Teacher’s Edition
Module 03
The Conversation as Interface
Multi-turn conversation as a refinement tool. Four types: add data, adjust tone, change format, polish details. Each turn changes one specific thing. Five turns can take generic output to something worth publishing.
Charter Oak Strategic Partners · Claude Mastery Program · Version 1.0 · Confidential · Not for distribution to participants
Modules 01 and 02 taught single-prompt skill. This module teaches the other half: using conversation to refine output across multiple turns. A five-turn conversation that takes five minutes of writing often produces better results than a single prompt that takes twenty minutes to perfect. The key insight: each turn in a conversation should change one specific thing. When people try to change everything at once, the conversation drifts. When they change one lever at a time, the output converges on what they want.
The four refinement types are: adding data (giving Claude more facts), adjusting tone (changing voice, formality, or energy), changing format (restructuring the output shape), and polishing details (final-pass improvements to word choice, accuracy, or completeness). Naming these types gives participants a vocabulary for what they are doing across turns, which makes them deliberate instead of random.
Opening — 3 minutes
“You now know how to write a strong first prompt. Claude’s real power shows up in conversation. Think of each follow-up message as a steering wheel adjustment. You have four ways to steer.”
“One: add data. Give Claude more facts it did not have. Two: adjust tone. Make it sharper, warmer, more formal, more casual. Three: change format. Turn the paragraph into a table. Turn the report into an email. Four: polish details. Fix a specific sentence. Tighten the opening. Add a number you forgot.”
“Each turn should do one of these four things. When you try to do all four at once, Claude does none of them well.”
Live Demo: Five-Turn Refinement — 15 minutes
demo-data/module-03/multi-turn-conversation-demo.md— Full five-turn sequence with refinement type labels.The case study starts with a bare request: write a customer case study for Greenfield Manufacturing. Turn 1 output is generic. Turn 2 adds specificity: the customer is a 200-person manufacturer, the result was an 87% reduction in QA review time, the person quoted is Maria Flores, Director of Operations. Turn 3 adjusts tone: make it more conversational, less corporate. Turn 4 changes format: restructure as a Q&A instead of a narrative. Turn 5 polishes: tighten the opening, make the call-to-action more specific, add a pull quote.
Each turn is labeled by refinement type in the demo file. Narrate the labels as you go so the room connects the theory to the practice.
Open Claude. Run the first prompt. As the output appears:
“This is our starting point. Generic. Correct, but useless. Now watch. Turn 2 is a data turn. I am giving Claude facts it did not have.”
Send Turn 2. Wait for output.
“Better. Specific. But the tone is corporate brochure. Turn 3 is a tone turn.”
Send Turn 3. Wait.
“Now the voice sounds human. But the format is wrong for our use case. I want a Q&A, not a narrative. Turn 4 is a format turn.”
Send Turn 4. Wait.
“Close. The structure works. A few rough edges to sand off. Turn 5 is a polish turn.”
Send Turn 5. Wait.
“Five turns. Five minutes. The output went from generic to publishable.”
“At which turn did the output cross the line from ‘I would not use this’ to ‘I could work with this’?”
Most people say Turn 2 or Turn 3. The shift from unusable to usable happens when you add real data. The shift from usable to good happens when you refine tone and format.
Anti-Pattern Demo: The Conversation Gone Wrong — 7 minutes
demo-data/module-03/conversation-gone-wrong.md— Five-turn failure with analysis and single-prompt fix.The anti-pattern shows five turns where the user says “make it better,” “no, more professional,” “actually, start over,” “can you make it more engaging but also more formal,” and “I don’t know, just make it work.” The conversation produces progressively worse output because each turn gives Claude contradictory or empty instructions. “More professional” and “more engaging” pull in opposite directions without clarity about what either means. “Start over” erases accumulated context. “Make it work” conveys frustration but zero information.
The demo file includes a single structured prompt that would have produced the desired output from the first turn. This is the punchline: sometimes the best multi-turn strategy is a better first turn.
“Now let me show you how most people actually use conversation with Claude.”
Read the five turns aloud from the demo file. Do not paste them into Claude. Reading them aloud is more effective because the room hears the frustration escalate.
“Turn 1: ‘Write me a case study.’ Turn 2: ‘Make it better.’ Turn 3: ‘No, more professional.’ Turn 4: ‘Actually, start over.’ Turn 5: ‘Can you make it more engaging but also more formal? I don’t know, just make it work.’”
Pause. Let the room laugh or cringe. Then:
“What went wrong? Everything. ‘Make it better’ says nothing. ‘More professional’ is undefined. ‘Start over’ throws away whatever Claude learned about your preferences. And ‘more engaging but also more formal’ is a contradiction unless you specify what each of those words means to you.”
“Here is the fix. One prompt. One turn.”
Show the single structured prompt from the demo file. Let the room see that all the refinement could have been avoided with a better starting point.
Paired Exercise: Alternating Refinement Turns — 15 minutes
“Pairs. Pick a topic. One of you writes the initial prompt. The other person writes each refinement turn. Minimum four turns. Each turn must use a different refinement type. You cannot use data twice, tone twice, format twice. You must cover all four types.”
“The constraint forces deliberate refinement. No more ‘make it better.’ Every turn has a purpose.”
Some pairs will write turns that barely change anything. “Add a title” is not a format turn. “Make it a little shorter” is not a constraint turn. Push them: “What specifically should change in this turn? What was wrong with the previous output that this turn fixes?”
If a pair finishes in three minutes, their turns were too shallow. Send them back in.
Debrief — 10 minutes
“Two questions. First: at what point in your conversation did the output become something you would actually use? Second: when is it better to start a new conversation entirely instead of continuing to refine?”
The first question teaches them to notice the inflection point. For most pairs, it happens around Turn 2 or 3.
The second question is harder. The answer: start fresh when your context has become contradictory, when you want Claude to approach the problem from a completely different angle, or when you have accumulated so many instructions that Claude is trying to satisfy all of them simultaneously and satisfying none.
| Segment | Activity | Time |
|---|---|---|
| Opening | Four refinement types | 3 min |
| Demo | Five-turn refinement, live | 15 min |
| Anti-Pattern | Conversation gone wrong | 7 min |
| Paired Exercise | Alternating refinement turns | 15 min |
| Debrief | Discussion | 10 min |