A business owner I spoke with recently had just paid an agency four figures for something called an "AI-readable version" of their website. A parallel set of pages, built in a plain-text format called markdown, sitting quietly behind the scenes so that ChatGPT and Google's AI could "read the business properly." The pitch was compelling. AI doesn't understand normal websites the way humans do, they were told, so you need a special translation layer just for the machines.
It's a good story. It isn't true, and Google said so again this month.
What Google actually said
On 10 July, Google's Search Advocate John Mueller weighed in again on the trend of building separate "AI-only" versions of websites, this time through markdown files or the increasingly popular llms.txt format, a text file some consultants claim tells AI models how to read your site. His position hasn't softened. Google Search doesn't use llms.txt, markdown mirrors, or any special machine-readable file to understand or cite a business. It reads your normal website. The same one your customers see.
Mueller's argument is blunt. Why would an AI system want to see a version of your site that no actual customer ever sees? A second, machine-only copy of your website isn't a shortcut. It's a second website you now have to build, update, and keep in sync forever, on top of the one you already have. Bing's search team has said much the same thing, adding that these parallel pages tend to get built once, then forgotten, then quietly go stale and broken while it's nobody's job to notice.
There's a sharper edge to it too. Search engines have a name for showing one version of a page to a crawler and a different version to a human. It's called cloaking, and it's against the rules everywhere, whether or not that was the intention.
Why this keeps getting pitched anyway
The reason this offer lands on so many desks is that the underlying fear is real. Google's AI Overviews, the summarised answers that now sit above the traditional blue links, appear on somewhere between 48% and 65% of all searches depending on which study you read. More broadly, well over half of all Google searches now end without a single click to any website. For a business that has spent years building organic traffic, that's a genuinely frightening number, and fear is exactly the condition under which a "special AI fix" sounds worth paying for.
The frustrating part is that the fear is pointed at the wrong solution. Building a machine-only mirror of your site doesn't make AI trust you more. It doesn't get you cited. Google's own guidance is that AI systems draw from the same well-built, well-structured page a person would land on, plus what's said about your business elsewhere on the internet, in reviews, articles, and directories. There is no separate side door.
The businesses being cited in AI answers right now aren't the ones with a clever technical workaround. They're the ones whose actual website says something worth quoting.
What's actually working instead
The data on who gets cited by AI Overviews and ChatGPT is consistent, and none of it involves markdown files. Content written by a named, credible person outperforms anonymous or generic copy. Pages updated within the past six to twelve months are cited far more often than stale ones, because freshness signals that someone is actually running the business. And specificity beats breadth every time. A page that clearly answers one real question a customer has, with a genuine opinion or a piece of first-hand experience in it, gets picked up ahead of a page that vaguely covers everything.
Structured data, the behind-the-scenes markup that tells search engines what a page is actually about, still matters and is worth doing properly. Google even added new options this year for product pages to state their own category more precisely. None of that is the same thing as building a parallel website. It's a small, sensible addition to the one website you already have.
There's also a real payoff for getting this right. Brands that do get cited inside AI Overviews see meaningfully more clicks, including paid clicks, than businesses that don't show up in the answer at all. The gap between being cited and not being cited is becoming one of the more important competitive lines in a business's marketing, roughly where ranking on page one sat a decade ago.
The practical takeaway
If someone offers to build you a separate AI-optimised version of your website, ask what problem it's actually solving. In most cases, the honest answer is none. The better use of that budget is auditing the website you already have. Is the person behind each page named and credible? Is the content current? Does the opening paragraph of your key pages actually answer the question a customer is asking, in plain language, before you get to the sales pitch? Is your business showing up with real, specific reviews and mentions elsewhere on the internet, not just on your own domain?
None of that is glamorous. It's also exactly what good marketing has always required. AI hasn't changed the fundamentals of getting found. It's just made the businesses that never bothered with the fundamentals a lot easier to spot.
Sources: Search Engine Journal, "SEO Pulse: Google Adds Social Reporting; Mueller Warns Against Markdown," 10 July 2026; Search Engine Land, "Google & Bing don't recommend separate markdown pages for LLMs"; Google Search Central developer documentation; SparkToro Zero-Click Study 2026; Search Engine Land AI Overviews citation and CTR research 2026.