In the SEO and AI community, there is an increasingly intense discussion about the idea of creating special Markdown (.md) or JSON pages intended exclusively for LLM systems. However, according to the views of representatives of the largest search engines, such a practice is not recommended and can bring more harm than good.
The reason for the discussion was a question that Lily Ray asked on the social network Bluesky, where she stated that there is more and more talk about serving special URLs for AI bots, with content adapted to LLMs, while users get a different version of the same page.
Google and Bing warn about “cloaking”
John Mueller from Google responded to this, pointing out that LLM systems train and work on standard HTML pages from the very beginning. In his opinion, there is no technical reason to serve content to LLMs that users cannot see, and such a practice could easily be seen as a form of cloaking, which is against Google’s rules.
In later comments, Mueller was even more direct, judging the idea of mass conversion of sites to Markdown as pointless, noting that modern models understand HTML and even images without problems.
A similar attitude comes from Microsoft’s Bing team, which warns that such a strategy would only increase the load of the crawl process, because search engines would check the correspondence between page versions anyway. He emphasized that Bing relies on structural data such as Schema tags and that AI systems already understand classic websites very well, with the message that “less is often more” in SEO.
The reason why this topic is important lies in the fact that part of the SEO community is looking for shortcuts for better visibility, both in classic search and in AI systems. However, as Lily Ray herself later pointed out on LinkedIn, serving different content to AI bots and users directly enters the zone of violation of long-standing search engine rules and can lead to serious negative consequences.
Basically, the message from Google and Bing is clear: optimization for LLMs should not mean parallel versions of content, but high-quality, unique HTML that is the same for both users and bots, writes SEL.