// The metric · Programmatic SEO & GEO
The Retrieval Readiness Score
One 0 to 100 number for the question every tracker skips: before AI decides whether to cite you, can it even find, read, and understand you?
// Why this exists
Everyone measures what AI says. Almost nobody measures whether AI can reach you.
Everyone measures how often AI mentions you. That's the effect. RRS measures the cause: whether AI can reach, read, and resolve you. So you can fix it.
// What it measures
Nine dimensions, one score.
Four are judgement calls, each backed by its own free diagnostic. Five more come off a single live fetch of your page.
Access
weight 18Can AI crawlers even reach you?
The gate. If GPTBot, ClaudeBot, Google-Extended, PerplexityBot and the rest are blocked in robots.txt or by a managed firewall rule, nothing downstream matters. You cannot be retrieved, so you cannot be cited.
Assessed by AI Crawler Audit →Extraction
weight 16Can a clean answer be lifted from your page?
Retrieval works at the passage level, not the page level. If your content only makes sense read top-to-bottom, a model can't lift a self-contained chunk into an answer. We score how cleanly your page fragments into retrievable pieces.
Assessed by Chunk Previewer →Answerability
weight 16Does the content actually answer the question?
Generative engines favor passages that state the answer plainly and early, backed by quotable specifics: statistics, named sources, direct claims. We score how directly your page answers the questions your market asks.
Assessed by Answerability Score →Entity Resolution
weight 14Does the model know who you are?
Models synthesize from entities, not strings. If your brand, product and category aren't defined clearly and consistently, the model can't confidently attribute a claim to you, and hands the citation to someone it does understand.
Assessed by Entity Check →Measured on the page itself
- Structureweight 12 Tables, question-form headings, plain definitions, statistic density and FAQ schema. These are the features that make a passage liftable as an answer rather than merely readable.
- Freshnessweight 8 A visible publication or updated date, and a machine-readable one. Undated pages are treated as stale, and staleness is the single best evidenced reason a page stops being quoted.
- Token costweight 8 The bytes an engine must ingest to reach your answer. Between a cheap retrieval and an expensive one the spread is roughly 4.7x, and every extra thousand tokens is another reason to route around you.
- Timeweight 8 Measured response time on a live fetch. Slow responses do not fail outright, they lose ties against a faster source that says the same thing.
- Query alignmentweight 15 Optional, and only scored if you supply the buyer question the page should win. We compare your headings against it. A page can be about a topic without answering the question, and that gap is invisible until you name the question.
And three gates, which cap rather than count
- Retrievable AI crawlers are not blocked in robots.txt. If they are, the score is capped: you cannot be cited if you cannot be fetched. Trigger: Access scores below 50. Capped near the Access score itself.
- Readable without JavaScript The served HTML carries real text before any script runs. A page that renders to an empty shell looks fine to a human and blank to a cheap crawl. Trigger: No extractable text in the served HTML. Capped at 25, or 45 for a near-empty shell.
- Same content on mobile The mobile document is not materially thinner than the desktop one. Where it is, part of your answer is invisible to a large share of fetches. Trigger: Mobile document under 60% of the desktop text. Capped at 65.
The weights above sum to 115 rather than 100 on purpose. Query alignment carries a weight of 15 but only scores when you supply a buyer question, so most scans never spend it. Whatever was measured is renormalised over its own weights, which is why a domain that returns four dimensions gets a score built from four weights rather than four scores and four zeros, and why a partially measured score is not comparable with a fully measured one.
// Entity resolution, visualized
A brand the model recognizes as a thing.
Models synthesize from entities, not strings. The more your brand resolves (in Wikidata, the Knowledge Graph, your own schema, and corroborating mentions), the more confidently AI can attribute a claim to you, and cite you.
// How it's scored
Transparent by design.
- Each dimension is scored 0 to 100 from the same checks that power our free tools: no black box.
- Gates cap, they don't average. Blocked crawlers, a page that needs JavaScript to show any text, or a mobile version missing most of the content are failures nothing downstream can compensate for, so each one puts a ceiling on the total.
- Nothing unmeasured is scored as passed. If we can't read a dimension we say so and leave it out, and if fewer than three of the four judgement dimensions came back you get a partial scan with no headline number rather than a flattering one built from half a rubric.
- Weights are relative, not a percentage of 100. Every weight is published beside its dimension above: Access carries the most at 18, Freshness, Token cost and Time the least at 8. Whatever was actually measured is renormalised, so a missing dimension neither counts as zero nor quietly reweights the rest.
- Citation is measured separately. Whether you actually get cited is the outcome. We track that with Citation Gap and Answer Monitor. RRS measures readiness; those measure results.
See where you stand.
We benchmark Retrieval Readiness for crypto and fintech brands first, but the score works for any site. Book a free audit and we'll run every dimension on your domain and hand you the number, with the fixes ranked.
Get your Retrieval Readiness Score →