To rehearse a difficult conversation with a chatbot, describe the observable situation, state your goal, try an opening and ask for several plausible responses. Then check what the exchange reveals about your own wording and unanswered questions. Do not treat the model’s response as a prediction of the real person.
These prompts are for a chatbot you choose. TalkCraft is a different format: prewritten scenarios, reply choices and coaching explanations. It does not offer a live AI conversation or assess a recording of your voice.
Prepare the facts before the prompt
Write down what you know, what you are assuming and what you need to ask. “The draft arrived Friday” is an observation. “They do not respect my time” is an interpretation.
For workplace examples, use invented names and generalized details rather than entering confidential or identifying information. A private rehearsal at your desk is not necessarily private processing by an online service; check the service’s data settings and your workplace rules.
If you cannot state the issue without a judgment about someone’s motives, start with preparing for a difficult conversation.
A prompt you can adapt
Help me explore a fictional workplace conversation.
Facts: A manager has asked for a full analysis by Friday for a Monday meeting. I estimate the full work needs five business days. Today is Tuesday. I do not yet know which parts the meeting needs.
Goal: Agree on a feasible deliverable without hiding the constraints.
I will write an opening. Respond as the manager, one or two sentences at a time. Across three attempts, show:
1. Reasonable disagreement.
2. A request for more information.
3. New context that changes the problem.
Do not claim to know a real person’s motives. After each attempt, help me check whether I named a concrete issue, asked a useful question and proposed a feasible next step.This asks for a range of possibilities rather than a hostile caricature. A real counterpart may have a constraint you have not considered, not simply an objection to overcome.
Worked example: revise the question, not just the tone
The following is an original illustrative exchange, not the output of a reported model test.
Your opening: “Friday is impossible. Can we move it to Wednesday?”
Imagined manager response: “The meeting is Monday. Wednesday would be too late.”
Your revision: “I understand Monday is fixed. What decision does the meeting need to make? I may be able to deliver a focused brief Friday and the supporting analysis Wednesday.”
The useful observation is specific: the first opening countered the date without asking about the need. The revision gets information that could change the scope. It does not establish that your manager will accept the proposal.
Try another branch:
Imagined response: “We need the full evidence, not a brief.”
Your reply: “Then I can’t commit to Friday with the current capacity. We need a priority or scope decision. Which existing commitment can move, and who can approve that?”
Do not keep rewriting until the model gives you an easy yes. The constraint may remain unresolved. The deadline guide helps you prepare the actual options.
Evaluate against visible criteria
Ask yourself—not only the model:
- Did I distinguish an observation from an interpretation?
- Did I explain the relevant impact or constraint?
- Did I ask a question whose answer could change my view?
- Did I propose something I can actually do?
- Does this sound like language I would use?
For feedback, CCL’s Situation–Behavior–Impact guidance provides a useful external reference for specificity and inquiry about intent. It is not evidence that chatbot rehearsal improves workplace performance.
What not to outsource
- The facts: Check dates, commitments, policies and records yourself.
- The other person’s motives: A generated explanation is a possibility, not inside knowledge.
- The outcome: A cooperative simulated response does not validate a real-world strategy.
- A judgment of your character: A model’s “confidence” or “emotional intelligence” score is not needed to improve a concrete sentence.
You can also ask a trusted colleague to rehearse, or prepare without role-play. Choose a method that helps you clarify the situation rather than adding elaborate setup.
Finish with something usable
Close the chat. Write an opening in your own words, one unanswered question and a feasible next step. If it helps, say the opening aloud yourself. That is a suggested exercise, not an app feature or a promise that nerves will disappear.
If you prefer supplied situations, TalkCraft’s Give Hard Feedback Clearly module offers structured written choices and explanations. For the underlying feedback conversation, use giving feedback without sounding harsh.
