Most people prompt an AI like they'd ask a stranger for directions: one sentence, vague destination, disappointing results. The fix isn't a magic phrase — it's giving the model a brief, the same way you'd brief a freelance writer. After publishing dozens of AI-assisted posts, we've settled on a three-layer structure we call the prompt stack.
Layer 1: Persona#
The persona layer tells the model who is writing. Not a job title — a voice.
You are a practical blogger who writes for small-site owners. You explain things plainly, avoid hype, and admit trade-offs. Sentences are short. Jargon gets defined the first time it appears.
Voice instructions beat adjective lists ("be engaging, be professional") because they give the model rules it can apply to every sentence. Two details matter most:
- Reading level. Say who the reader is and what they already know. "Beginners who have never touched AdSense" produces very different text than "publishers comparing ad networks".
- Anti-patterns. Name what you don't want. We explicitly ban "in today's fast-paced world", "unlock", "delve", "game-changer", and the habit of ending every section with a tidy three-part summary.
Layer 2: Context#
The context layer is everything the model cannot guess about your situation:
- Topic and angle. Not "write about email newsletters" but "why a 500-subscriber newsletter beats 10k pageviews for most bloggers".
- Audience and goal. Who reads the blog, and what should they do after finishing the post?
- Source material. Notes, links, data points, personal anecdotes. This is the single highest-leverage input you can provide — it's also what makes the draft yours rather than statistically average internet text.
- Structure request. Either paste an outline or ask for one first and approve it before any prose gets written. Outline-first drafting catches wrong assumptions while they're still cheap to fix.
If you only change one habit this year, make it this: collect inputs before asking for output. Ten bullet points of raw material will improve a draft more than any prompt engineering trick.
Layer 3: Constraints#
Constraints are the format contract. Ours usually reads like this:
- 900–1,200 words
- H2 sections only, no H1 (the template supplies the title)
- Short paragraphs: two to four sentences
- At most one bulleted list per section
- No conclusion that repeats the intro; end with one concrete next step
- Include the provided anecdotes verbatim where marked
Constraints do double duty: they shape the draft and make editing mechanical. When a draft violates a constraint, you don't debate taste — you point at the rule.
Keeping the stack reusable#
Save the whole thing as a template with placeholders:
[PERSONA]
You are {voice description}...
[CONTEXT]
Topic & angle: {angle}
Audience: {who}
Goal: {reader action}
Source notes:
{paste your research}
[CONSTRAINTS]
{your format rules}
Fill the placeholders per post and your drafts converge toward a consistent voice. When the voice drifts — and it will — tighten the persona's anti-pattern list rather than rewriting drafts by hand.
