Improve AI search readiness (GEO)
The AI search readiness tab scores whether an AI assistant can fetch each published page whole, parse it, and pull concrete answers out of it. Each page gets a score from 0 to 10 across seven checkable factors, plus a worklist of the pages most worth fixing. The score is also called GEO (Generative Engine Optimization).
The score audits the page itself. It does not measure, and does not predict, whether AI engines actually cite you. Observed AI activity is reported separately, as history.
Before you begin
Publish a site first. The tab appears only for a project with a publication target, and pages with no published version have no score.
Scores are computed at publish time. After you improve a page, publish again to see its score change. There is no re-scan control on this tab.
Open the tab
Go to Analytics in the sidebar.
Open the AI search readiness tab.
Read the readiness breakdown
The Readiness breakdown card shows the average score across all scored pages, then how many pages sit in each band:
Band | Score | What it means |
|---|---|---|
AI-ready | 7 to 10 | structured for clean retrieval and parsing |
Partially ready | 4 to 6 | parseable but with weak areas |
Hard for AI to use | 0 to 3 | too thin, oversized, or unstructured |
"AI-ready" here is a score band. In AI-ready published sites the same phrase describes the machine-readable endpoints every published site exposes. The endpoints are always there; the band is per page and earned.
Under the bands, a line reports what AI agents actually did over the last 30 days: page fetches, on-page questions, and site-index fetches of llms.txt and the sitemap. With no observed traffic it reads No AI agent activity observed in the last 30 days.
Work the Pages to improve worklist
The Pages to improve card lists every page scoring below 7. Rows are ranked by observed AI demand over the last 30 days, highest first, then by lowest score. Pages AI agents are already pulling get fixed first. With no observed demand anywhere, the order falls back to worst score first.
Each row shows:
Column | Contents |
|---|---|
Page | the page title. Choose it to open the topic in the editor. A page with no linked topic shows as plain text. |
Score | the page's score, to one decimal place. |
AI demand (30d) | total observed demand, broken down as fetches and asks. |
Biggest opportunity | the weakest factor, its value out of 10, and the fix for it. |
Work top to bottom: open the page, apply the fix for its weakest factor, then publish again. The fix shown maps directly to how that factor is scored. Five of the seven factors are measured deterministically from the page. Semantic completeness and answer density are AI judgments, rated from the page's first 2,000 characters.
When every page reaches 7 or higher, the card reports All {n} pages are AI-ready (7+). Nothing to fix.
The seven factors
Factor | Weight | How it is measured | Suggested fix |
|---|---|---|---|
Semantic completeness | 20% | whether the page answers the question its title implies (AI judgment) | Make sure the page answers the question its title implies. |
Structural clarity | 18% | heading count, paragraph length (20 to 80 words is ideal), and list usage | Add headings and a list; keep paragraphs around 20 to 80 words. |
Answer density | 17% | direct answers versus filler and hedging (AI judgment) | Lead with the answer; trim filler and hedging. |
Fetch size | 15% | Markdown size, full marks up to 50,000 characters and sliding to 1 at 100,000 | Split this page into focused topics under ~50,000 characters — it is large enough that AI agents truncate it mid-read. |
Schema potential | 10% | the topic type, where task, how-to, FAQ, troubleshooting, and tutorial score highest, reference, concept, and overview score mid, and no type at all scores lowest | Set a topic type — task, how-to, or FAQ score highest. |
Internal linking | 10% | cross-reference count, where 4 to 8 links scores highest | Add cross-references to related topics (4 to 8 is ideal). |
Concrete evidence | 10% | numeric specifics against vague quantifiers, per 100 words | Replace vague quantifiers with exact values — numbers, limits, defaults, versions. |
A weak factor also decides the action a health scan raises for the page. An oversized page is asked to be split, a thin page to be expanded, and the rest to be updated, so the advice on this tab and the item in Prioritize what to write next always agree.
What this tab does not do
It has no refresh, re-score, or export control. Publishing is the only way to a new score.
It reports observed AI activity as history. It makes no claim about search rank, traffic, or citations.
See also
Triage content gaps: find content readers want but cannot find
AI queries and crawler activity: what readers ask and which AI crawlers ingest your docs
AI-ready published sites: the endpoints AI engines read
Published site SEO: schema.org markup and canonical URLs