Should you serve Markdown to AI bots? The 2026 publisher debate

For the past year, more and more sites have started offering Markdown versions of their pages for AI systems to read. The idea is simple: strip away the JavaScript, ads, and styling, and give a model the plain, structured text underneath, the way tools like llms.txt and Cloudflare's Markdown-for-Agents work propose. In early August 2026 the trade press focused the question sharply. Digiday ran the case for and against markdown for AI bots, quoting publishers, ad-tech founders, and data brokers on both sides. It is the clearest snapshot we have had of where the industry actually stands.

If you write and publish — a post on LinkedIn, an article on Medium, a long X Article — this debate is not abstract. It decides whether the Markdown file you happen to have is an asset or a liability: something that helps machines find and cite your work, or something that quietly leaks it. More practically, it is a reminder that Markdown is now the lingua franca of a two-audience web: one audience of humans reading formatted rich text, and one of machines reading plain, structured source. The trick is publishing to both from a single source. That is exactly the workflow we have been recommending all year.

The case for: agents prefer Markdown, and it is cheaper

The "for" side starts from a mechanical fact about how AI systems read the web. When a model fetches a traditional HTML page, it has to push through JavaScript, ads, design elements, video, and photos before it reaches the underlying text. Markdown removes all of that. The content arrives pre-structured, with headings, lists, and links already marked up, so the model does less work to interpret it.

That efficiency has a name in the industry: the token economy. "HTML has a lot of tags and JavaScript and CSS and things that don't have to do with the content," TollBit co-founder and CEO Toshit Panigrahi told Digiday. "In terms of markdown, it makes it friendlier, the inference costs drop. Crawl budgets last longer." TollBit, which converts publisher pages to Markdown for AI companies, reported an average 90% reduction in tokens when content is converted to Markdown. Fewer tokens can mean lower compute costs — and some publishers are betting that efficiency eventually turns into better visibility in AI answer engines.

There is a revenue angle too. Some publishers are already trying to monetize the AI-bot traffic they get. Digiday reported that Time began serving ads inside its Markdown files last month, and Mobian co-founder Jonah Goodhart described prompting ChatGPT and Perplexity to pull a Time page and confirming the agents saw and passed along the ad content — an early test of treating AI bots like an impression you can sell against.

The case against: easier to scrape, hard to prove it pays

The "against" side is the mirror image. The same property that makes Markdown attractive to AI — it is clean and easy to read — is exactly what makes it easy to scrape without permission or control. Many publishers have spent the last two years trying to slow or block AI bots with detection tools and Cloudflare protections. Handing the bots a tidy Markdown copy walks that work back.

"When you take a website, you convert it to markdown, and you offer it to OpenAI or to Claude, it now has that data. You've now lost control... you've leaked your information into that model," Ed Zyszkowski, co-founder and CEO of Personal Digital Spaces, told Digiday. His analogy: it is like tearing out the walls of your house to make it more convenient for a commercial tenant, when you might prefer to rent out specific rooms on your own terms.

The other strike against Markdown is that the promised benefit is hard to prove. GEO platform Promptwatch analyzed over 1.6 million citations in AI answers and found Markdown did not increase the likelihood of getting content cited. And Google has weighed in directly: its generative-engine-optimization (GEO) guidelines say Markdown is not needed for Google search, and that the search engine reads regular HTML. Google later softened its language, noting that llms.txt files are "completely fine" to maintain for other services, but that Google Search ignores them — they neither help nor harm ranking. If you want to reach humans through Google, Markdown alone does not move the needle.

What the debate means for creators

Publishers with big pages can afford to experiment with Markdown endpoints and ad units in plain text. For an individual creator, the sensible takeaway is simpler and, frankly, more useful: keep Markdown as your single source of truth, and treat AI visibility and human reading as two separate output formats.

You do not need to rewrite how you publish. The Markdown file you draft your post in is already the machine-friendly version — structured, portable, and easy for any AI system to read if you choose to expose it. The human version is the same content converted to rich text. Whether you publish to LinkedIn, X Articles, or Medium, you are converting the same Markdown source into the formatted post your readers actually see. The 90%-token-economy argument only reinforces the point: plain-text structure is the format machines like, and it happens to be the format that keeps your work portable and yours. For the privacy trade-offs, our client-side versus cloud converter comparison walks through why keeping conversion local matters.

A two-audience publishing workflow

Here is a workflow that gives you both audiences without extra work. It assumes you already draft in Markdown, which we covered in our Markdown beginner's guide.

Step 1 — Draft once in Markdown. Your source of truth is a plain-text file with headings, bold, lists, and links. This is the format AI systems read best, and it is the version that lives forever, independent of any platform.

Step 2 — Decide if you expose the Markdown to bots. If you run a site and want to be friendly to AI readers, you can publish an llms.txt or Markdown copy alongside your HTML. Just know, per Google's guidelines, that this is for AI services, not for Google ranking — and if you are wary of uncontrolled scraping, you can skip it entirely.

Step 3 — Convert to rich text for humans, on your device. This is where the human version is made. Open the Markdown in md2rich, which turns it into formatted rich text with working headings, bold, lists, and links. Conversion runs entirely in your browser — nothing is uploaded, no account is required — so the "leaked into a model" concern never applies to this step.

Step 4 — Paste into the platform editor. Copy the formatted output and paste into LinkedIn, X Articles, or Medium. The formatting comes along as native rich text, so it looks hand-formatted rather than pasted.

Two examples: one Markdown source, two outputs

Here is how the same Markdown produces a clean AI-readable source and a formatted human post. First, the Markdown source:

# Publishing for humans and machines

**The takeaway:** one Markdown source, two audiences.

- Machines read it as clean, structured text (fewer tokens)
- Humans read it as formatted rich text after conversion
- The source stays portable and yours either way

That exact file is what an AI system would consume if you exposed it — headings, bold, and bullets already structured. Now the same content after conversion to rich text is what your reader sees on LinkedIn: a bold headline, an emphasized opening line, and a clean bulleted list, rendered natively by the platform. A second example with a link:

# Keep your writing format-agnostic

A Markdown file is just text, so it outlives any app.
Write once, publish with formatting: [md2rich](https://md2rich.com/)

After conversion, the link becomes a real clickable rich-text element, and the heading renders as a formatted headline. The machines and the humans both get what they need from the same Markdown — which is the whole point of the debate, resolved in one small workflow.

The bottom line

The publisher debate over Markdown for AI bots is really a debate about control and payoff: whether the efficiency of plain text is worth the ease of scraping, and whether AI visibility payback is real yet. For a creator, you do not have to pick a side to benefit. Writing in Markdown already puts you in the best position for both futures. If AI-first reading wins, your source is ready. If human platforms keep winning, your conversion path is ready. Either way, the format that gives you both is the same one you already use.

FAQ