Methodology

What we check

AI agents are starting to browse, shop, and book on behalf of users. Your website either works for them or it doesn’t. We measure which.

The framework

We score four things: can an agent find your site, navigate it, read the content, and complete a task. Each pillar measures how well your site supports that step without relying on visual cues.

1. Discoverability

Before anything else, an agent needs to know what your site is and what it offers. We check for the machine-readable signals agents use to orient themselves — an llms.txt summary, structured data, an AI-crawler policy, and a current sitemap — so they can understand you without guessing from the visual layout.

1.1llms.txt file

A plain-text index that tells agents what your site does and where the important pages live.

1.3Structured data

Schema.org / JSON-LD markup describing your content as products, articles, organisations, or events.

1.6Sitemap

A current XML sitemap so agents can see every page at once instead of crawling their way there.

+ 2 more criteria

2. Navigability

Once an agent has found your site, it has to move through it without visual cues. We measure whether the HTML structure, link text, and URL patterns are clear enough for an agent to follow routes, skip dead ends, and understand the page hierarchy at each step.

2.1Semantic HTML

Proper landmarks — nav, main, article, section — and a clean heading hierarchy so agents can read the page as a tree.

2.3Link quality

Descriptive link text that names the destination, not “click here” or “read more,” and no javascript: void links.

2.5URL structure

Clean, predictable URLs that reflect content hierarchy, so agents can guess routes rather than crawl blind.

+ 3 more criteria

3. Content Quality

Agents don’t render pixels — they read source. We check whether your content is available in formats agents can ingest efficiently, without wading through ads, scripts, and interface chrome, and whether the page actually delivers its content in the markup rather than behind JavaScript.

3.1llms-full.txt

A single markdown file with your core content, so agents can ingest the whole site in one request instead of crawling page by page.

3.3Clean HTML

Core content visible in the source, not hidden behind accordions, click-to-reveal patterns, or images without alt text.

3.4Content depth

A strong ratio of real content to navigation, ads, and boilerplate, so agents don’t burn context on chrome.

+ 1 more criterion

4. Task Clarity

This is the hardest layer. We check whether your site tells agents what actions they can actually take — not just what they can read — through machine-readable task manifests, descriptive button text, and direct URLs for key flows. Humans fill in the gaps from visual cues; agents need it spelled out.

4.1Task manifest

A published list of agent-invocable actions via llms.txt instructions, agents.json, AWP, AGENTS.md, or WebMCP attributes.

4.2Action descriptions

Button and CTA text that names the outcome — “Enrol in course” — not just “Submit” or “Go.”

4.3Flow shortcuts

Deep links for transactional flows — checkout, signup, contact — so agents can trigger them directly rather than clicking through.

Scoring

What your score means

Drag the slider to explore the five bands of agent readiness.

0/ 100

Agent-Fragile40–59

Agents can partially interact, but key flows break. The structural foundations are there — the machine-readable signals and task guidance aren’t.

Test your site

Generate your agent readiness report →