GEO score checker

What the score measures

The score is the weighted sum of 14 checks across three levers, each weighted by how much failing it actually costs you a citation. A check that cannot be measured is reported as not checked and removed from the denominator rather than scored as a zero.

Retrieved: 25 points

Whether a crawler can fetch and parse the page at all. This is table stakes, and it is also where the fastest wins live.

• AI crawler access (8): whether robots.txt allows the search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot) and eleven other AI tokens, weighted so a blocked search crawler costs more than a training crawler you opted out of.
• Content in initial HTML (7): whether the main content exists before JavaScript runs.
• Indexable and accessible (6): no noindex, a self-referencing canonical, no paywall between the crawler and the answer.
• llms.txt (4): the smallest check in the model, deliberately.

Ranked: 15 points

Whether the page reads as a source worth trusting. An engine resolves a page to an entity before it decides whether to quote it.

• Entity and authorship clarity (7): Organization or Person schema,sameAs links that corroborate it, a named author, a link to an About page.
• Freshness (4): a date a reader can see, and dateModified in the schema.
• Topical focus (4): one H1 that agrees with the title, and a real meta description.

Selected: 60 points

Whether there is a passage an engine can lift. This carries the most weight because it is where almost every page loses, and because it is the hardest thing to fake.

• Answer-first structure (13): the heaviest single check. Does each section answer its own heading in the first sentence?
• Chunk extractability (10): paragraphs of 15–60 words that stand alone, without a bare “this” or “it” pointing at something that got chunked away.
• Heading hierarchy (9) and scannable formatting (9): descriptive headings, no skipped levels, lists and tables instead of enumerating prose.
• Citable specifics (8): figures, dates, prices, named tools. Answer engines quote specifics and skip adjectives.
• Q&A structure (7) and structured data (4).

How to read the result

Work top-down and stop when the effort stops paying. The result is already ordered for that: blocking issues first, then the fix list sorted by points recovered with cheaper fixes breaking ties.

1. Blocking issues, if any. A noindex or a blocked OAI-SearchBot costs few points but makes the page uncitable outright, so it gets a banner rather than a deduction.
2. The fix list. Five items at most. Each names the points it recovers and how much work it is.
3. The full breakdown. Every check, with the evidence behind its verdict: which sections bury their answer, which crawlers are blocked, how many paragraphs run long.

The fix prompt button copies all of that as instructions for a coding agent. It carries the evidence, not just the advice, which is the difference between “rewrite the opening of these three named sections” and “improve your structure”.

What this checker cannot tell you

It cannot tell you whether you will actually be cited. Citation depends on the query, the competing sources, and each engine’s own ranking, none of which any tool can see from outside. What it measures is whether the page is eligible: retrievable, attributable, and structured so an answer can be lifted from it.

Two more honest limits:

• Some checks are heuristics and say so. Render-dependency infers from the HTML rather than measuring a render. The summary text names it when a verdict is inferred.
• llms.txt has no proven impact. A 300,000-domain study found no correlation between having one and being cited, and no major vendor has committed to reading it. It is worth 4 points out of 100 because it is free and harmless, not because it works.

Where to go next

If the checker flagged crawler access, the AI crawler checker shows the exact robots.txt lines to change. If it flagged structured data, the schema generator builds valid JSON-LD you can paste in.