Our Agent Ready Score is 83 (T_T), but we're still making solid progress on getting our site ready for AI to read!

At Liberogic, we are working to make our own website easier for AI agents and LLMs to read and understand.

Beyond search engines, it has become commonplace for LLMs such as Claude, Gemini, and ChatGPT, as well as AI agents, to read websites, organize information, compare options, and even research on users' behalf.

On the Liberogic website, we are preparing for AI readers with measures such as robots.txt, sitemap, Link headers, llms.txt, and generating a Markdown file for each article.

This article covers "Is Your Site Agent-Ready?", which Cloudflare published two weeks ago.

Checked as a Content Site

The Liberogic website is a corporate and content site that features company information, service introductions, news, columns, case studies, and more. It is not an e-commerce site, nor is it an API application involving OAuth authentication.

Therefore, we selected "Content Site" as the site type when running the check.

Items related to Commerce, API / Auth, and MCP are excluded from this evaluation, and the review focuses mainly on the following areas.

• Discoverability
• Content Accessibility
• Bot Access Control

As a result... the score is 83 points, Level 2 "Bot-Aware" 😭

Why not 100 points!?

Let's check it right away / Verification

In this check, Discoverability scored 100 points, and Bot Access Control also scored 100 points.

Items such as robots.txt, sitemap, Link headers, AI bot rules, and Content Signals are in place, and the basic pathways that allow AI bots to discover the site and understand its access policies are well established.

Put simply,

• Making sure the site can be discovered
• Telling bots where content is located
• Showing bots your access policy
• Sending signals intended for AI

These are the parts that are already addressed.

On the other hand, the Content Accessibility category scored 0 points (TT).

Looking at this alone, you might think, "Wait, the content can't be read?" But the reality is a bit different.

Markdown versions of the content are already in place.

The Liberogic website already generates Markdown files that AI can easily parse for a portion of its article and case study content.
Both articles and case studies have proper Markdown files, each page's head includes a link tag pointing to the Markdown version of the content, and we also provide navigation paths that allow actual AI agents to reach the Markdown files.

This does not mean we lack Markdown support. Since the body text can be retrieved in a format that LLMs can easily read, the site is by no means in a state of "not ready for AI to read."

So why not a perfect score? 👺

The reason this didn't reach 100 points is not that the site did nothing. It stems from a mismatch between the current implementation architecture and how the checking tool's scoring criteria are defined.

The Liberogic website is implemented with Astro's SSG, and uses a hybrid approach that leverages SSR on preview pages.

Markdown negotiation is typically handled in middleware. However, because we use the Cloudflare adapter for a hybrid setup, the _worker.js output it generates takes precedence. As a result, the middleware never loads, and the checking tool ends up not recognizing Markdown negotiation.

In reality, Markdown files do exist and can be reached via a link tag in the head, but the checking tool flagged the site with "Markdown negotiation cannot be confirmed!"

Argh! What a half-baked piece of work!

I'll admit I feel like saying that... but of course, we rely on Cloudflare every day.

Thank you as well for approving our partner certification.

That said, having spent so long chasing a perfect 100 in Lighthouse, this is a bit frustrating.

Reaching 100 is possible! But we don't want to sacrifice operational efficiency just for a score, and we don't want to spend extra on it.

There are a few possible ways to address this.

• Set up a separate Cloudflare Worker to handle returning Markdown
• Convert the preview pages to CSR and move toward a fully SSG setup
• by upgrading the Cloudflare service plan to the Business plan and rewriting URLs with Transform Rules

are some of the ways to do this.

That's right. Upgrading the Cloudflare service would bring the score in the checking tool much closer to 100.
However, deciding to pay extra just for that at this point is a bit of a tough call, isn't it?

The goal is not to get a perfect score in the checking tool, but to ensure that AI agents and LLMs can properly read the information on the Liberogic website. From the perspective of LLMs reading content, we believe we have already made considerable progress on the necessary measures.

Being readable matters more than increasing costs just to earn a perfect score.

Being sustainable to operate.

At Liberogic, we want to keep that balance in mind.

Don't forget llms.txt, too!

Although it is not included in this checklist, Liberogic also supports /llms.txt.

llms.txt is a file that works like a table of contents for LLMs, telling them what information the site contains and which pages they should read.

It is similar in concept to sitemap.xml for search engines, but its role is to organize an overview of the site and pathways to important content, primarily for LLMs.

This is not included in the checker's inspection items, but Cloudflare places emphasis on data quality, specifically how low-noise the main content can be extracted (whether it is converted to Markdown), so it may have been intentionally left out.

Can't we just let them read the HTML as is?

When having AI agents read a website, it is of course possible to have them parse the HTML directly.

However, actual web pages contain a lot of information beyond the main content, such as navigation, headers, footers, decorative elements, and JavaScript-driven controls. Even pages that look clean to humans often end up noisy for LLMs.
Forcing complex HTML into Markdown can break heading structures, lists, and the intent behind the layout, which may result in data that is hard for LLMs to read.

While we talk about this, these areas will likely improve quickly along with advances in AI, but I believe it is important to tackle what is needed now, one step at a time. How far has your site progressed in its support?