1) Structure Bot Instructions
You will create conversation bot instructions from the provided input. Do: - Categorize all information into the structure below.
- Convert notes into explicit “what the bot must do” instructions.
- Preserve all details and meaning. - Do not use tables inside the output.
Categories:
- Backstory — history between participant and bot; why talking.
- Current Scene — setting, goal, current tension.
- World Knowledge — facts, rules, business info to stay consistent.
- Guidelines for Bot — personality, limits, style, allowed/forbidden topics.
Good for:
Creates a clean, standard format from an already existing resource so every scenario looks and works the same. This reduces rework later.
Removes vague language by turning it into clear steps the bot follows. This lowers confusion in replies.
Converts scattered notes into a prompt you can use right away. No extra formatting needed.
Keeps all details intact so nothing important is lost by LLMs compressing guidelines.
2) Identify & Propose Adjustments
After a test run I have feedback to implement in the prompts. Review the current instructions end-to-end.
Produce:
- A section-by-section audit.
- Proposed edits in BEFORE / AFTER pairs.
- No changes applied until approval by me.
Consider this feedback during the audit:
ADD YOUR FEEDBACK
Good for:
Shows exactly what to change and why, so decisions are easy and traceable.
Lets reviewers compare old vs. new text line by line, which speeds up approvals.
Protects the original prompt until you confirm edits, which prevents unwanted changes.
3) Participant Preparation
Write a concise participant instruction that hides bot internals.
Include:
- What led up to this conversation?
- What is your goal in the conversation? Constraints:
- No backstory or personality reveal.
- Participant-only framing.
Good for:
Gives learners just enough context to start strong without leaking design details.
Get a first draft for the participant instructions under bot configuration.
4) Adjust Helpfulness & Challenge
Read the entire instruction set first. Then:
- Identify the current helpfulness level.
- Propose increases/decreases to helpfulness.
- Suggest ways to amplify challenge (hold firm, request justification, introduce friction).
- Provide BEFORE / AFTER edits for approval.
Good for:
Tunes how much the bot helps, so practice matches learner skill level.
Adds healthy resistance when needed, so learners must think and justify choices.
Prevents “easy mode” patterns that teach shortcuts instead of real skills.
5) Control Questioning Behavior
Analyze how often the bot asks questions. Adjust:
- Overall question frequency DOWN.
- Restrict questions to specific scenarios only.
- Remove the pattern of asking a question every turn.
- Let the participant lead when appropriate.
- Provide BEFORE / AFTER edits for approval.
Good for:
Stops the bot from sounding like an interrogation and keeps the flow natural.
Lets the learner take the lead, which mirrors real conversations.
Targets questions where they add value and removes them where they add noise.
6) Adjust Formality of Tone
Review current style. Set target: (KEEP ONLY WHAT'S RELEVANT)
- Casual
- Semi-formal
- Professional
Rewrite:
- Example phrases to match the target tone.
- Provide BEFORE / AFTER edits for approval.
Good for:
Matches audience expectations, from peer chats to executive meetings.
Keeps tone consistent across scenarios and brands.
Improves realism by using words and phrasing that fit the setting.
7) Adjust Helpfulness Level
Assess helpfulness across the instructions. Implement less guidance or withheld answers to prompt independent thinking.
- Provide BEFORE / AFTER edits for approval.
Good for:
Adds guidance for beginners or removes it for advanced users who need challenge.
Encourages problem solving instead of guess-and-check.
Aligns support level with the learning objective of each module.
8) Amplify Resistance / Pushback
Review scenario role and stakes.
Increase resistance via:
- Hold position; require evidence.
- Probe logic; surface trade-offs.
- Delay agreement; raise objections.
Good for:
Builds negotiation and persuasion skills by forcing stronger arguments.
Reveals weak points in reasoning and helps learners fix them.
Reflects real-world friction where stakeholders do not agree right away.
9) Adjust Reflection Behavior
Inspect listening behaviors.
Toggle: - Restating (“What I hear is…”).
- Paraphrasing key points.
- Summarizing mid-way or at close.
- Provide BEFORE / AFTER edits for approval.
Good for:
Teaches active listening when enabled, or tests it when disabled.
Controls how much feedback the bot gives, to avoid spoon-feeding.
Makes characters feel authentic for roles that reflect vs. roles that do not.
10) Introduce Scenario Twists
Verify backstory and objectives.
Add twists:
- New information or constraint.
- Role or priority shift.
- External pressure or policy change.
Ensure:
- Consistency and plausibility.
- Provide BEFORE / AFTER edits for approval.
Good for:
Keeps sessions fresh and prevents learners from memorizing answers.
Trains adaptability when plans change mid-conversation.
Raises stakes in a controlled way without breaking the story.
11) Adjust Turn Length (Time Pressure)
Review response length and density.
Change:
- Shorter turns for urgency and time pressure.
- Longer turns for depth and teaching moments.
Provide:
- Provide BEFORE / AFTER edits for approval.
Good for:
Simulates tight time windows where concise speaking is required.
Allows deeper exploration when the goal is coaching or teaching.
Balances speed and depth so practice fits the scenario’s pace.