Stay, Steer, or Start Over

Subtitle: Decide what the next conversational move should be

What you will make

Create a short decision record showing how you found one mismatch in an imperfect chatbot response, made one focused correction, tested the response with a constructive challenge, and chose the next conversational move. The goal is not a perfect answer. It is to recognize the state of the work and decide deliberately whether to continue, steer, summarize, restart, or stop.

Lab materials

• 📓 Notebook · Lab Notebook — An optional place to record the work and download a local copy.
• 📎 Resource · Learner guide
• 📎 Resource · Stay, Steer, or Start Over?
• 📎 Resource · Gallery
• 📎 Resource · Facilitator guide

Before you start

Level and format: Beginner; individual, pair, observation, or facilitated demonstration

You need the 📎 Resource · scenario cards, the 📎 Resource · worksheet, or the 📓 Notebook · local Lab Notebook. You may use an approved general-purpose chatbot, work from a supplied scenario, or observe a facilitator. A live chatbot account is not required.

Use one modest, low-risk task. Do not enter private, client, medical, legal, financial, employment, account, credential, safety, reputation-sensitive, or another person’s information. If your task becomes consequential or uncomfortable, switch to a supplied scenario card.

The core idea

An imperfect first answer does not automatically mean “start over.” Conversation quality is not measured by length or polish. The useful move is the one that fits the current state of the work while keeping you—not the tool—responsible for the goal and decision.

Choose the next move

Complete the activity here. Open the optional 📓 Notebook · Lab Notebook beside this page when you want a place to record your work.

1. Find one mismatch

   Use a low-risk conversation of your own or choose a 📎 Resource · scenario card. Read the initial answer and identify one assumption, omission, or direction that does not fit the real goal.

   Write down the mismatch before trying to fix it.

2. Clarify the actual goal

   Ask the chatbot what it believes you are trying to accomplish. Compare that with your actual intent.

   Complete this sentence:

   The actual goal is ___, not ___.

3. Steer with one focused correction

   Correct one assumption, narrow the scope, preserve a key idea, or change the format. Do not rewrite the whole task unless the current thread has lost the goal beyond repair.

   Save the correction and note what changed in the response.

4. Add one constructive challenge

   Ask for one counterargument, missing perspective, failure point, or weak assumption.

   Notice whether the challenge improves the answer, reveals a limit, or points to something that must be checked outside the conversation.

5. Decide what happens next

   Choose one move:

   • Continue because the thread is aligned.
   • Steer with a clearer direction.
   • Summarize and preserve the useful parts.
   • Restart with a reviewed summary.
   • Stop because the goal has been met or the conversation should not continue.

   Record why that move fits the current state of the work. If you restart, write the reviewed summary you would carry forward instead of discarding useful context without naming what you kept.

Check your result

• Goal: Can you state the actual goal and the mismatch you found?
• Correction: Did one focused correction move the response closer to the goal?
• Challenge: Did the challenge reveal useful information, a weak point, or a limit?
• Decision: Can you explain why your selected next move fits better than the alternatives?
• Verification: Can you name what must be checked outside the chatbot?
• Privacy: Can you explain what you withheld and why it was unnecessary?

A smoother answer is not proof that it is accurate, complete, safe, or appropriate. Check facts and commitments separately. Do not use this short activity to outsource a consequential decision or justify sharing unnecessary information.

You are done when

You have retained each part of this record:

• The low-risk task or supplied scenario
• The actual goal and the mismatch I found
• One focused steering correction and what changed
• One constructive challenge and what it revealed
• The next move I chose and why it fits
• One item requiring human review or outside verification
• One thing I deliberately did not share

Try another version

These are optional extensions, not additional completion requirements.

• Apply the same five moves to a different supplied scenario
• Compare two focused corrections without changing the underlying goal
• Ask a partner to choose the next move, then compare your reasons
• Repeat the exercise later or with another tool and record variation without declaring one move universally best