The PM Prompt Guide.
Prompts are briefs, not questions.
Most people use AI like a search engine. Type a question, get an answer, either use it or complain it was not good enough. The people getting real leverage treat AI like a collaborator, and prompting like a craft. Here is the craft, compressed.
Ricardo Ramirez
Founder, Sprintt · Product Builder
TL;DR · 4 takeaways
- 01
Shift from questions to briefs. Context, constraints, format.
- 02
Every good prompt has four parts: role, context, task, output format.
- 03
Iterate. AI is a conversation, not a vending machine.
- 04
Constraints are instructions, not restrictions.
From questions to briefs.
A question asks for an answer. A brief provides context, constraints, and a clear definition of what good looks like. The difference in output quality is not incremental. It is the difference between a Google result and a senior analyst's memo.
The people who get great output from AI are not using secret prompts. They are writing better briefs. That is the entire skill.
Prompt quality is output quality. The model is not failing you. The brief is.
Four parts of a good prompt.
Every prompt that produces useful output has the same four elements. When an output misses, one of these is underspecified.
- 01
Role
Who should the model think it is? "You are a senior product strategist reviewing a pricing proposal" produces different output than no role at all.
- 02
Context
What do you know. What constraints exist. What you have already tried and ruled out. The more grounded context, the less the model has to guess.
- 03
Task
Be specific. Not "write something about onboarding" but "write a three-paragraph executive summary of the onboarding problems surfaced in the five interviews below."
- 04
Output format
Tell it how the answer should look. Bullet points. A table. A 150-word memo. If format matters, specify it, or you will get a wall of text.
Templates for PM work.
These are prompts I use weekly. Paste, fill in the brackets, iterate. They are shaped, not magic.
You are a senior UX researcher. I am sharing N customer interview transcripts. Identify the top five friction points mentioned, group them by theme, note how frequently each appears, and flag anything a single customer said that is unusually sharp. Output as a table: Theme, Frequency, Representative Quote, Why it matters.
You are a PM writing a PRD for engineering. Feature: [description]. Problem: [problem]. Constraints: [constraints]. Known unknowns: [open questions]. Write a PRD with sections: Overview, Problem, Goals, Non-Goals, User Stories, Open Questions. Keep it under 600 words. Assume the reader is an engineer who will ship this in two weeks.
I have N roadmap candidates below. For each, I will give you an estimate of Reach, Impact, Confidence, and Effort. Apply RICE scoring, show your reasoning per item in one sentence, and output a ranked list. Flag any item where my confidence looks too high relative to the evidence I provided.
You are a PM summarizing a sprint to a VP of Product. Under 150 words. Cover: what shipped, what did not ship and why, the top risk to address next sprint, one customer signal worth noting. Direct tone. No hedges. No corporate voice.
The four mistakes.
- 01
Being vague
"Make this better" tells the model nothing. Tell it what better means. More concise. Less hedging. Written for a non-technical audience. Be specific about the direction.
- 02
One-shot thinking
Treat AI like a conversation, not a vending machine. If the first output misses, refine. "Close, but too formal. Rewrite paragraph two in a direct voice, cut qualifiers."
- 03
Accepting the first draft
AI first drafts are starting points. The model hedges, over-qualifies, pads. Edit it down ruthlessly. Push it to be sharper. The second pass is where the value is.
- 04
No constraints
Unconstrained prompts produce generic output. Word counts, tone, format, audience. These are instructions, not restrictions. Give the model a shape to aim for.
Advanced moves.
Once the basics are automatic, these techniques unlock output quality most people never see.
- 01
Ask it to think out loud
"Before you respond, list your assumptions." The model often hides reasoning. Make it visible. You will catch the wrong ones before they ship.
- 02
Steelman the opposite
"What is the strongest argument against this approach?" Use AI to pressure-test your thinking before a meeting where you cannot afford to be wrong.
- 03
Persona interviews
"You are a skeptical CFO reviewing this business case. Ask me the three hardest questions." Simulate the reaction before the real room gets in.
- 04
Iterative critique
"Here is the output. Here is what I do not like about it. Revise with these specific changes." The conversation IS the prompt. Each turn is a refinement.
- 05
Constraint stacking
Layer constraints. "In 120 words, in the voice of the CEO, without using the word 'leverage,' rewrite this." Specificity is where clarity lives.
Write a brief. Read the output. Describe what is wrong in one sentence. Rerun. Three turns in, you have a draft worth keeping.
If this was useful, pass it on.
Ricardo Ramirez
Product Builder and Founder of Sprintt. Advising product teams on AI strategy and operating models.