Can a prompt replace NeuronWriter?

Build a content editor that scores a draft against the pages currently ranking for a target query, with the formula fully visible.

Corpus collection: for a target query, fetch the top 10 organic results through a commercial SERP API behind a swappable provider module, never by scraping. Then fetch each result page politely, honouring robots.txt, and extract its main content — strip navigation, footer, cookie banners and sidebars with a boilerplate-removal pass, because scoring against a competitor's navigation menu produces nonsense. Store the corpus with its collection date; a corpus older than a few weeks is stale and the interface should say so.

Term extraction: from the corpus, extract unigrams, bigrams and trigrams, drop stopwords, lemmatise, and compute for each term: the number of corpus pages containing it (document frequency), and the median count per page among the pages that use it. Rank terms by document frequency first — a term nine of ten pages use is a stronger signal than one that appears thirty times on a single page.

The editor: a plain writing surface with the term panel beside it, updating as the user types, debounced. For every term in the target list, show the median usage across the corpus, the draft's current count, and a state: missing, under, in range, over. Group terms into essential (in eight or more corpus pages), recommended (five to seven), and optional. Clicking a term highlights every occurrence in the draft.

The score: a single 0-100 number, and immediately beneath it, the arithmetic — how many essential terms are covered, how many recommended, how far the draft's length sits from the corpus median, and the weight of each component. Let the user adjust the weights. A score whose derivation is hidden is a score nobody can argue with, and arguing with it is exactly what a writer should be doing.

Guard against the obvious failure: flag any term used more than twice the corpus median as over-used, and show a warning that writing to satisfy a term list produces text nobody wants to read. The tool measures coverage; it does not certify quality.

Headings: extract the corpus's H2 and H3 texts, cluster them, and present the common subtopics as an outline suggestion, with the source pages listed for each.

Export: the draft as Markdown, and the term list plus scores as CSV.

Out of scope: generating or rewriting text with a model, rank tracking, backlink data, and CMS publishing.