John Mueller (Google) explained this week on Bluesky why Google serves its developer documentation also in plain Markdown. The purpose: helping AI tools (Copilot, Cursor, code assistants) parse the content without cleaning HTML. Search Engine Journal covered the thread and opens an interesting discussion for technical blog creators: should your blog have a Markdown version? Spoiler: probably not. Mueller was very clear on when yes and when no.
Mueller's framework: discovery vs functionality
Mueller distinguished two distinct purposes of internet content:
- Discovery: people find you. Classic SEO. Optimizing for Google Search, AI Mode, Bing, ChatGPT.
- Functionality: people (or an AI) can use your content to solve a task. Here Markdown helps because it removes noise and leaves only processable text.
For programming docs, both goals coincide: the developer searches "how to use X API" (discovery) and then a coding agent reads your documentation to generate code (functionality). That's why Google serves both formats.
For a typical blog (lifestyle, marketing, opinion, general tutorial), discovery is 95% of the value and functionality is marginal. Mueller says it without rounding off: for most sites, Markdown versions won't produce measurable results against competitors.
When it DOES make sense to serve Markdown from your blog
| Blog type | Serving Markdown helps? | Why |
|---|---|---|
| Own API or library docs | Yes | Coding agents consume it literally |
| Technical step-by-step tutorials with code | Maybe | AI assistants can reproduce your steps |
| Cheat sheets, quick references | Maybe | Useful for contextual auto-completion |
| Opinion, marketing, lifestyle blogs | No | No use case justifies the cost |
| Generic B2B company blogs | No | What matters is well-done classic SEO |
What DOES work to rank in AI search
Mueller called Markdown optimization a "temporary crutch" and refreshed real advice: what works in AI search is the same that worked in well-done classic SEO.
- Clear semantic structure: single H1, H2s that delimit sections, H3s that detail. AI parsers read structure, not tricks.
- Schema markup: Article, FAQPage, HowTo, Recipe, Product. Still the common language between your HTML and crawlers.
- Authoritative and well-attributed content: visible citation ("according to...") and byline with credentials move the needle on who AI cites.
- Performance: Core Web Vitals still weigh. A slow blog gets cited less.
Mueller was direct: "sites should evaluate agentic optimization based on measurable results for their specific context". Translation: don't invest time optimizing for LLMs until you have data proving that traffic reaches you. Meanwhile, basics done well win.
The mistake many technical blogs are making
After reading this kind of threads, several blogs have added .md versions of every article, llms.txt tags, and routines to serve Markdown to user-agents detected as AI. The problem: these patches add complexity without evidence of measurable improvement in their real traffic. It's fascination engineering, not return engineering.
Operating rule: before touching your publishing pipeline to serve alternative formats to AIs, look at your Google Search Console and your blog's analytics. If your human organic traffic is healthy and growing, don't touch. If it's stagnant, the bottleneck is content or structure, not format.
What to do if you have a technical blog
What is worth it
- Keep your code snippets in pre/code blocks with language-x class. Both humans and AI understand it well.
- Use descriptive headings ("How to configure X step by step") instead of creative ones ("The secret behind X").
- If you publish docs of your own library, consider serving them also in Markdown format without ceremony (you don't need a complicated process, a /raw/ route per article is enough).
What is not worth it
- Creating duplicate .md files of every marketing or opinion post.
- Implementing AI bot detection with differentiated responses (technical overhead + maintenance).
- Buying "GEO" (Generative Engine Optimization) SaaS without first having good classic SEO.
"For non-developer sites, Markdown versions won't produce measurable results against competitors" (John Mueller, paraphrasing the Search Engine Journal recap). One of the most sober pieces of AI search advice of the year.
Conclusion
The hype around serving Markdown to AIs is understandable but miscalibrated for most blogs. Mueller cleanly separates discovery (who you want to reach) from functionality (what task you facilitate). Only when both converge (technical docs with coding agents) is the effort worth it. For the average blog, it's still more profitable to invest in clear structure, solid data, and visible authorship.
If you want a blog platform that already comes with schema markup, semantic headings, and well-formatted code blocks without touching configuration, try Vlogerly: basics done well, ready from the first post.
