google’s AI surfaces while the brand itself is left out of the actual recommendation. Track more than citations: check whether the brand is represented, framed positively, and actually recommended.
What matters in an AI prompt: intent or keywords?
Summarizing Peec AI’s analysis, prompt wording often matters less than the underlying intent for brand visibility. Build representative prompt groups around intents, personas, markets and constraints instead of chasing phrase variants.
Zero-click searches: highest in the UK, lowest in Germany
Similarweb clickstream data shows meaningful market differences: the UK has the highest zero-click rate, Germany the lowest and the highest share ending in a click. Fishkin suggests the EU’s anti self-preferencing rules may be helping keep searchers clicking.
The 3-layer AI search diagnostic and measurement matrix
A free workbook mapping Presence, Readiness and Business Impact metrics so AI search reporting becomes actionable rather than a pile of isolated screenshots. A useful structure if your AI reporting still leans on one-off citation checks.
Headline formats and Google Discover: what 3.4 million articles reveal
Quote-led headlines outperform in some cases, but the effect swings by publisher, audience, topic, language and surface. The takeaway: test headline formats against your own Discover data, not against generic benchmarks.
There is no AI shortcut
Five separate signals landed on the same message: Google said stop chunking, confirmed llms.txt and Markdown do nothing for Search, Ahrefs showed almost nobody reads llms.txt, and a rendering test showed assistants read raw HTML. The work that moves AI visibility is the unglamorous kind: crawlable, useful, well-structured HTML and genuine authority. Spend your effort there, and measure presence and recommendation, not just citations.
📡 On the Radar
Quick hits- AI Mode Google’s information agents went live in AI Mode for AI Ultra subscribers (12 Jun), adding a background monitoring layer to Search. link
- Volatility Heavy ranking movement the week of 8 to 12 June that most tracking tools barely registered; read GSC, not just trackers. link
- Google Ads Conversion-based customer lists turned on for Enhanced Conversions and Customer Match accounts from 17 June. link
- Bing Testing new news-box designs and colour selectors in product carousels. link
- Industry “SEO under pressure”: Fiorelli and Foley Carter on fewer clicks and proving value, with a calm playbook rather than panic. link
- Local Competitor review analysis as a research goldmine for content and AI visibility gaps. link
🔨 Tools
BookmarkFree AI share-of-voice tracker across ChatGPT, Claude, Gemini, Perplexity and AI Overviews, including the sub-queries your prompts fan out into. Bring your own API keys.
See how much AI traffic any site is getting from generative engines. Vendor tool, so treat figures as directional.
- State of Search Q1 2026, real clickstream behaviour across the US, EU and UK Datos x Rand Fishkin
- Building a community-led content strategy, reaching conversations keyword research misses Erin Simmons / Athens SEO
📅 On the Calendar
Coming upLily Ray, Mark Williams-Cook, Aleyda Solis, Navah Hopkins, Lazarina Stoy, Areej AbuAli and more.
A relaxed SEO networking event with padel, drinks and food. Play or just watch.
💬 Social Buzz
Where it clusteredThe “GEO is just SEO, stop the shortcuts” message dominated after Milan and the Ahrefs llms.txt study, with Danny Sullivan widely quoted pushing back on chunking content for AI on the Search Off the Record podcast.
A more sober thread: SEO is under real pressure from fewer clicks and AI-mediated discovery, and the shared response is to evolve reporting toward visibility, trust and business impact rather than panic. read
Milan takeaways spread fast through threads from attendees: chunking is useless, there is no reward for HTML validation, and subscription linking drove a +34% engagement lift in Google’s internal tests. recap
Cover only three things?
Ahrefs llms.txt study plus Google’s stance: the “no AI shortcut” story.
Bing AI Visibility Insights: first-party AI data arrives.
Lily Ray: citations are not recommendations.
That is the week. The thread tying it together: the lazy AI tactics are getting debunked in public, and the winners are the people doing the boring, durable work.
Until next week,