Surfer SEO

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 optimisation tool I actually need instead of Surfer SEO.

Be clear about the honest bit: **the scoring is simple statistics, and getting the search results is the hard part.** Everything below is arranged around that.

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
- queries: id, keyword, locale, device, fetched_at
- results: id, query_id, position, url, title, fetched_at, html_snapshot_id
- pages: id, url, word_count, headings_json, terms_json, fetched_at
- drafts: id, query_id, title, body_md, score, scored_at
- Snapshots kept, so a score computed today is reproducible next month

GETTING THE RESULTS, WHICH IS THE CONSTRAINT
- Search engines block automated queries. **Say this out loud rather than writing a scraper that works for a week and then silently returns nothing**
- The workable options, in order: paste the top ten URLs by hand for the query you care about; use a search engine that offers an interface for this; or fetch from an engine that permits it
- Whatever the source, cache aggressively and refetch on a stated interval. This data changes slowly
- If a fetch fails, the tool must say the analysis is stale rather than score against nothing

THE ANALYSIS, WHICH IS ARITHMETIC
- Extract main content from each ranking page — readability-style extraction, not the whole HTML, or your term model is full of navigation
- Term frequencies across the set, TF-IDF against a general corpus to find what is distinctive to this topic
- Median word count, heading depth and count, and the questions that appear as headings across several results
- **Report ranges, not targets.** 'Between 1,400 and 2,300 words' is true; '1,847 words' is false precision that makes writers do stupid things

SCORING A DRAFT, CAREFULLY
- Coverage of the distinctive terms, structure against the observed range, and the questions answered
- **Never score on keyword density.** It has not been a ranking factor for a very long time and optimising for it produces text no human wants to read, which is the actual thing search engines measure now
- Show what is missing as suggestions with the evidence — 'six of the ten ranking pages have a section on pricing' — rather than as a number to chase

WHAT TO REFUSE TO BUILD
- An automatic writer that fills the gaps. It produces exactly the thin, over-optimised content that recent search updates were built to remove, and it will cost you the traffic you are trying to get

WHAT MATTERS MOST
Ranges instead of targets, and evidence behind every suggestion. This category's failure mode is writers following a number off a cliff.