AI prompt tone practice exercise for product managers

The scenario

Customer service replies generated by AI often read as correct but cold: they answer the question and skip past the person asking it. The fix is not a longer prompt, it is a more deliberate one, and the fastest way to see the difference is to generate the same reply three different ways and put them side by side.

The exercise gives you a choice of real, common support situations: a refund request, a delayed delivery, a billing error, or a product bug. You pick one, describe the customer's problem in a single sentence, and write a baseline prompt asking an AI to respond to it. That first output is your benchmark, not your answer. From there, you rewrite the same prompt three times, once aimed at an empathetic tone, once at a professional tone, and once at a casual, friendly tone, running each version and keeping every output for comparison.

The comparison is where the actual skill lives. Three replies that all technically answer the customer's question can differ enormously in whether they sound like they were written by someone who cares, someone who is covering themselves, or someone who is just going through the motions.

What you are asked to do

The deliverable has three connected parts:

• Write the baseline prompt and its output for your chosen scenario, then produce three tone variants, empathetic, professional, and casual, each with its own prompt and output shown side by side.
• Score every output on a 1 to 5 scale across four dimensions: clarity, warmth, resolution, meaning whether it actually helps the customer, and brand fit, and note which specific words or instructions in each prompt produced the highest scores.
• Write one final, combined prompt that reliably produces a high quality, on-tone reply for this exact scenario, incorporating whatever you learned from comparing all three tones.

What a strong answer covers

The exercise's own objectives set the bar a submission has to clear:

• Does the baseline prompt actually get run and its output kept, rather than skipped in favor of jumping straight to the tone variants?
• Do the three tone variants read as genuinely distinct in voice, not the same sentence with a different adjective swapped in?
• Does the scoring identify which specific prompt element, a word, an instruction, a constraint, produced the highest scores, instead of just assigning numbers without explanation?
• Does the final combined prompt actually incorporate a lesson from the comparison, rather than restating the original baseline prompt unchanged?
• Would a colleague be able to reuse the final prompt on a different customer message from the same category and get a similarly on-tone result?

Skills this exercise practises

Structuring a prompt around role, context, constraints, and a stated success criterion, instead of typing a single instruction and hoping. Comparing AI outputs against a defined rubric rather than a gut reaction. Iterating a prompt based on what specifically worked, not a vague sense that one output "felt better." These map onto the bootcamp's own curriculum on prompt structure, evaluation, and iteration loops. If you want the same evaluation discipline applied to reviewing AI-written code instead of AI-written text, the AI code diff review exercise from Claude Code for Product Managers is a useful next step. For a wider view of the tools this kind of prompting applies to, see ChatGPT for product managers and best AI tools for product managers.

Which bootcamp this comes from

This exercise is the practical challenge from AI Prompting for Product, a one week, two live session bootcamp on Builders Camp covering prompt structure, research and synthesis prompts, strategy and roadmap prompts, and evaluation loops. Completing the practical challenge counts toward the bootcamp's completion requirement and its certificate, alongside the certification quiz.

Builders Camp runs this bootcamp both live and self-paced, included with the Builders Camp Membership alongside every other bootcamp, track, and masterclass.