Google has given a fairly clear warning about one of the more fashionable AI search ideas doing the rounds: creating separate markdown versions of website pages just for large language models.
According to reporting from Search Engine Journal, Google’s John Mueller and Martin Splitt discussed the idea on Search Off the Record and were cautious about it. Their view was not that markdown is bad in itself. It was more practical than that. If you create a second version of your website for machines while keeping the normal HTML version for people, you have doubled the amount of content you need to maintain.
For businesses in Bath, Somerset, Wiltshire, Dorset, Bristol and Gloucestershire, that is a useful reality check. AI search visibility matters, but it does not mean every local business needs a parallel machine-facing website. In most cases, the stronger move is to make the existing website clearer, better structured and easier to trust.
What Google was cautioning against
The discussion centred on the idea that publishers might create markdown copies of their pages because AI systems could find them easier to read. On the surface, that sounds tempting. Markdown is plain, lightweight and less cluttered than many modern web pages.
Martin Splitt’s point was that markdown is not a replacement for a good web page. People expect layout, images, navigation, colour and visual hierarchy. HTML has been built and tested around those needs for decades. If a business publishes markdown for users, it loses much of that experience. If it publishes markdown only for AI systems, it now has two versions of the same content to manage.
John Mueller added that parallel versions can be hard to debug. If the public web page breaks, a customer may complain. If the AI-facing version breaks, nobody may notice. A crawler or AI system might still find a page with some text on it and treat it as the intended version, even when it is stale, incomplete or wrong.
That is the real risk. Not markdown as a format, but the hidden complexity that creeps in when a website starts maintaining one page for humans and another for automated systems.
Why this matters for local businesses
Plenty of local organisations are now asking the same broad question: how do we show up when people use AI search tools instead of a normal Google results page?
That is a sensible question. Search is changing, and businesses should not ignore it. But the answer is rarely to bolt on a technical shortcut before the main website is doing its basic job well.
A Bath clinic, Bristol consultancy, Somerset retailer or Wiltshire venue still needs service pages that clearly explain what is offered, who it is for, where the business operates, what makes it credible, and how someone can take the next step. Those are not old-fashioned SEO concerns. They are also the sort of signals that AI-driven search and answer systems need when deciding what a business is about.
That is why practical AI search optimisation should start with the visible site. If the human-facing page is thin, vague, out of date or hard to navigate, a markdown duplicate is unlikely to be the thing that fixes the problem.
The problem with maintaining two versions
The awkward part of parallel content is that it sounds efficient right up until someone has to look after it.
Every time a service changes, both versions need updating. Every time a price, location, eligibility detail, opening hour or process changes, both versions need checking. Every time a page is redirected, noindexed, rewritten or merged, the machine-facing copy needs to stay in step.
That is manageable for a large technical team with strict publishing workflows. It is much less attractive for a regional business where website updates already have to compete with running the business itself.
It can also create a trust problem. If an AI system reads an old markdown version while customers see a newer HTML page, the business may end up being described inaccurately. That risk is not theoretical. Local service businesses often have details that matter: service areas, appointment types, delivery zones, accreditations, seasonal availability and contact routes. Stale versions can mislead people quietly.
What to improve instead
The more useful response is not to ignore AI search. It is to make the existing site easier for both people and machines to understand.
Start with page clarity. A good service page should make the offer obvious near the top, use descriptive headings, answer common customer questions, and avoid burying important information inside design elements that are hard to parse. If a human has to work too hard to understand what the page is saying, automated systems probably will too.
Next, look at structure. Clear HTML headings, sensible internal links, accurate title tags, useful summaries and well-written body copy all still matter. For many local businesses, improving website content writing and editing will do more than adding a second technical layer.
Then check evidence. AI search systems are often trying to resolve whether a business is real, relevant and reliable. Case studies, team information, location details, reviews, clear contact pages and consistent service descriptions can all help. This is where traditional SEO for Bath businesses and newer AI search work overlap more than people sometimes admit.
Finally, keep the site maintainable. A simple website that is accurate and regularly reviewed is usually stronger than a clever website with hidden versions nobody remembers to update.
When markdown may still have a place
None of this means markdown is useless. It can be a perfectly good internal writing format. Many teams draft in markdown before publishing in a content management system. Documentation sites, developer resources and knowledge bases may also use markdown as part of their normal workflow.
The caution is about treating a separate markdown layer as an AI SEO magic trick. If it creates a duplicate version of the site that users do not see, the business has to be honest about the maintenance burden and the risk of drift.
For most small and medium-sized organisations, especially those serving a local or regional market, that is probably not where the next hour of website work should go.
The practical takeaway
Google’s message is refreshingly boring, which is often a good sign. Do not make publishing more complicated than it needs to be. Do not create hidden duplicate versions unless there is a very strong reason. Do not chase AI visibility by neglecting the page real customers actually read.
For Bath and South West businesses, the better checklist is simpler: make your pages clear, useful, current and well structured. Explain your services properly. Link related pages sensibly. Keep important details accurate. Show enough evidence that both people and machines can understand why your business is a credible result.
AI search is changing how businesses get discovered, but it has not removed the need for a good website. If anything, it has made clarity and maintainability more important.
