SEO & GEO Readiness Scorer

Free tool by AAJ — score any URL the way an AI agent actually reads it

Paste a public URL and get a 100-point score, a letter grade and a prioritized fix list covering whether an AI agent can reach your page, read it without running JavaScript, and understand it well enough to quote. Fourteen measured checks across three stages, every weight published. The on-screen report is free and ungated.

Use the tool at aajconsult.com/tools/seo-geo-readiness-scorer. Every check, its weight and the changelog are published at the scoring standard.

Why the score is a funnel and not a checklist

Most SEO audits hand you a flat list of findings and let you guess which ones matter. This one is ordered, because the failures are ordered: an agent must reach a page before it can read it, and read it before it can understand it. A perfect FAQ schema on a page ChatGPT cannot fetch is worth nothing, and a scoring model that averages the two together will tell you that you are doing fine.

So the three stages are weighted by what they gate. Reach is worth 39 points, Read 25, Understand 36. And a weak early stage caps the whole score rather than being averaged away: if Reach comes in under 60%, the total is capped at 40 no matter how good the rest is. If Read comes in under 50%, the cap is 55. Fail both and the cap is 35. The cap is the point — it stops a site from collecting credit for structured data that no agent will ever see.

Reach — 39 points

An AI agent can fetch the page (20.5). The single heaviest check. The scorer requests the page the way an AI crawler does and records what comes back. If that fails, nothing downstream is measurable.

Retrieval bots allowed in robots.txt (9.9). Which named AI retrieval agents your robots.txt actually permits. This is the most common quiet failure: a robots.txt written for search engines that blocks, by omission or by a blanket rule, the crawlers that feed AI answers.

Missing pages return a real 404 (4). Soft 404s — a "not found" page served with a 200 status — teach a crawler that every URL on your domain is valid, which poisons what it decides to keep.

Time to first byte (3) and sitemap and robots hygiene (1.6). Slow first bytes cost you fetches under a crawl budget; a stale or unreferenced sitemap costs you discovery.

Read — 25 points

Server-rendered content volume (15). How much of your actual content exists in the HTML before any JavaScript runs. This is where client-rendered sites lose most of their score, and it is usually a surprise: the page looks complete in a browser and arrives nearly empty to anything that does not execute scripts.

On-page fundamentals (4). Title, meta description, headings — the basics, still worth points because they are still what gets parsed first.

Token efficiency — markup tax (3). The ratio of markup to text. A page carrying forty times more div soup than prose is expensive for a model to read, and expensive pages get truncated.

Markdown content negotiation (3). Whether your server offers a clean Markdown variant to clients that ask for one. This is the forward-looking check in the Read stage — the cheapest way to hand an agent your content without the markup tax.

Understand — 36 points

Answer-shaped content (12). Whether the page states an answer in a form a model can lift: a question-shaped heading followed by a direct answer, near the top, in prose rather than implied by a layout.

Structured data quality (10). Not merely present — correct. Valid JSON-LD with resolvable entity references, not a stack of schema blocks that reference nothing.

Citability signals (10). What makes a model willing to name you as the source: specific figures, attributed quotes, outbound citations to primary sources.

Freshness and authorship (2) and llms.txt (2). A visible date and a named author, and whether you publish an llms.txt content map.

Measured, assessed, and what the score refuses to include

Only measured checks score. Where something is a judgement rather than a measurement, the report says so and reports it separately instead of quietly folding an opinion into a number. Each finding is also tagged by effort — a quick fix or a project — so the fix list sorts by what you can ship this week against what needs planning.

The scorer also states its own limits on the report: it measures one HTTP response from one location. It does not measure model behaviour, ranking, or how likely you are to actually be cited. Anyone selling you a tool that claims to measure the last one is selling you a guess.

Grades

90 and above is an A: agents can reach the page, read it without JavaScript, and parse it into a grounded answer. 80 is a B, 70 a C, 55 a D, and below 55 an F — not ready. Because of the stage caps, a site with excellent content and a fetch problem lands in the F band, which is the honest answer even when it is not the flattering one.

Frequently asked questions

What does the scan actually request?

The public URL you enter and its public robots.txt, sitemap.xml and llms.txt — the same files any search crawler reads. It needs no access to your analytics, your server or any private data, and it does not store the page content.

Is it free, and is there a signup?

The scan and the full on-screen report are free with no account. The downloadable Markdown report asks for an email; the report you can read on the page does not.

Why did my score get capped?

Because an early stage failed badly enough to make the later ones moot. The report names which stage and what it scored. Fix that stage first — it is the only work that raises the ceiling.

My site scores well in Google. Why is this score low?

Because they measure different things. Googlebot renders JavaScript; most AI retrieval agents do not, and several of them respect a narrower robots.txt allowance. A site can rank well and still be invisible to the systems answering your buyers' questions.

Where do the weights come from, and do they change?

They are published in full on the scoring standard page, with a versioned changelog. When a weight changes, the standard changes with it — the page and the engine are checked against each other on every build.

Related resources

Agent Readiness Quick Check — AI Answer Share of Voice — GEO & AEO Playbook — AI Search hub

A score tells you what is broken. Someone still has to fix it.

Most of the heavy checks — server-rendered content, answer-shaped copy, structured data that resolves — are engineering and editorial work, not settings. AAJ does that work. Book a Growth Audit.

Run this in your agent

The same method runs as the free AAJ agent skill SEO, GEO & AEO Audit, so a coding agent can audit your pages from inside your own repo. Install it with npx skills add sarojkjha/aaj-marketing-skills --skill seo-geo-aeo-audit.