Retrieval Legibility
Retrieval legibility is how easily a machine can read, parse, and embed the actual facts on your page. It is our term for a plain but decisive reality: a retriever only knows what it can read, and a great deal of what humans see on a screen is invisible to the crawler underneath. A price rendered as an image, a phone number assembled by a script, a claim that only appears after JavaScript runs, all of it can be true and none of it can be retrieved. This page defines the term and shows why illegible facts are, for an AI engine, facts that do not exist.
Retrieval legibility is whether the facts on your page exist as readable, parseable text in the raw response a machine actually receives, because a fact a retriever cannot read is a fact it cannot embed, retrieve, or cite, however true it is.
What Retrieval Legibility Actually Is
Start with the gap that almost no one measures: the difference between the page a human sees and the page a machine receives. When you visit a site, your browser downloads the raw HTML, then executes the JavaScript, fetches data, paints images, and assembles the finished experience. Most AI crawlers, the retrievers behind ChatGPT, Perplexity, Claude, and the growing family of answer engines, do not do that work. They take the raw HTML the server returns and read the text in it. What they cannot see, they cannot use. Retrieval legibility is the measure of how much of your meaning survives that stripped-down reading.
Make it concrete. Imagine a service business with a clear, correct pricing table on its site. If those prices are set in plain text, "$180 for a standard inspection," a retriever reads them, embeds them, and can quote them when a buyer asks an AI what the service costs. Now imagine the identical prices baked into a designer's PNG. To a human the two pages look the same. To a retriever, one page states a price and the other page contains a picture of nothing readable. The fact is equally true in both. It is legible in only one.
The same fracture runs through the whole page. A phone number concatenated in a script to dodge scrapers is invisible to the retriever too. A specification split across an interactive tabbed widget, where only the first tab is in the initial HTML, loses everything behind the other tabs. A key claim that renders only after a client-side framework hydrates the page is, to a non-executing crawler, simply absent. Legibility is not about whether your facts are correct or even whether they are on the page. It is about whether they are present, in readable text, in the exact bytes a machine reads before any browser magic happens.
Watch: How Google Search indexes JavaScript sites - JavaScript SEO by Google Search Central, a clear explainer of why crawlers and browsers see different pages and what falls out of the raw HTML. Source: YouTube.
Why Legibility Decides Whether AI Can Cite You
Retrieval is the first gate in every AI answer. Before a model reasons or writes a word, it pulls passages of text, converts them into embeddings, and matches them against the question. That pipeline runs on text. If a fact never enters the pipeline as text, it is never embedded, never chunked into a retrievable passage, and never surfaced as a candidate answer. Legibility sits upstream of everything: relevance, ranking, and citation all operate only on the facts that made it into readable form.
This is why legibility is more brutal than most visibility problems. A poorly worded but readable claim can still be retrieved and improved on later. An illegible claim gets no such chance. It is not ranked low; it is absent from the ballot. When a prospective customer asks an AI engine "what does this cost" or "do they serve my area" or "how long have they been in business," the engine answers from the text it could read. If your answer lived in an image, a script, or an un-rendered component, the engine either omits you or, worse, fills the gap with a competitor whose facts were stated plainly. Truth that a machine cannot read does not lose gracefully. It simply does not compete.
How to Make Your Facts Legible
The fixes are unglamorous and mostly free. The goal is simple: every fact you want an AI to know should exist as plain, server-rendered text in the raw HTML.
Get the fact out of the image
Prices, hours, phone numbers, addresses, and named claims belong in real text, not baked into a graphic. If a fact currently lives only in an image, a PDF, or a chart, restate it as readable text nearby. A picture of a price is decoration; the price itself has to be typed.
Render the fact on the server
Content that only appears after client-side JavaScript runs is invisible to non-executing crawlers. Server-side rendering or static generation puts your key facts into the initial HTML. Frameworks like Next.js, Nuxt, and SvelteKit make this the default rather than the exception.
Test what the machine actually reads
View the page source, disable JavaScript, or run a plain curl request, then search that raw output for your critical facts. If a price or claim is missing from the raw text, it is missing from retrieval. Reinforce it with structured data so the fact is stated in an explicit, machine-first format too.
What This Means for Your Website
Retrieval legibility reframes a whole category of design decisions that used to be purely aesthetic. The elegant hero image with the offer typeset inside it, the slick single-page app that assembles content on the fly, the clever obfuscated contact details: each of these can quietly delete your most important facts from the machine-readable web. The remedy is not to make your site uglier. It is to make sure that behind every beautiful presentation there is a plain-text version of the fact a retriever can read.
Pair legibility with meaning. Legible text that is vague still retrieves poorly, which is why legibility works hand in glove with semantic clarity: state the subject explicitly, keep terminology consistent, and write self-contained passages so each readable fact also lands near the questions buyers ask. Legibility gets the fact into the machine's field of view. Clarity makes sure it is understood once it is there. You need both, and legibility comes first, because a fact that is never read cannot be misread or correctly read: it is simply gone.
An AI engine does not reward the truest page. It rewards the page whose truth it can read. If your best facts live in images, scripts, or un-rendered components, you are not ranked low: you are invisible in the exact moment a buyer is asking about you.
How AIOInsights Reads This Signal
Retrieval legibility is one of the first things AIOInsights checks, because it is upstream of everything else. Our evaluation looks at the raw content a machine receives, not the rendered page a designer sees, and asks whether your load-bearing facts survive that reading. Do prices, contact details, service areas, and core claims exist as plain, readable text? Are they present in the initial HTML rather than dependent on client-side execution? Are they reinforced by explicit structured data and organized into clean, self-contained chunks that a model like SBERT can embed cleanly? This work feeds the AI Discoverability pillar, and it connects directly to fact legibility, our finer-grained check on whether each specific fact is machine-readable.
Every one of these checks is real and deterministic. We do not guess and we do not invent a score that drifts each time. We read what a retriever would read and report, plainly, which of your facts are legible to a machine and which have quietly disappeared. Fixing them is often the fastest, cheapest gain available: the truth is already on your page, it just needs to be readable.
Check Whether AI Can Read Your FactsKeep reading the lexicon: Structured Data, Chunking, and SBERT.
See whether AI systems can actually read and cite the facts on your site.
Free Trust Check