A Human Version and an AI Version: Why Every Page on Our Site Has a Markdown Twin

As of this week, every page on our site has two versions: the one you're reading right now, and a plain-text Markdown version built for the AI agents and tools that crawl the web looking for content. We added a small toggle at the top of every page to switch between the two, plus a file that lists all of our pages for those same tools. Before recommending anything to a client around AI search visibility, we'd rather test it on our own site first. Here's what we built, why, and just as important, what it doesn't do.

What is the HU | IA toggle at the top of every page?

It's a two-position switch, visible in the header and in the menu. "HU" opens the page as you're seeing it now: layout, images, navigation, everything. "IA" opens the same page as plain Markdown, no styling, with headings, paragraphs and links written in simple syntax that any text-processing tool can read without parsing HTML. Both versions exist for every published page on the site, not just blog posts: services, case studies, zone pages, all of it. We didn't try to be subtle with the label. A visitor who clicks it out of curiosity should understand right away what just changed, without needing to know what a Markdown file is.

What is an llms.txt file, and why did we add one?

An llms.txt file is a plain-text list of your pages, a bit like a sitemap, but written to be read by a language model rather than parsed by a traditional search engine. Ours lists every published page and post, organized by language and page type. We also built a second, more detailed file that gathers our definitions, our offers and our case study results along with their sources. The idea behind both files is simple: if a generative AI tool is trying to figure out who we are and what we do, we'd rather hand it a clear, current version than leave it to guess from a page full of menus, carousels and scripts.

Does Google require an llms.txt file or a Markdown version of your pages?

No, and we want to say that plainly because it's a common source of confusion right now. Google Search Central's generative AI optimization guide, updated in July 2026, is explicit on this: there is no Google requirement for an llms.txt file, no AI-only markup, no Markdown pages, no required chunking format, and no benefit to rewriting content purely for AI systems. So we didn't build this thinking it would improve our Google ranking. We built it because our AI search visibility work is about helping small businesses get found and cited by tools like ChatGPT or Perplexity, and we didn't want to sell something we hadn't tried on ourselves first.

How did we keep these Markdown pages from hurting our own search rankings?

This is the part that took the most care. Publishing hundreds of duplicate pages, even in a different format, can easily create a duplicate content problem if it isn't handled properly. So every Markdown page sends a noindex, follow header: it stays reachable to anyone with the link, including AI tool crawlers, but we're explicitly telling Google not to index it as a separate page. None of the Markdown pages appear in our XML sitemap, and every Markdown version points back to the canonical URL of the matching HTML page, which stays the reference version for regular organic search. That's the kind of detail we check before rolling anything out at scale: better to work it out once, on our own site, than to find out the hard way on a client's.

Should a small business without this kind of setup worry about it?

In most cases, no. Most Montreal-area small business owners we work with don't need to build a whole dual-page system to rank well or to be readable by an AI tool. What actually matters, and what helps both regular search and visibility in AI-generated answers, is a lot more basic: does your site respond properly with no server errors blocking access, is your main content real text rather than just images, and is your business name, address or service area, and phone number written the same way everywhere on your site. Those are the same fundamentals we check first on any engagement, well before Markdown or llms.txt ever come up. A site built on a stable content management system with a custom theme, rather than a page builder loaded with extra scripts, already starts ahead on this front. That was true of the site we rebuilt from scratch for DF Médico-Esthétique, where the priority wasn't Markdown at all, just a clean, fast site that's easy to keep updated.

Here's what we check before we ever get near an llms.txt file, and what you can check yourself in a few minutes:

1. Open an important page on your site in a private browser window and confirm it loads with no error, no unexpected redirect, and no login wall.
2. Try to select the main text on the page with your mouse. If it won't select because it's actually an image or an embedded PDF, a search engine likely has trouble reading it too.
3. Compare your business name, service area, and phone number across your site, your Google Business Profile listing, and your social accounts. Even a small mismatch raises doubt about which version is correct.
4. Confirm your XML sitemap exists, usually at /sitemap.xml or something close to it, and that it actually lists your important pages, not just the homepage.
5. If your site runs on a page builder stacked with scripts and widgets, ask whether it's slowing your mobile load time enough to matter. That's a measurable problem, unlike the absence of an llms.txt file.

None of those five checks take deep technical knowledge, just a few minutes and an honest look at what a visitor, or a crawler, actually sees when they land on your site.

Will we know if this changes anything for our own visibility?

Not yet, and we'd rather say that than make up a number. The system only went live this week; we have no data so far on how often AI tools actually pull our Markdown pages, or on whether it changes how often we get mentioned in AI-generated answers. We'll track it the same way we track a Google Ads account or a Google Business Profile listing: by watching what actually moves in the data, not by assuming a technical change automatically produces a result. If there's something concrete to report in a few months, that'll be its own article, not a claim today.

If you're wondering how, or whether, your business shows up in the AI tools your customers are already using, that's exactly the kind of question we look at in our AI search visibility work.

Drafted with AI assistance, checked and published by Marven Salgado.