Clearscope

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Build me the content grader I actually need instead of Clearscope.

Same category as any content optimiser, and the same honest constraint: **the term model is only as good as the corpus behind it, and you do not have their corpus.** What you can do is make the analysis transparent, which theirs is not.

STACK
- Node 20+ with Fastify, server-rendered HTML with a little vanilla JS
- SQLite through better-sqlite3, WAL mode
- Caddy in front

THE DATA MODEL
- topics: id, keyword, locale, sources_json, built_at
- terms: id, topic_id, term, idf, coverage, mean_count — coverage is the share of ranking pages containing it
- drafts: id, topic_id, title, body_md, grade, graded_at, snapshot_md
- grades: append-only, so an editor can see a draft improving rather than one number replacing another

BUILDING THE TERM MODEL
- Take the top ranking pages for the topic — pasted, or fetched where permitted — extract the main content, and compute term statistics across the set
- Include multi-word phrases up to three words; single words alone miss most of what a topic model is for
- IDF against a background corpus you keep locally; without it every page scores highly on 'the'
- **Store coverage per term, not just frequency.** A term in nine of ten ranking pages is a topic requirement; a term appearing forty times in one page is that page's obsession

GRADING, TRANSPARENTLY
- The grade is coverage-weighted: what share of the high-coverage terms does the draft use at all
- **Show the arithmetic.** The whole reason to build this rather than buy it is that a writer can see why the grade is what it is, instead of chasing an opaque letter
- Cap the contribution of any single term, so repeating one word cannot raise the grade. Otherwise you have built a keyword stuffing machine with a scoreboard
- Report the grade with the sample size behind it. Ten pages is a weak model and the interface should say so

THE WRITING SURFACE
- A plain markdown editor with the term list beside it, used terms struck through, and the grade recomputed on a debounce
- Group terms by coverage: essential, common, occasional. A flat alphabetical list is unusable
- A word count with the observed range from the ranking set, as a range

WHAT TO SAY OUT LOUD IN THE INTERFACE
- **A high grade is not a ranking.** It means the draft covers what the ranking pages cover, which is a floor and not an edge. The thing that ranks is the part none of them have
- Never let the tool suggest inserting a term into a sentence. That is how text becomes unreadable

WHAT MATTERS MOST
Visible arithmetic and a per-term cap. Transparency is the only advantage a homemade grader has over a bought one, and the cap stops it rewarding the wrong thing.