AI Readiness Score
Rates your highest-traffic pages Strong, Moderate or Weak on how easily a generative engine can lift a citable answer out of them. Assessed by code, not AI — reproducible, and free to run.
What it measures
Every week we take your top 20 pages by Search Console clicks, fetch each one (respecting robots.txt, ~1 request per second), and score eight factors. The three heaviest are the content interventions measured as most effective in GEO: Generative Engine Optimization (Aggarwal et al., KDD 2024); the rest cover how cheaply a passage can be extracted.
| Factor | Weight | What counts as met |
|---|---|---|
| Cites external sources | 18 | 3+ distinct external domains linked from body content. Social profiles, share buttons, link shorteners and app-store links are excluded — they are navigation, not evidence |
| Contains concrete statistics | 18 | 5+ figures with a unit, percentage or currency |
| Includes quotations | 12 | At least one quoted block |
| Question-form headings | 14 | 2+ headings phrased as the questions people ask |
| Answer-shaped structured data | 14 | FAQPage, QAPage or HowTo (Article scores half) |
| Lists or tables to lift from | 10 | 6+ list items, or any table |
| Author and date visible | 8 | Both an author and a published/modified date |
| Answers before it digresses | 6 | Under ~120 words between the H1 and the first section |
Why a band, not a number
We report Strong / Moderate / Weak rather than a score out of 100. The factor list is grounded in published research, but the weights and thresholds are not — we chose them because they looked reasonable, and we have never calibrated them against whether higher-scoring pages actually get cited more often. A two-significant-figure score would imply a resolution we have not demonstrated, and would make “41 vs 46” look like a real difference when it is an artefact of where a threshold happens to sit.
The individual findings do not depend on the weighting. “This page has no statistics, no quotations and no FAQ schema” is a true, checkable statement however you score it — so that is what the report asks you to act on.
How the site band is calculated
Pages are combined into one band using an average weighted by Search Console clicks, so a weak page nobody visits does not count as heavily as a weak page earning thousands of clicks. That weighting needs first-party Search Console data, which is why a standalone AI-visibility tool cannot produce it. Strong is the top band, Moderate the middle, Weak the lowest.
Two adjustments worth knowing: a page under ~300 words is marked thin and can never rate Strong however well structured it is — there is simply little for an engine to quote. And a page returning 4xx/5xx is flagged Unreachable rather than Weak, because an unreachable page among your top URLs is a different problem.
What we deliberately do not claim
This is a checklist, not a prediction. We do not claim that improving these factors will get you cited.
C-SEO Bench (Puerto et al., NeurIPS 2025) evaluated nine conversational-SEO methods across six domains and found most were ineffective or actively harmful to ranking, with a traditional retrieval-ranking baseline measuring 7.6× more effective in retail. It also found gains shrink as more sites adopt the same tactics. We report these as signals because they are cheap to fix and grounded in published work — not because they are guaranteed to move anything.
How we crawl
- Top 20 pages by clicks only — never a full-site crawl.
robots.txtis fetched and obeyed before every request.- Roughly 1 request per second to a single host, identified as
MetricsTab-Bot. - Once a week. A full run is about 20 seconds of load on your server.