51% of B2B software buyers now start their research in AI, not Google. Only 4% of readers click through. And 97% of llms.txt files never get read.

Edition 7, and mostly on schedule.

GEO is Generative Engine Optimisation, the unglamorous name for getting your brand mentioned by AI when someone asks it a question. It also goes by AI visibility, AI search, and half a dozen other things. Label doesn’t matter much. What matters is that the AI answer is increasingly the thing your buyers see first, and often the only thing they see.

Reading time: about 6 minutes. If this is useful, pass it on to someone who’d get something from it. And tell me what’s off, missing, or you want more of. Just DM me. Edition 6 is here if you missed it.

Russ

---

4%. That’s the share of readers who click through after getting news from an AI chatbot.

The Reuters Institute published its 2026 Digital News Report on 16 June, with the launch event on 18 June. 27 markets, roughly 100,000 people surveyed. The single number I keep coming back to: only 4% of readers who get news from AI chatbots say they always or often click through to the original source. That is against 19% for search engines and 17% for social. Around 10% of global audiences now say they get news from AI chatbots at all, and that share is the fastest-growing part of the whole report.

My take. This is the fact that reshapes the brief. You are not chasing referral traffic from AI, you won’t win that fight, the click has gone. The question you’re now trying to answer is what the AI is actually saying about your brand to the nine and a bit out of ten readers who never leave the answer. That is your shop window now. Most brands still have no idea what’s in theirs. If you sell to businesses and you’re worried your inbound is quietly softening this year, this is a big part of why. It also explains, cleanly, why publishers have stopped moaning and started organising: their traffic has left the building, and they know exactly where it went.

Sources: Reuters Institute Digital News Report 2026 · Press Gazette.

51%. That’s how many B2B software buyers now start their research in an AI chatbot more often than in Google.

G2 published its 2026 AI Search Insight Report a couple of weeks ago. 1,076 B2B decision-makers, global sample, fieldwork in March. The headline number: 51% of buyers now start software research in an AI chatbot more often than in Google, 71% use AI chatbots somewhere in the process (up from around 60% seven months earlier), and 93% say AI chatbots have fundamentally changed how they research.

It gets sharper on the impact. 8 in 10 buyers say AI chatbots accelerated their purchasing decision. 69% chose a different vendor than they’d expected because of how the AI framed the options. 85% think more highly of a brand that gets cited by AI. 33% actively bought from a vendor they had never heard of before the AI mentioned it. Trust-signal-wise, the number one confidence signal for buyers reading an AI answer is a citation from a third-party review site.

My take. This is the Reuters number, translated into B2B software buying. Buyers used to Google their way to a shortlist. Now half of them prompt their way to one, and two thirds of them end up choosing a vendor from that AI-generated shortlist rather than the list they walked in with. If you sell B2B and you cannot answer the question “what does ChatGPT say to my buyer when they ask about my category”, you do not have a marketing gap, you have an air gap between where buyers are and where you’re showing up. The review-site finding is worth a second read too: it is the same lesson we keep landing on, which is that other people’s content, especially third-party validation, is what tips the answer your way.

Source: G2 2026 AI Search Insight Report.

The UK’s regulator wave keeps building. A “Stealth Crawler Prohibition” bill is next.

A UK MP is drafting a Stealth Crawler Prohibition bill, modelled on a New York state bill from earlier this year. It would require AI bots scraping websites to declare who they are and what they intend to do with what they take. It has UK news industry backing. The bill is deliberately narrow: it does not regulate AI models, it regulates how the bots identify themselves. The New York debate referenced a roughly $1 billion grey market in unlicensed content trading, which gives you a sense of the scale of what’s currently invisible.

It sits inside a stack of adjacent moves in June alone. 31 UK websites started using county court contracts to set per-article fees of around £500 for unauthorised scraping. The CMA forced Google to give sites AI Overviews opt-outs without punishing them in normal search. Cloudflare switched to default-blocking unauthorised AI crawlers. Common Crawl has taken legal warnings. The Data (Use and Access) Act got Royal Assent on 19 June, and the ICO is now updating its AI guidance under it. Publishers, in the space of a few weeks, have moved from complaining to organising, contracting, and now legislating.

My take. Publishers have stopped arguing about whether unpaid AI access is theft and started building the toolkit to bill for it, prove it, and prosecute it. The filter tightens each month. Coverage in publishers with strong licensing positions will keep flowing into AI answers. Coverage in publishers who are still letting everything through unpaid, or whose content is now sitting behind default-block Cloudflare rules, quietly won’t. If your PR programme is not thinking about which outlets you target on those grounds, you are leaving a lot of visibility on the table. This is fixable, and it is exactly the sort of thing a decent measurement programme should be pointing you at.

Source: Press Gazette.

97%. That’s the share of llms.txt files that AI crawlers never read.

Ahrefs analysed 137,000 domains on 15 June. Of the valid llms.txt files they found, 97% got zero fetch requests from AI bots in May 2026. GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Google-Extended almost never touch the file. They crawl HTML directly. Adoption of llms.txt is up roughly 5x in a year. The crawlers are not reading it. The day after Ahrefs published, Google separately confirmed that AI-focused site files, if they are crawled at all, get no special treatment in Search or AI results.

My take. A lot of brands have been sold llms.txt as an AI visibility fix this year. The data now says, cleanly, it is not one. That does not mean you should not have one. It means it will not, on its own, change what AI says about you. The things that actually shift the answer are the same things that already shifted it: coverage in the sources the AI already trusts, content structured in ways it can extract, a body of third-party validation. This is the sort of thing a proper measurement programme should be able to tell you specifically for your category, not from a blog post about the average of 137,000 sites.

Sources: Ahrefs · Search Engine Roundtable.

At Known & Cited

Heads down on a proper product rebuild here. Nothing dramatic on the surface, and nothing that changes what current clients get in the near term. Under the bonnet, we are rebuilding the K&C products so they line up cleanly with the standards that matter in this space, the ones that are quickly becoming the reference points for anyone doing serious AI visibility work.

If you’re a current client, the short version: things are going to get better and more comprehensive over the next few weeks. Nothing changes on price, cadence, or how we work together. Just what lands in the report, and how usefully it reads.

Due to go live in the next week or two. I’ll say more once it’s out. If you are on the fence about starting, and would prefer to see the new version before we talk, drop me a line and I’ll flag you the moment it’s live.

NOT AI

Quick note before the Grand Prix bit. A piece I wrote for Mediaplanet’s Health Awareness (the supplement that runs with the Guardian) went live on 26 June. It’s on how I’ve been using AI as a stage 4 cancer patient. It is the FC:AI side of what I do, and it is a slightly more personal read than most of what lands in here. Have a look if it’s your kind of thing.

Right. On to the actual not-AI bit.

I am off to the British Grand Prix this weekend. It also happens to be my birthday, so I have quietly wrapped the two things together and declared it a long weekend I did not have to justify to anyone.

Four nights away. Four nights of the kids not being my problem, which sits somewhere between “guiltily excited” and “actually excited, thanks”. I have also convinced myself that this is a responsible use of a birthday, on the basis that dads are supposed to have interests, and I have decided motor racing is now mine.

I fully expect to come home lighter of pocket, having spent most of it on Lewis Hamilton merchandise I do not need and cannot really justify. A cap, obviously. Possibly a scarf, in July, in the sun. There is no world in which I return without a lanyard. Cause, effect.

Be Known. Be Cited.

See you next time.