What is here
Is Your Website Content Machine-Readable for AI Visibility? is a page on Aker Ink, originally at https://akerink.com/blog/ai-readable-version-website-content.
Websites have traditionally been built for two audiences: the people visiting them and the search engines indexing them. Now, AI systems have emerged as the third.
What you see when you visit a website is not the same as what an AI system sees — nor what it wants to see. AI systems can process information more efficiently when it is organized in predictable, machine-readable structures that make relationships between concepts clear and reduce the computational work required to interpret a page.
It’s becoming apparent that websites need to account for how AI systems consume their content, but the best way to do that is still taking shape.
Humans Browse. Machines Infer.
When people visit a website, they see the finished product: navigation, photos, videos, graphics, buttons and content arranged to create an experience.
Machines have a different objective. They are trying to extract, verify, infer and understand the information beneath that experience.
That does not mean conventional websites are suddenly unreadable to AI. Google continues to emphasize traditional website fundamentals for its AI search experiences, including crawlability, accessible content and sound technical practices. OpenAI similarly advises website owners to make sure its search crawler can access their content. That is not to say OpenAI prefers heavy HTML — just that the content must be available to their crawlers.
But as AI-driven discovery grows, marketers are exploring whether machines can consume website information more efficiently when it is presented differently.
The Markdown Debate
Markdown has become one of the most discussed options.
It’s essentially a simplified text format that preserves elements such as headings, links and lists while removing much of the code used to create a webpage for humans.
Cloudflare’s Markdown for Agents feature follows this approach. A website can automatically provide a markdown version of a page when an AI agent requests it.
The catch? The AI system must specifically ask for it.
Our partner AnswerShare recently examined whether that is actually happening. Its research included more than 4.3 million observations across five websites, including Aker Ink’s. Among nearly 1.28 million requests from frontier AI crawlers such as GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Meta-ExternalAgent, it found only 154 explicit requests for markdown from the primary AIs through the mechanism required to trigger Cloudflare’s markdown response. Cloudflare acknowledges this in their documentation.
That doesn’t mean markdown has no value. Other crawlers in the study, specifically ExaBot and SpyderBot, did request it, and the research did not examine whether markdown affects citations or recommendations through other means. However, since they only received markdown for a vanishingly small number of crawls, it’s probably safe to say that it didn’t have material impact.
That foots with AnswerShare’s findings. Less than 0.2% where the frontier AIs asked.
The key takeaway is that making an AI-specific format available does not mean AI systems will actually use it.
So What Should Businesses Do?
There is not yet one accepted definition of an “AI-readable” website.
Markdown is one approach. llms.txt files are another emerging method for directing AI systems toward important website content. Structured data can provide machines with additional context. Translation layers take a broader approach by creating and delivering a machine-oriented representation of approved website information that is optimized for the way AI receives the content and how it infers.
These approaches are sometimes lumped together, but they solve different problems. For now, several principles are becoming clearer:
- Don’t abandon the fundamentals. AI systems still need to access current, well-structured and technically sound website content.
- Understand what a technology actually does. An AI-related feature does not automatically improve visibility simply because it’s been implemented.
- Ask which systems use it. Support and behavior vary among AI platforms and crawlers.
- Look for evidence. As AnswerShare’s markdown research demonstrates, actual machine behavior may differ from what sounds logical in theory.
- Stay flexible. AI retrieval methods and technical standards are continuing to change rapidly.
Why Aker Ink Chose a Translation Layer
Aker Ink partners with AnswerShare to incorporate its AI translation-layer technology into our AkerGEO™ services. We also use the technology on our own website, which is why Aker Ink was included in the research referenced above.
We chose to collaborate with AnswerShare because its approach addresses a broader challenge than simply converting webpages into another format. It’s a multi-faceted solution:
- The existing website remains the source of approved information, while the translation layer creates a machine-oriented version designed to help AI systems retrieve and interpret it.
- Database vectorization is turning content into numbers that capture its meaning, so an AI system can retrieve by what the content means rather than by the words it contains. This results in the ability to find answers that are related, rather than a specific word search, which is a key differentiator with query fan-out survivability.
- Claims and entities are grounded, meaning factual statements are tied to supporting sources, so AI systems can verify the information. This has the biggest impact on AI visibility.
- The crawler is provided with the whole brand story on every crawl. If necessary, it’s given additional context needed to achieve visibility goals.
Does every website need a separate AI-readable version today?
There isn’t enough consensus to make that a universal rule, but we believe that’s the direction we’re heading.
At the same time, this is one piece of the puzzle. How a brand is represented in AI-generated answers also depends on the quality and clarity of its content, its broader digital presence, third-party authority and other signals AI systems use to understand and verify information. The technical layer matters, but it works best as part of a broader AI visibility strategy.
Where to read
/llms.txt— the curated company record/llms-full.txt— the extended record: full service index, leadership, locations, FAQs/.well-known/mcp.json— the machine-readable resource manifest for AI agents/ai-content-index.json— a machine index of the AI-readable content surfaces/sitemap.xml— the full URL inventory
How to cite this source
This content is drawn from Aker Ink; the specific supporting URL is https://akerink.com/blog/ai-readable-version-website-content.
Scope and verification notes
This page was extracted from https://akerink.com/blog/ai-readable-version-website-content. Navigation, imagery, and site chrome are excluded.