← Back to the grader

239 real landing pages, graded.

The hero copy of 239 well-known landing pages — YC companies, dev tools, SaaS, AI products — extracted and scored 0–100 by the deterministic grader built here. No LLM, same input → same score, always.

The single biggest finding: 195 of the 239 pages — 82% — contain no number anywhere in the hero. Not a rounded marketing figure, not a customer count, nothing. The em-dash panic is a weak tell by comparison. If a page reads generic, that's usually why.

Hero copy extracted on 2026-07-24 by a plain no-JS fetch (first h1 + first subhead + first CTA) · deterministic scoring · full method, script and CSV in the MIT repo. Scores judge the extracted hero text against fixed heuristics — not the product, not the whole page.
Scored with static-fetch-regex-v1, the rule set of 2026-07-24. Three rules were tightened on 2026-07-25: a digit inside a name, a version or a year is no longer a quantified claim; a button arrow is no longer an emoji; an acronym such as SQL or CLI is no longer shouting. Re-scoring this corpus with the current rule would move 42 of the 239 rows — caps 33 → 6, emoji 7 → 2, nonum 195 → 203, median 79 → 83.

That is not an erratum and these rows are not stale. A dataset is scored by the rule of its date; re-scoring an archived row with a later rule produces a different dataset, not a correction. The grader applies today's rule, so pasting a row from this table into it can legitimately return a different score — that is the rule change doing its job, but you should know which of the two you are running.

Where does your page land?

Paste your headline, subhead and CTA — instant, nothing stored. The grader applies the current rule set, so your score and this table are not scored identically; see the note above.

● Grade my landing page Dataset + method (MIT) →