SMF·SCALENUT
Can a prompt replace Scalenut?
AI writing — AI drafting, research and content optimization
Exhibit tracking slip
Verdict
Scalenut's Cruise Mode is a pipeline: keyword, then research, then outline, then draft. Building it as a pipeline with an approval gate at each stage is straightforward and produces a much better result than one long prompt, because the errors compound otherwise. The two things it cannot have are the SERP data that seeds the research — a paid API call per keyword — and the topic clustering built on a keyword database nobody outside a data vendor owns.
Exhibit A — The prompt
Received on31.07.2026Build a staged keyword-to-draft pipeline with an approval gate between every stage. Never run end to end unattended.
Stage 1 — Research. Given a target keyword, fetch the top 10 organic results through a commercial SERP API behind a swappable provider module, never by scraping. Fetch each page politely, strip boilerplate, and extract: the H2 and H3 headings, the word count, the questions asked (sentences ending in a question mark or beginning with a question word), and any statistics (sentences containing a number with a unit or a percentage) with their source page. Present this as a research digest, not a summary — the raw headings and questions are what the next stage needs.
Gate: the user reviews the digest, removes irrelevant competitors, and approves.
Stage 2 — Outline. Cluster the collected headings by embedding similarity and label each cluster. Propose an outline from the clusters, ordered by how many competitors cover each and in what position. Attach to each proposed section the competitor headings it came from and the questions that belong under it. The user edits freely — reorder, rename, delete, add — and approves. The outline is the artefact that most determines the final draft, so make editing it fast.
Gate: approve the outline.
Stage 3 — Draft. Generate section by section, not in one call. Each section's prompt receives: the article's target keyword and audience, the outline entry with its own questions, the two adjacent section headings for continuity, and the text of the previous section already accepted. Generate one section, show it, let the user accept, regenerate or edit, then move to the next. Section-by-section with the previous accepted text in context is what stops the repetition and drift a single long generation always produces.
Stage 4 — Review. Assemble the draft and report: which of the collected questions it answers and which it does not, which statistics it used and whether each is attributed to its source page, word count against the competitor median, and heading structure against the approved outline.
Running state: every stage's output persisted, so a pipeline can be resumed days later. Nothing is regenerated silently — a rerun of any stage is explicit and its previous output is kept.
Citations: every statistic carried into the draft keeps a link to the page it came from, and the review stage flags any number in the draft that cannot be traced back to the research. A model inventing a plausible statistic is the most damaging failure mode here and the pipeline should catch it rather than trust it.
Use Anthropic or OpenAI, whichever key is configured. Store everything in SQLite.
Out of scope: keyword volume or difficulty from any source, topic clustering across a keyword database, rank tracking, and CMS publishing.Opening prefills the prompt — press enter to run it.
Exhibit B — What you lose
- B.1 the SERP and keyword data, which remains a per-query cost
- B.2 topic clustering across a keyword universe you do not have
- B.3 the content-score model tuned against real ranking outcomes
- B.4 publishing straight into a CMS
Exhibit C — Why people still pay: workflow, data, and model tuning
Because the pipeline is only as good as what seeds it, and the research stage runs on datasets that cost more to license than the subscription costs to buy.
Questions
Can I import my Scalenut projects?
Drafts export as text and documents, so finished work comes across. Briefs and research digests do not export in a reusable form, and re-running the research for a keyword costs one SERP call, so there is little reason to migrate anything in progress.
Why gate every stage instead of running it end to end?
Because errors compound. A research stage that picked up two irrelevant competitors produces a skewed outline, which produces a draft about the wrong thing, and you discover it after generating three thousand words. Approving the outline takes two minutes and saves the whole run.
What does it cost to run?
One SERP call per article, a fraction of a cent, plus generation — section by section on a long article comes to perhaps ten to thirty cents. A dozen articles a month is a few dollars against a forty-nine dollar plan.
What is the one thing that does not survive the rebuild?
Knowing what to write about. Scalenut clusters a whole keyword universe and tells you which topics to cover in what order; this pipeline starts from a keyword you already chose. The strategy layer sits on data you cannot buy at this scale.
Related tools
Receipt