When AI can't read you, it makes you up
Three findings from this summer all point the same way. In an AI-mediated market, your reputation is your retrieval strategy.
The short version: If you want to get found on AI search, stop tinkering with your website and start earning your reputation. AI tools describe your business from whatever the wider web already says about you, so the founders who get cited are the ones other credible sources talk about. The optimisation tricks sold as 'generative engine optimisation' don't reliably work, and some make things worse.
AI agents hit a wall on a third of B2B sites, and when they can’t read yours, 58% of what they say about you comes from third parties, against 12% when they can. Siteline, 2026.
Only 28% of people trust AI-generated answers, against 70% for traditional search. YouGov, 2026.
A review of 45 studies found no GEO technique reliably improves visibility, and some cut a page’s AI retrieval by up to 16%. arXiv, 2026.
What happens when an AI can't read your website?
A few weeks ago a group of researchers did something obvious that nobody had bothered to do. They pointed AI agents at a hundred B2B software companies and asked them to do what a buyer now does forty times a day: go and find out about this company, what it costs, what it is good at. Then they watched what happened.
In roughly a third of the sessions, the agent hit a wall. A page it couldn't read, or wasn't allowed to. But that didn't stop it... It didn't come back and say 'sorry, I couldn't find out.' It went off and asked around, and 58% of what it then reported about those companies came from third-party sources, against just 12% when the site behaved itself.
So when the machine can't read you, it doesn't stop talking. It makes you up, out of whatever the rest of the internet happens to be saying: the review site, the old forum thread, the directory entry someone filled in badly in 2019 and never touched again.
Do people even trust what AI tells them?
Now add a second finding, from YouGov this month. Only 28% of people say they trust the answers an AI assistant gives them. 70% still trust a traditional search engine. People have worked out how to use these tools at frightening speed, they're asking them everything, and then not believing a word of the answer without a second opinion.
That's not a reason to ignore AI. It means the machine's say-so, on its own, closes nothing. What moves a sceptical buyer is who the machine cites. The named source. The credible third party. The founder whose view already carries weight in the room. Trust doesn't come from the tool. It transfers from the reputation the tool is pointing at.
Does generative engine optimisation (GEO) actually work?
We all know what a certain kind of marketer will say to all this. Great, so we optimise for the machines. There's a whole gold rush selling founders 'generative engine optimisation', the promise that a handful of clever tweaks will get you cited.
So I went and read the evidence, because someone has to. A survey published this month pulled together forty-five separate studies on it. Its finding is blunt: no technique reliably improves your visibility across AI platforms, and some of the popular rewrites actually reduced a page’s chances of being retrieved, by as much as 16%. If you sell this stuff, that ought to sting. Founders are paying to be seen less.
This is not a surprise to anyone who has spent time in reputation. There is no schema hack that substitutes for being genuinely worth citing. The model is trained on what the world already says about you. If the world says little, or says it badly, no clever rewrite of your homepage changes the underlying material. You have to go and earn the sources.
If the world says little, or says it badly, no clever rewrite of your homepage changes the underlying material. You have to go and earn the sources.
So what actually gets you found by AI?
Here's where all three findings meet, and it's an old idea wearing a new coat.
For twenty years I've watched reputation do the same stubborn thing. It forms in the loudest, worst moments and then it loiters for years, long after the facts have moved on, whether it belongs to a country, a company or a founder. It's the story other people hand you before you've bothered to look for yourself. What's changed this summer is that the story is now machine-readable and repeated instantly, to a buyer you'll never meet, by a system they half-trust and constantly double-check.
The retrieval strategy is the reputation strategy. They're the same job now. And it breaks, usefully, into things you can actually do.
Make yourself readable, first, and I mean that literally. If an agent can't get through your own front door, you've handed the narration to strangers.
Then, and this is the real work, make sure the story waiting out there is one you'd be happy to be described by. Get cited by sources a wary buyer already believes.
Put the founder's actual thinking into the world under their own name, because Edelman and LinkedIn find that 75% of decision-makers have gone off to research a company on the back of exactly that.
Be the kind of business other people want to be caught recommending.
That combination, the readable site, the earned third-party credibility and the founder’s own voice, is what It’s a Shovel calls the Founder Visibility Stack. None of it is fast. None of it is a hack. That’s the point. The moat was always reputation. AI has simply automated the part where the market decides who gets mentioned, and handed the summary to a machine that will fill any silence you leave with other people’s words.
The question that actually matters isn't whether your site is optimised. It's if the internet described you tomorrow, without you in the room, would you recognise the company it described? Would you like it?
That's the work. It always was.
Jessica Whitcutt is the founder of It’s a Shovel, a reputation and visibility consultancy. She is MCIPR and spent her career in reputation, including a former FTSE Top 20 role, helping founder-led businesses get found and trusted across search, AI and the channels where B2B decisions are actually made. More about Jessica.
Frequently asked questions
What is generative engine optimisation (GEO)?
It's the practice of trying to make your content more likely to be cited by AI search tools and assistants. The trouble, per a 2026 review of 45 studies, is that no GEO technique reliably works, and some rewrites actually make a page harder for AI to retrieve.
How do AI tools decide what to say about my company?
Largely from what already exists about you across the web, not just your own site. When an AI agent can't read your site, a 2026 benchmark found 58% of what it says about you comes from third-party sources. Your reputation elsewhere is doing the talking.
How do I get my business found and cited on AI search?
Make your own site genuinely readable to agents, then earn credible third-party mentions and publish the founder's own thinking under their name. AI repeats the story the wider web already tells, so the work is making that story accurate, credible and worth repeating.
Is traditional SEO still worth doing with AI search?
Yes. Most people still trust traditional search far more than AI answers, 70% against 28% in YouGov's 2026 data, and a readable, well-structured site is the foundation both search and AI draw on.
How long does building AI visibility take?
Longer than any hack promises, which is exactly why it holds. Reputation accrues slowly, one credible mention at a time, and that durability is the point.