Most (ChatGPT) prompts fail because they’re wishes, not briefs.
Good outputs start with good inputs. Prompting isn’t magic; it’s a good brief.
✦ The 7 deadly sins of prompting (to avoid at all costs).
✦ Iteration beats length: prompt quicker, not longer.
✦ The R-E-X prompt works 99% of the time.
Let’s retrain how you prompt in this 10-minute guide.
The 7 sins and their fixes
1. No context
Why it fails: the model guesses your job, audience, and constraints. Guessing multiplies errors.
Quick fix: add role + task scope + constraints.
Before: “Analyze this.”
After: “You are a product analyst. Analyze the attached transcript for founders from seed to Series A to find the vision outliers. Output a 5-bullet decision memo. Max 180 words.”
2. Vague instructions
Why it fails: You didn’t define success.
Quick fix: Define success and acceptance tests.
Before: “Write about marketing trends.”
After: “Write a 1,000-word brief on the three most important B2B AI marketing trends for Q3 2025. Include one data point per trend with a source and a one-line implication.”
3. Treating it like Google
Why it fails: Asking questions is level 1. Give it directives.
Quick fix: Change questions into jobs with deliverables.
Before: “What are good onboarding ideas?”
After: “Draft a 5-step onboarding flow for a B2B SaaS. Include the email subjects, timing in days, and one KPI per step.”
4. Asking for everything at once
Why it fails: One giant ask hides failures and creates spaghetti outputs.
Quick fix: Split into steps and chain outputs.
Before: “Create our GTM plan, website copy, and investor memo.”
After, step 1: “List the 5 core customer jobs-to-be-done with a one-line pain for each.”
After, step 2: “Using the chosen JTBD 2 and 4, write homepage H1 options (5) within 8 words each.”
After, step 3: “Expand H1 #3 into a 150-word hero section.”
5. Not iterating
Why it fails: It’s a chat. So have a chat with it.
Quick fix: Critique-then-revise loop is the key.
Before: “Write the article.”
After, step 1: “List 5 potential angles for my article about [topic] to get eyeballs from [audience]. Focus on [niche].”
After, step 2: “Using the chosen angle, let’s write 10 SEO-ready titles with their one-liners summaries.”
After, step 3: “Using the chosen titles and one-liner, let’s create the outline of the article to…”
6. No format or tone
Why it fails: models default to generic structure and bland voice.
Quick fix: force the shape and the voice.
Before: “Announce the feature.”
After: “Write a LinkedIn post. 220 words. Hook (2 lines), 3 bullets, one CTA. Tone: direct and practical, no buzzwords, plain English.”
PS: If I were to write a Linkedin post, I’d use easygen instead of chatgpt.
I explained how here.
7. No examples
Why it fails: Examples are how models learn your taste.
Quick fix: Add 1–2 gold standards (and optionally one anti-example).
Before: “Write a landing page.”
After: “Model the tone and density on these two snippets [paste]. Avoid this anti-example [paste]. Keep sentence length under 16 words.”
The R-E-X Prompt
The R-E-X prompt is to define a role, give examples & set expectations in your (ChatGPT, or other AI) prompt.
Role
Define who the model is and the constraints of the job. Domain. Audience. Risk tolerance.
Examples
Paste one or two gold outputs to imitate. Add a short note on why they work. Optional anti-example.
Expectations
State format, length, tone, any banned words, a scoring rubric, and the iteration loop.
R-E-X 3-step checklist
Write the Role line.
Paste the Examples.
Set Expectations: format, word range, tone, rubric, and loop.
Role: Senior marketing analyst for non-technical leadership. Plain English. No hype.
Example to imitate: “CTR 1.8% vs 1.2% (+0.6pp). CPC $2.45 vs $3.10 (−21%). So what: shift budget to exact-match.”
Inputs I’ll paste: timeframe; channels; our metrics; benchmarks (or “none”). If anything’s missing, ask one precise question and stop.
Task: Compare us vs benchmarks; flag any KPI with ≥15% gap (better or worse).
Return in order: Summary (≤5 lines + one “so what”); Comparisons by channel inline; Findings (5–7 lines with likely cause + confidence H/M/L); Actions (top 3 with impact %, test days, difficulty L/M/H); Notes (methods + data risks).
My Prompt Maker
I made a GPT to help you write better prompts.
It’s free, and I don’t get anything from it.
Go to Prompt Maker.
Write your (bad) prompt.
You will receive a (better) prompt.
Self-critic: sometimes, the Prompt Maker makes it too long.
What I love to do is to use it and trim it before copying/pasting the prompt.
If this whole process feels too long, prefer this:
Write a quick, clear prompt.
Ask ChatGPT to ask you clarification questions.
You’re “forced” to make it clearer = writing a better prompt.
Like anything, the more you prompt, the better you are at it.
That’s it for this Sunday’s edition.
✦ For the full archive of my blogs: https://docs.google.com/document/d/1pWuMCBVQo1zKcgKltX_BZxAr31KgxmOlp3Vzvmc5Hxc/edit?usp=sharing
✦ To join my member-only chat:
Fantastically insightful article. Was not expecting the Prompt Maker to be integrated within GPT. Brilliant.
Thank you.
The full archive for the blog is not accessible