When AI Asks What You’ll Do, Keep the Answer Provisional

Imagine you ask an AI assistant to help plan a small online workshop. It asks, ‘How likely are you to run it next month?’ You say, ‘Probably.’ The assistant produces a launch calendar. By the end of the conversation you are choosing a registration deadline, although you still do not know whether your co-host is free. The assistant has done useful work, but the conversation moved from forecasting to planning without anyone making a decision.

A question about what you might do can be useful precisely because it makes you consider the action. That also makes it a poor substitute for a decision. When you use AI to think aloud, it helps to distinguish three things: how likely an action seems, whether you want it, and whether you are ready to commit. You can answer all three differently without contradicting yourself.

Asking about an intention can change the situation

In a 1993 paper in the Journal of Consumer Research, Vicki Morwitz, Eric Johnson and David Schmittlein studied whether asking about purchase intentions affected later purchases. They reported higher subsequent purchase rates after a single question in the settings they studied. Repeated questions did not simply amplify the effect: among people whose initial intentions were low, repeated measurement was associated with a lower tendency to buy. The researchers also reported weaker effects with prior experience of the product.

A later set of experiments by Morwitz and Gavan Fitzsimons examined how questions about general purchase intentions could affect choices among specific products. In one experiment, university students chose among Canadian chocolate bars that were then unavailable in the US market. The researchers explored whether the question made existing attitudes towards options easier to bring to mind. That is one account examined in a consumer setting; it does not establish what an AI coaching question will do.

Neither paper studied people talking to a chatbot. The practical inference is narrower: when a conversation asks you to imagine taking an action and immediately starts building around it, do not assume the answer was a neutral reading of a settled intention. It might be the first time you considered the action in detail. That can be helpful. It is also a reason to leave room to reconsider.

Give different answers different jobs

Suppose the workshop looks likely because your audience has asked for one. You may still prefer to run a shorter trial, and you may be unwilling to announce any date until the venue and co-host are confirmed. A forecast belongs in your planning notes. A preference can guide the options you compare. A commitment belongs in the calendar or in a message that other people can act on.

This distinction matters most when an AI tool produces several outputs from the same conversation. The draft calendar might be useful for checking lead times; a draft invitation may help you see whether you can explain the workshop in two sentences. Keep both labelled as provisional. A document can be ready for review even when its underlying proposal is not ready to share.

You can ask for that separation directly: ‘I’m exploring whether to run a workshop next month. Treat my estimates as provisional. Give me one plan for running it, one for delaying it and the information that would change my decision.’ The point is to keep the comparison alive. If the tool writes only the launch plan, ask what would make the smaller trial or a later date preferable.

Use the question on purpose

Sometimes you want a prompt to move you from vague interest to a specific action. If you have already decided to invite a collaborator, asking when you will send the invitation can help you choose a time. Say explicitly that you have made the decision, and specify which part is still open. The assistant can help with a draft or a reminder while leaving the relationship and timing decisions with you.

If you are still deciding, ask what has to be true for the next step to make sense. In the workshop example, you might need confirmation from your co-host and an estimate of preparation time. Let AI turn those unknowns into questions you can answer. The MaryChuks guide to asking collaborators what they actually prefer offers a useful companion here: a planning conversation cannot settle somebody else’s availability or priorities.

The same distinction applies when you ask your audience a question. ‘Would you come to a workshop?’ can start a conversation and help you learn what people want. It is not the same evidence as registrations after you share a date and a clear description. Keep those numbers separate in your notes, and avoid describing expressions of interest as confirmed attendance.

Before you act on the conversation

• Name what you gave the assistant. Was it a rough prediction, a preference, a condition or an explicit decision? If the summary upgrades it, correct the wording.
• Keep the missing information visible. Write down which fact, conversation or cost would change the plan. Give the uncertainty an owner and a sensible time to revisit it.
• Choose what may leave the conversation. A private outline can help you think. A public date, booking or invitation changes what other people can reasonably expect. Approve that step separately.

At the end of a useful planning chat, you may have a clearer picture of what you want and still decide to wait. That is progress. The test is whether you can say what you know now, what you still need to learn, and which action you have actually chosen.

Sources

• Morwitz, Johnson and Schmittlein, ‘Does Measuring Intent Change Behavior?’ Journal of Consumer Research, June 1993.
• Morwitz and Fitzsimons, ‘The Mere-Measurement Effect: Why Does Measuring Intentions Change Actual Behavior?’ Journal of Consumer Psychology, 2004.