Agent-Readiness measures whether an AI agent can correctly read, understand, and paraphrase your page without human assistance. It has almost nothing to do with whether the web page looks good or not, and it only partially overlaps with the human user experience. A website that looks beautiful and smooth in the browser may be a mess of noise that cannot be cleanly extracted by agents such as ChatGPT, Perplexity, and Claude. What tools like isitagentready do is to quantify this invisible gap into a score, letting you know how much available information is left on the page in the eyes of the machine.
What exactly does agent readiness measure?
Traditional SEO assumes that readers are human beings looking at images through a browser. Agent readiness assumes that the reader is a program: it gets the HTML source code, plain text and structured data returned by the server, rather than the full screen rendered by JavaScript. This difference is critical. Many websites put their key content after the front-end framework renders it so that people can see it when they open it, but the first-hand response the agent gets is an almost blank skeleton. What proxy readiness measures is how much information your page can retain without relying on rendering or human eyes, just machine-readable signals.
Taking it apart, it actually asks three things: whether the content can be extracted cleanly, whether the semantic structure is clear, and whether the machine can judge what the page is about and who endorses it. These three things simultaneously determine whether an agent will quote you and whether they will put your brand name, numbers, and conclusions in the right places when quoting. Moving it to the wrong location is worse than not being cited.
How to read isitagentready scores
isitagentready gives an overall score plus a set of sub-checks. The total score is not a vanity metric, it corresponds to how high the hit rate is when an agent crawls this page. A low score usually does not mean that the content is poorly written, but that the content is packaged in a way that makes it unreadable and misaligned for machines. What really matters is where the sub-items fall.
- Semantic HTML: Are the heading levels (h1 to h3) correct, and are paragraphs using real tags rather than a bunch of nested divs.
- Content extractability: Does the main text exist in the initial HTML, or does it have to wait for JavaScript rendering to appear?
- Structured data: Is there Schema.org markup (Article, FAQPage, Organization) to make fields directly readable by machines.
- Text-only friendliness: Whether a clean text or Markdown version is provided, with clear boundaries between paragraphs.
- Crawl permission: robots.txt and llms.txt are blocked or guided by AI agents.
- Authoritative signals: Machine-parsable trust cues such as author, publication date, source citations, etc.
Each item is not all-or-nothing, but has degrees. For example, a page may have a very clean semantic structure and get high scores, but if the key figures are all placed in pictures and there is no text replacement, this item will fall off. The result is that the agent can read your paragraph, but cannot copy the number you most want to be quoted.
What actually happens when an AI agent reads a page
Imagine Perplexity receives a question and decides to scrape your page. It sends a request that is close to plain text, gets back the HTML, throws away the styles and scripts, cuts the remaining content into segments, and then determines which segment answers the question. The whole process is completed in a few seconds, and no one is around to help it guess "what should be written on this picture." If your focus is hidden in a carousel, accordion, or block that requires a click to expand, agents are likely to skip it.
This also explains a common phenomenon: the same piece of content, written in a clean article structure, will be quoted, but disappear after being inserted into interactive components. The agent doesn't complain, it just quietly picks another source.
Common losing points
When running agent readiness checks for customers, the recurring problems are actually very concentrated:
- The content is rendered using client-side JavaScript, the initial HTML is almost blank, and the agent gets the skeleton and walks away.
- Use visual styles to pretend to be titles, and enlarge and bold divs so that machines can’t see the hierarchy.
- Key data, specifications, and prices are made into pictures, and there are no text or table versions to read.
- There is no structured data at all, or the Schema type is filled in incorrectly or does not match the screen content.
- The FAQ is presented as a purely visual accordion, but has no corresponding FAQPage markup.
- In the absence of llms.txt, the agent can only rely on guesswork to decide which pages to crawl and which to skip.
None of these require rewriting the content. They are all structural and engineering issues of "how to hand over existing content to machines." This is actually good news. Agent readiness is a small number of optimizations that are low-cost to change, but the return is directly reflected in the citation rate. It is much more cost-effective than rewriting an entire batch of articles.
How many points do you think you have passed? What are you going to do next?
There’s no universal pass mark, but here’s a pragmatic standard: If an agent can read only your initial HTML and spell out “who you are, what this page is about, what the key conclusions and numbers are, and who endorses it,” you’re roughly ready. The score is for locating gaps, not for framing.
The practical priorities are clear: first make sure the content is readable in the initial HTML, then add semantic tags and title levels, then add the right structured data, and finally handle llms.txt and authority signals. The order is reversed. First, pile up a bunch of Schemas but let the content not be rendered. This is equivalent to labeling things that cannot be read. If you want to know what your website looks like in the eyes of agents and which gaps are the biggest drag on citation rates, you can book a 30-minute GEO diagnosis and we will walk you through it with a complete agent readiness check.
