A prompt that produces a full blog post instead of a vague outline must specify four things before the model writes a word: the audience, the angle, the structure, and the constraints — and it must ask for a draft, not a plan.
Most people get an outline because they asked for one, or because they gave the model a topic and nothing else. The fix is a short written brief you paste in front of every request.
The brief
A brief is five lines. Line one names the reader ("freelance designers who invoice hourly"). Line two states the angle ("why hourly billing punishes fast workers").
Line three lists the sections you want, in order. Line four sets length and tone ("900 words, plain English, no jargon"). Line five names what to avoid ("no bullet-point lists, no 'in today's world' openings").
That is it. You are not writing a novel — you are removing the model's freedom to guess. When the model has to guess your audience, it defaults to the broadest possible reader, and broad readers produce broad writing.
Why vague prompts drift
Language models predict the most likely next chunk of text given what came before. If your prompt looks like a topic heading, the most likely thing that follows a topic heading is an outline. That is the statistical shape of the input.
So "write about remote work productivity" reliably returns a numbered list of subtopics, because that is what thousands of similar strings precede. Give the model a different shape — a brief with a named reader and a stated angle — and the likely continuation changes. This is also why asking for "a blog post" and then complaining you got an outline is unfair: an outline is a legitimate interpretation of an ambiguous request.
According to our AI tool database, the current generation of assistants handles very long inputs — ChatGPT's GPT-5.5 and Anthropic's Claude Opus 4.8 both carry a 1M-token context window — so length is not your constraint. Specificity is.
Outline before draft
Do it in two passes, not one. First ask for the section headings only, based on your brief. Read them.
Fix the ones that are wrong. Then paste the approved headings back and say "now write section two, 200 words, in the voice described above." Working section by section gives you a checkpoint every few hundred words instead of discovering at the end that the whole piece drifted.
It also sidesteps the most common failure: the model writes a strong introduction, gets bored, and turns the back half into filler. A concrete illustration: suppose you are writing a help-centre article on resetting a two-factor authentication device. Your brief names the reader as "a customer who just got a new phone and cannot log in," the angle as "do this in five minutes without calling support," the sections as "what happened, what you need, the steps, if it still fails," and the constraint "no screenshots, no jargon."
A vague prompt on the same topic returns a generic explainer about 2FA. The brief returns something a frustrated customer can actually follow, because every sentence has a job.
Where this fails
A brief does not fix a topic nobody wants to read, and it does not make the model accurate. If your brief asks for a statistic, the model may supply a confident number that does not exist — the brief controls shape, not truth. Every factual claim still needs checking against a real source before you publish.
The method also costs time up front: writing five good lines takes longer than typing a topic, and on a throwaway internal note that trade may not be worth it. Long briefs can backfire too. Past a certain point the model starts treating your constraints as suggestions and quietly drops the ones it finds inconvenient, so keep the brief tight and re-state the two or three rules that matter most in each follow-up message.
And if you are working with a tool whose pricing or plan limits you are unsure about, check the vendor's own page — those details change often. This approach is a drafting method, not a publishing method. It gets you a structured first draft; you still own the editing, the fact-checking, and the judgement about whether the piece deserves to exist.