An effective blog-post prompt has seven parts — role, audience, topic, source material, tone, length, and output format — and you write them as plain labelled lines in one message, not as a clever sentence.
Here is the template you can copy and fill in: "You are a [role, e.g. freelance tech writer]. Write a blog post for [audience, e.g. small-business owners who have never used AI]. Topic: [one sentence]. Use only these facts: [paste your notes, links, or quotes].
Tone: [e.g. plain, friendly, no hype]. Length: [e.g.
900 words]. Format: [H2 subheadings, intro, 3 sections, FAQ of 3 questions, no bullet lists]." That single block gives the model everything it needs, and it takes about two minutes to fill in.
The rest of this answer explains why each line matters, shows a filled-in example, and covers the one line most people forget.
The mechanism is simple: a language model fills whatever gaps you leave with the most statistically likely filler. Leave out the audience and it writes for a generic reader. Leave out the source material and it invents plausible details.
Leave out the format and you get a wall of text you have to restructure by hand. The role line sets vocabulary and level of expertise. The audience line sets how much you explain versus assume.
The source line is the one that keeps the piece honest — when you paste your own notes, quotes, or links and tell the model to use only those, you remove its need to guess. Our AI tool database, which tracks 360 AI tools with a pricing and capability snapshot recorded at verification time, exists because tool facts change fast; the same logic applies to any topic.
If a claim is not in your pasted notes, the prompt should tell the model to say so rather than fill the space.
Here is a filled-in example for a post about choosing a first AI writing tool. Role: "You are a technology writer who explains things to people who are not technical." Audience: "Freelancers who write for clients and have never paid for an AI tool."
Topic: "How to pick a first AI writing assistant without wasting money." Source material: three bullet points you wrote yourself, plus the instruction "Do not state any price, plan name, or usage limit that is not in these notes. If you need a number you do not have, write [CHECK PRICE] instead."
Tone: "Direct, no marketing language, short sentences." Length: "800 to 900 words." Format: "Intro, three H2 sections, one worked example, closing checklist."
Notice the [CHECK PRICE] instruction. That is the highest-value line in the whole prompt. It turns the model from a confident guesser into a marker of gaps, and you can search for the bracket afterwards and fill each one from the vendor's own page.
According to our AI tool database, snapshots of tool pricing and capabilities are recorded at a point in time, which is exactly why a prompt should never be trusted to supply a current figure on its own.
Two things to add once the first draft comes back. First, ask for a revision in the same thread rather than starting over: "Keep the structure, cut the intro to two sentences, and replace every adjective that sounds like advertising." Models hold context well within a conversation, so a follow-up is cheaper and more accurate than a fresh prompt.
Second, keep a fact ledger. Before you publish, list every number, date, name, and quotation in the draft and write next to each one where it came from — your notes, a page you opened, or nowhere. Anything marked "nowhere" gets cut or replaced.
This step is where most AI-assisted drafts fail, and it takes about five minutes. If you want a deeper routine for catching invented details, the guide on how to stop ChatGPT from making up fake facts in your content walks through it step by step. One more tip that saves real time: save your best prompt as a reusable template with the role, tone, and format lines already filled in, so future posts only need the topic and source material swapped in.
Writers who do this report the setup cost drops to almost nothing after the first two or three posts.
Where this fails. A well-built prompt fixes structure and tone, not judgement. It cannot tell you whether your angle is interesting, whether your audience already knows the basics, or whether a source you pasted is itself wrong.
Long prompts also have a practical ceiling — paste twenty pages of notes and the model starts ignoring the middle. Keep source material to the few facts the post actually needs. And if you are writing about a tool's pricing, plans, or limits, the prompt can only organise what you give it; the vendor's own page is the only reliable source for current numbers, because those change without notice.
Finally, a prompt that produces a clean draft on a familiar topic will often produce something flat on a topic you know well, because you notice the missing nuance. That is a signal to add more of your own notes, not to write a longer prompt.