- Why 90% of bad AI results are prompt problems, not model problems
- The 5 building blocks: Role · Context · Task · Format · Constraints
- 3 ready-made prompt templates to copy (Content, Analysis, Repurposing)
Most "the AI can't do that" moments are actually prompt problems. The model is rarely the bottleneck — the prompt gives it too little to answer well consistently. This tutorial gives you a repeatable scaffold that turns lucky hits into reliable results.
Building Block 1 — Role
Give the model a clear role. "You are an experienced B2B performance marketer" instantly narrows the response space and lifts tone, vocabulary and assumptions to the right level. Without a role, the AI answers generically — the average of everything.
That's the first and cheapest lever. The other four building blocks — and the three ready-made templates — are in the full version:
Full version free with email
Building blocks 2–5 plus 3 copy-paste prompt templates. Sign up once, unlocked immediately — and you get new tutorials first. No spam.
Building Block 2 — Context
Everything the AI shouldn't have to guess: audience, product, tone, previous attempts. Context is the difference between "write a LinkedIn post" and a post that sounds like you.
Building Block 3 — Task
One clear, single assignment with a verb: "Write," "Analyse," "Compare." Multiple tasks in one prompt water down the result — split them.
Building Block 4 — Format
Say exactly what the output should look like: length, structure, table, bullet list, tone. "Max. 120 words, 3 bullets, no buzzwords" beats any vague ask.
Building Block 5 — Constraints
The "don'ts". This is where you get brand consistency: "No filler like 'in today's fast-paced world', no exclamation marks, British English."
Template 1 — Content
Role: You are a B2B content strategist for fintech. Context: Audience = marketing directors, tone = operator, no-nonsense. Task: Write a LinkedIn post about [TOPIC]. Format: max. 140 words, hook in line 1, 1 concrete example. Constraints: no buzzwords, no hashtag spam, no exclamation marks.
Template 2 — Analysis
Role: You are a sceptical performance analyst. Context: Here is the campaign data: [DATA]. Task: Find the 3 most important insights + 1 risk. Format: Table (Insight | Evidence | Action). Constraints: only what the data supports, no speculation.
Template 3 — Repurposing
Role: You are a multi-channel editor. Context: Source text = [ARTICLE]. Task: Turn it into 1 LinkedIn post, 3 X posts, 1 newsletter teaser. Format: separate per channel, each with its own hook. Constraints: no 1:1 repetition, keep the core message.
The Quality Check
Before you use a result: would a good freelancer have delivered the same thing with this brief? If not, what's usually missing is Context or Constraints — the two building blocks people forget most often.
More on orchestrating many prompts/agents: How I installed 184 AI agents with Claude.