# Agent Readiness Score: What It Measures > An agent readiness score rates whether AI agents can reach and use your site. What the checks measure, why evidence-backed ones weigh more, and a free check. Page: https://trylectern.com/blog/agent-readiness-score By Edgar Li, October 24, 2026. An agent readiness score is a single rating of how well a website works for AI agents: the assistants and tools that fetch pages on a person's behalf. Most versions combine several checks, such as whether robots.txt lets the agents in, whether the firewall serves them, and whether the page content is readable without a browser. The number is only as good as the checks behind it and the weight each one gets. Lectern's site audit gives one: a readiness score from 0 to 100 that puts 90% of its weight on checks with evidence behind them and at most 10% on best practice. Its free check, [Can AI agents read your site?](https://trylectern.com/agent-access), gives no grade; it reports what it found for each agent and names the rule behind each block. This post explains what such a score should measure, how to weigh the checks, and how to read any score you are shown, including ones from other tools. ## Key Takeaways - A readiness score is a summary. Read the checks under it, because two sites with the same number can fail in different places. - Checks that have evidence behind them should count for most. Checks that only follow a best-practice list should count for little. - The strongest evidence is direct: a robots.txt rule that names an agent, or a request that names an agent and gets refused while a browser is served. - A request sent from a checking tool is one reading. A firewall that verifies bots by address may treat the real agent differently. - A high score does not mean AI assistants will cite you. It means they are able to read you. ## What Agents Need From a Site An agent can only use a page it can fetch and read. Three things decide that, in order. 1. **robots.txt.** The file at yourdomain.com/robots.txt lists which crawlers may fetch which paths. Each assistant uses a different crawler for different jobs. See [which AI crawlers exist](https://trylectern.com/blog/ai-crawler-blocking). 2. **The edge.** A CDN or firewall can refuse or challenge a request whatever robots.txt says. This is how many sites block AI assistants without meaning to. 3. **The page.** Directives such as `noindex` on the homepage tell crawlers not to use it, and content that only appears after scripts run may not be seen by a fetcher that does not run them. These are the things a readiness check can observe from outside. They describe access, not quality. An agent that can read a thin page still has little to cite. ## Two Tiers of Checks Not every check on a readiness list has the same standing. We separate them into two tiers. **Evidence-backed checks** test something that directly decides whether an agent can read the page, and the result can be shown. - robots.txt blocks a named agent, with the exact line quoted. - A request naming ChatGPT-User, Claude-User or Perplexity-User was refused, or met a challenge page, while a browser request to the same page was served. - The homepage carries a directive such as `noindex` that tells crawlers not to use it. **Best-practice checks** are things people recommend because they seem sensible, with little or no measured effect on whether assistants read or cite a site. A file in a particular format, a metadata convention, or a checklist item copied from another checklist belongs here. We looked at one of these, [llms.txt, and the data on whether crawlers use it](https://trylectern.com/blog/llms-txt-doesnt-work). A fair score gives the first tier most of its weight and the second very little. If a score counts a missing best-practice file as heavily as a firewall that refuses ChatGPT-User, it will rank sites in the wrong order, and a team could spend a week on the file while the firewall stays shut. ## What Lectern's Free Check Reports The free check at [/agent-access](https://trylectern.com/agent-access) reads a site from outside and reports: - Which tracked AI crawlers robots.txt allows or blocks, grouped as search, AI search, Gemini grounding and AI training, with the rule that decided each. - Whether the homepage carries directives that tell crawlers not to use it. - Whether requests naming ChatGPT-User, Claude-User and Perplexity-User were served the same page as a browser, and which edge, such as Cloudflare, refused them when its headers say so. - Fix steps when an agent's request was refused. It sends those requests from Lectern's own servers under each agent's name. It reports them as what those requests got, never as what the real assistant gets. It does not check for llms.txt or other well-known files, and it gives no overall grade. Those are in the site audit, below. ## The Readiness Score in Lectern's Site Audit Lectern's site audit, which every workspace can run, including on the free plan, has an agent-readiness section with a score. - **Two tiers.** Checks in the evidence tier carry 90% of the score. Best-practice checks, such as llms.txt and other well-known files, carry at most 10% however many there are. Each check is labelled with its tier. - **What the evidence is.** A check is in the evidence tier when published data on agent runs links it to how those runs went, not only to how often a file was fetched. Four checks qualify today: the robots.txt policy for agents, content readable without JavaScript, whether requests naming the agents are served, and pricing stated on the site. That data is correlational, and each check says so. - **No penalty for what could not be read.** A check Lectern could not read, or one that does not apply to the site (no API, nothing sold), is left out of the score rather than counted as a fail. - **No number without evidence.** If no evidence-tier check could be read, the section says "Not scored" instead of building a number from best practice alone. - **Separate from visibility.** The readiness score does not change the audit's main score or any visibility measure. ## How to Read Any Readiness Score 1. **Ask what is in it.** A score with no list of checks cannot be acted on. 2. **Find the evidence-backed failures first.** One refused agent request matters more than five missing best-practice items. 3. **Treat training blocks separately.** Blocking a training crawler is a legitimate choice and does not stop assistants from citing a site. Blocking a search or user-triggered agent does. 4. **Re-run it.** Firewall rules and robots.txt change in unrelated deploys. We found [our own robots.txt blocking AI crawlers for five months](https://trylectern.com/blog/we-blocked-ai-crawlers-for-five-months) because nobody re-read it. 5. **Do not read access as visibility.** Whether an assistant names your brand is measured separately, in [AI share of voice](https://trylectern.com/blog/ai-share-of-voice). ## Where Readiness Fits Readiness is the agent-experience half of the picture: can agents do their job on your site. Answer engine optimisation is the other half: do the answers mention you. [AX vs AEO](https://trylectern.com/blog/ax-vs-aeo) covers how they relate. Run the [free agent access check](https://trylectern.com/agent-access) on your domain. If it finds a block, fix that first. To track access and AI answers together every week, see [Lectern's plans](https://trylectern.com/pricing).