# Legible vs Cloudflare
Markdown for Agents

## Content quality: HTML scraping vs CMS-native

The cleaner the source content, the better the AI outcome. Cloudflare starts from rendered HTML and has to infer what matters — stripping away navigation, scripts, and presentation code to find the actual content. This works reasonably well for simple pages, but complex layouts, JavaScript-rendered content, and interactive elements often leave artifacts in the output.

Legible starts closer to the original content source by connecting directly to your CMS API. This means it works with the structured data your team already created — titles, body content, metadata, categories — rather than trying to reverse-engineer that structure from rendered HTML. The result is consistently cleaner output with fewer formatting artifacts, which means AI systems get a more reliable representation of what your content actually says.

## Token efficiency: why ~2,800 beats ~5,000–8,000

AI systems process content in tokens, and every model has a limited context window. When a typical blog post consumes 15,000 tokens as raw HTML, the AI system spends most of its budget on navigation menus, footer links, scripts, and structural markup before it ever reaches your actual message. Cloudflare's Markdown conversion reduces this to roughly 5,000–8,000 tokens, which is a real improvement.

Legible takes it further by starting from CMS-structured data rather than rendered HTML, typically producing around 2,800 tokens per post. This additional reduction matters because it means the AI system can read more of your content in a single pass, produce more accurate summaries, and allocate more of its context budget to understanding your message rather than filtering out noise. For sites with hundreds of pages, the compound effect of better token efficiency can meaningfully improve how well AI systems represent your brand across all interactions.

## Discovery files and AI metadata

Markdown conversion is one piece of the AI visibility puzzle, but it is not the whole picture. AI systems also need discovery signals — files and metadata that tell them where to find your most important content and how to prioritize it. Legible automatically generates and maintains llms.txt and ai-sitemap.json, two emerging standards that help AI crawlers understand what your site offers without scraping every page.

Cloudflare's Markdown for Agents feature does not include discovery file generation, so teams using it still need to create and maintain these files manually or go without them. For teams that publish frequently, keeping discovery files in sync with actual content changes is a maintenance burden that tends to be forgotten until it causes problems.

## Analytics and visibility

One of the biggest gaps in most AI content strategies is measurement. Standard analytics tools like Google Analytics do not track AI crawler visits because these bots typically do not execute JavaScript. Without visibility into which AI systems are reading your content, how often they visit, and which pages they prioritize, it is difficult to know whether your GEO efforts are producing results.

Legible includes AI-specific crawler analytics that identify visits from ChatGPT, Claude, Perplexity, Google AI, and other systems. This gives teams a clear feedback loop: publish content, see AI systems consume it, and measure changes over time. Cloudflare offers general bot analytics through its dashboard, but it is not specifically designed for tracking AI content consumption patterns or providing the kind of GEO-focused insights that marketing teams need to make informed decisions.

## Publisher control and permissions

As AI systems become more capable and more prevalent, publishers increasingly want granular control over how their content is used. Can an AI system cite a specific page? Can it use the content for training? Should certain content be available for retrieval but not for scraping? Legible provides per-content permission controls and coordinated policy signals through Content-Signal headers, structured metadata, and crawler-specific directives. This gives publishers a clearer way to communicate their intent than relying on blunt instruments like robots.txt alone. Cloudflare's approach focuses on blocking or allowing AI access at the site level, which is useful but does not provide the same level of per-content granularity that many publishers need.