Prompting for Product People: A Practical Daily Workflow

A no-hype guide to using AI prompts across discovery, synthesis, planning, and delivery. Includes practical prompt patterns and guardrails for better product decisions.

I am not a developer. I am not a data scientist. I do not have an opinion about whether GPT-5 is better than Claude 4 in benchmarks. What I do have is a very busy week, six clients in three time zones, two workshops to run, and a backlog of writing, planning, and prototyping that would have buried me eighteen months ago. The reason it doesn't bury me now is that I use AI tools deliberately, daily, and in a few specific patterns that have made an enormous difference to how much I can ship without losing sleep.

What follows is not a list of prompts to copy. It's a description of how a non-engineer product person actually folds AI into a working week. Take what's useful. Discard the rest. The patterns are the point, not the specific tools.

The thinking partner, not the answering machine

The biggest mistake people make with AI tools is treating them like a search engine: type in a question, accept the first answer, move on. The way I use them daily is closer to a thinking partner. I draft a position. I ask the model to argue against it. I ask it to find the weak premise. I ask what a smart skeptic would say. That conversation, run for fifteen minutes, almost always sharpens the original idea. The AI's actual answer is rarely the output. The output is my own thinking, made better by the friction of having to defend it.

When I was preparing the keynote on AI for golf leaders last year, I had a rough thesis I liked. I spent forty-five minutes asking the model to attack it from five different angles: the skeptical CEO, the technical skeptic, the experienced operator, the journalist, the bored audience member. By the end, three of my favourite slides had been rebuilt and one had been thrown out entirely. The keynote was better because the dialogue forced me to find the holes I couldn't find on my own.

Writing first drafts faster, polishing them slower

I write a lot. Articles, proposals, workshop briefs, internal docs. The first draft used to be the slowest part. Now I dictate a rambling, unstructured version of what I want to say into a voice note, transcribe it, and ask the model to give me back the same thinking in a clean structure. It is never publishable. It is always a useful starting skeleton that takes me from blank page to second draft in twenty minutes instead of three hours.

But here's the discipline: I rewrite every sentence in the second draft. AI prose without human editing has a recognisable, slightly soulless cadence. Too smooth, too even, too eager to be helpful. If you publish it untouched, your readers will notice within a paragraph and trust you less. The model is the typewriter. The voice has to be yours.

Prototyping content before prototyping pixels

When I'm shaping a new product surface, I now start in plain English with the AI before I open Figma. I describe the user, the moment, the goal. I ask for ten different headline candidates, ten alternative button labels, ten alternative empty-state messages. Within thirty minutes I have a content prototype that exposes most of the structural decisions before I've drawn a single rectangle. This used to take me two days. It now takes an afternoon, and the resulting designs are better because the words came first.

Aaron Walter calls this "content-first design" and it's existed as an idea for years. AI just made it cheap enough to actually do. If you're a product designer who isn't doing this yet, try it for one project. Words first, then layout. That order almost always produces a sharper result.

Research synthesis at speed

User interview synthesis used to be the most time-consuming part of any research project: listening back, tagging quotes, clustering themes. I now feed transcripts into a model and ask it to surface patterns across them, then verify by listening to the original audio for the moments it flags. The AI gets me to themes in twenty minutes that would have taken six hours of manual work. The human verification step is non-negotiable; otherwise you risk false patterns. But the speed-up is real.

A discipline I've added: I never let the AI make the meaning. I let it find the candidates, and then I do the meaning-making myself. That distinction matters. Pattern recognition is mechanical. Insight is human. The tool is for one and not the other.

Five prompts that earn their place every week

Below are five prompt patterns I run almost every week, in different shapes, across different tools. They are not magic. They are working habits.

1. "Read this and give me three things that are unclear to a stranger." Useful for any draft, any deck, any proposal. Always finds something I'd missed.
2. "Now argue against the position I just took, as charitably as possible." Brutal, useful, and consistently sharpens my thinking.
3. "Rewrite this for someone who doesn't know our jargon." Ruthless plain-English filter for anything I'm about to send to a non-design audience.
4. "Give me ten alternative versions of this headline / button / question, varying tone and length." Best way to break out of the first idea trap.
5. "Pretend you're a busy CFO. What's the one thing you'd want to know after reading this?" Surfaces the value proposition I forgot to put up top.

What I will not use AI for

There are things I deliberately don't delegate to AI, because the cost of getting it wrong is worse than the time saved. I don't use AI for final-draft writing where my voice matters. I don't use it to make hiring or feedback decisions about people. I don't use it to talk to clients in real time. I don't use it as a substitute for sitting with users. The pattern is consistent: anywhere the relationship, judgement, or accountability is the actual product, I do the work myself. Anywhere the work is mechanical, repetitive, or first-draft, I lean on the tools.

If you're a product person who hasn't yet found your AI rhythm, start with one habit, not ten. Pick one task you do every week that drains you. Try the AI assist version of it for a fortnight. If it sticks, add the next habit. If it doesn't, drop it without guilt. The point is not to use AI for everything. The point is to use the time it gives you back to do the part of your job that only you can do.

In short

Practical checklist

• Start each prompt with the decision you need support on.
• Add context sources and mark how each should be used.
• Specify output format, audience, and quality bar.
• Include constraints and non-negotiables explicitly.
• Run a quick self-critique prompt before sharing outputs.

When to use this approach (and when not to)

• Use this approach when decisions carry customer, revenue, or delivery risk.
• Use it when multiple teams need consistent quality standards.
• Use it when you need repeatable outcomes, not one-off output.

• Do not paste generic prompts without your decision, audience, and constraints in the brief.
• Do not ship first-draft model tone where your judgement and reputation are on show.
• Do not skip iteration—keep examples of weak outputs and tighten the brief until they improve.

Frequently asked questions

What is the biggest prompting mistake in product work?

Jumping to generation before defining intent, context, and constraints. Prompts work best when treated as structured briefs.

Do product people need advanced prompt tricks?

Not usually. Clear problem framing, grounded inputs, and explicit output expectations beat clever phrasing in most real workflows.

What is a good prompt quality check?

Can another team member run the prompt and get a useful, decision-ready output using the same context and constraints?

When should product people avoid prompting?

When problem framing is unresolved. Clarify the question first; prompting cannot rescue vague strategy.

How do prompts support collaboration?

Shared prompt patterns create consistent analysis quality and reduce debate over format rather than substance.