Verified against Claude · 2026-07-23
Draft a full post from an outline without losing the point of view
Turns an approved outline plus real source material into a full draft that argues something specific and keeps one throughline, flagging where the source material ran short instead of inventing facts to fill the gap.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are drafting a full blog post from an approved outline. Your job is to write the argument the outline sketched, not just fill in prose under each heading — every section should advance the same point of view, and nothing in the draft should be invented to cover a gap the source material didn't actually fill. APPROVED OUTLINE H1: Why Your Onboarding Emails Are Getting Ignored. H2s: The 3 seconds you actually get, What "personalized" actually means in an inbox, The one metric that predicts unsubscribes, A 4-email sequence that held a 61% open rate. SOURCE MATERIAL Internal case study: client X redesigned their 4-email onboarding sequence in March, open rate went from 22% to 61% over the next quarter, unsubscribe rate dropped from 4.1% to 1.3%. No data on click-through rate available. POINT OF VIEW / THESIS Most onboarding sequences fail because they front-load feature explanations instead of answering "why did I sign up" in the first email. AUDIENCE B2B SaaS marketers who already run an onboarding sequence and are looking to fix underperformance, not build one from scratch. TONE REFERENCE "Most welcome emails read like a table of contents. Nobody signed up to read a table of contents — they signed up because something specific was broken." (short, declarative, contraction-heavy) TARGET LENGTH 1,600-1,800 words DRAFTING RULES Follow the outline's heading order and structure unless a specific problem with it surfaces while drafting — if you deviate, say so explicitly rather than silently reorganizing. Every factual claim, statistic, or specific example in the draft must trace back to something in the source material; if a section in the outline needs a supporting fact the source material doesn't contain, write the section around what you do have and flag the gap explicitly rather than inventing a plausible-sounding statistic, example, or quote to fill it — a fabricated detail that reads smoothly is a worse outcome than an honestly incomplete section, because the incomplete section gets caught before publish and the fabricated one usually doesn't. Keep the stated point of view as the connective tissue running through every section, not just something mentioned in the intro and abandoned — each section should either advance the thesis, provide evidence for it, or address a complication to it; a section that could be dropped into a completely different post arguing the opposite thesis without needing a single word changed is a sign the point of view didn't actually make it into that section. Match the tone reference at the sentence level — its actual sentence length pattern, contraction use, and directness — not just an adjective summary of it; two drafts both aiming for "conversational but authoritative" can sound completely different depending on how that gets executed, so anchor to the actual reference text, not the label. Vary sentence length and paragraph rhythm deliberately; a draft where every paragraph runs three sentences of similar length and structure is one of the most identifiable signs of an unedited AI draft, and it reads as monotonous even when every individual sentence is fine. End sections with a bridge into what comes next rather than a hard stop that reads as the model finishing a checklist item and moving on — the post should read as one continuous argument with headings as signposts, not eight independent mini-essays stapled together. OUTPUT FORMAT 1. The full draft, formatted with the outline's headings. 2. A short list of every place a fact, example, or statistic was needed but the source material didn't support one — state exactly what's missing. 3. One sentence confirming which single thesis the whole draft argues, so an editor can check it against every section directly.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Requiring every fact to trace back to the source material, with an explicit flag instead of an invented fill-in, targets the specific failure mode that makes AI-assisted drafting risky for publishable content: a model asked to write a complete-sounding section will produce a plausible statistic or example to cover a gap rather than leave visible white space, and a plausible fabrication is strictly worse than an honest gap because the gap gets caught in editing while the fabrication often survives all the way to publish, where it becomes a factual claim under the publication's name. Treating the point of view as connective tissue rather than an intro-only mention addresses a specific and common defect in AI-drafted long-form content: sections that are individually competent but collectively directionless, because each one was generated somewhat independently against its own heading rather than against the running argument — the test named in the prompt, whether a section could be lifted into a post arguing the opposite thesis unchanged, is a concrete, checkable signal of exactly this failure rather than a vague appeal to "cohesion." Anchoring tone to an actual reference passage instead of an adjective label works because voice is a function of concrete, countable choices — sentence length, contraction frequency, how often a sentence starts with a conjunction — and two people (or two model runs) both told to write "conversational but authoritative" will produce measurably different prose, while both shown the same reference passage converge much closer to the same register, because there's now something specific to pattern-match against instead of an adjective every writer interprets differently. The explicit instruction to vary sentence and paragraph rhythm targets a documented, easily-spotted tell of unedited AI prose — a run of same-length sentences and same-length paragraphs that reads as metronomic rather than written by someone actually thinking through an argument — and naming it directly, rather than trusting general "good writing" instructions to prevent it, catches a pattern that otherwise survives to the final draft because no single sentence in isolation looks wrong; the problem is only visible in the aggregate rhythm across a paragraph or page.
Verified against
Claude Claude Sonnet 5 · 2026-07-23
ChatGPT GPT-5.1 · 2026-07-29
Changelog
- 2026-07-23 — Initial publish, verified against Claude (Claude Sonnet 5) and ChatGPT (GPT-5.1).
Need this built into your business?
If a prompt isn't enough — SEO, built and maintained for you — that's Scult's day job.
EXPLORE SEO
