Using AI to Prepare for a Hard Meeting in Ten Minutes

By Trevor L Taylor

The best ten minutes before a hard meeting go to the other side of the table. Write down what the other person is worried about, in their own words, and the conversation starts from their concerns rather than from your opening. A model can draft that list in about a minute, and that is the one part of the preparation worth handing to it.

Picture a founder with ten minutes between calls and the difficult conversation next. A price increase, a missed deadline, a partner who wants to change the terms. The instinct is to rehearse an opening and hope the rest follows.

Rehearsal feels productive because it is the part you control. It also prepares you for the version of the meeting where nobody interrupts, which is never the meeting you are about to have.

How do you use AI to prepare for a difficult meeting?

Describe the situation in a few plain lines, then ask the tool to list the concerns and questions the other person is most likely to raise. Prepare for that list instead of a speech.

The list is the asset. Three or four likely concerns, ranked by how badly each one would derail the conversation, give you something a rehearsed opening cannot.

It also changes what you listen for. Someone who has already written down the objection about pricing hears it the moment it arrives, instead of recognizing it two days later.

The list changes your first move as well. Naming the other person's main concern in your opening, before they have to raise it, turns the first two minutes from a pitch into a conversation about the thing they came to discuss.

The ten minute sequence in order

The sequence matters more than the tool. Two minutes go to the outcome you want and the two facts you are certain of, written plainly enough that a stranger could repeat them back.

Three minutes go to your own questions, before the model writes any. A question tool built at MIT Sloan asks people to write their own questions first, before the model contributes any, and it is used by individuals preparing for a meeting. The order protects the part of the work that is yours.

Then three minutes for the model. Ask for the concerns the other person is most likely to raise, ranked, with the reason each one would matter to them rather than to you.

Ask it for one more thing: the question you would least like to be asked. A model with no stake in the outcome produces that question faster than a colleague will, because a colleague is being polite and it is not.

The last two minutes are yours again. Choose the first thing you will say, choose the first question you will ask, and decide in advance what you are willing to give up.

What goes into the description shapes everything after it. Two lines on what you want, two on what you believe they want, one on the history, one on the constraint you cannot move. Leave names and private details out of it.

Ranking is the step people skip, and it is the one that makes the list usable. A conversation has room for three concerns answered well and no room for nine answered briefly, so the order decides what you actually prepare.

Rank by consequence rather than by likelihood. One concern ends the meeting, one slows it down and one is a detail somebody raises because it is easy to raise, and only the first of those deserves a prepared answer.

Deciding the concession cold is the other quiet advantage. A limit chosen with ten minutes of clear thinking holds better than one invented under pressure, and it keeps you from giving away something you only valued once it was gone.

Preparation that survives contact with another person is a set of questions, never a speech.

There is a prompting habit worth keeping here. Ask the tool to argue the other position as strongly as it can, because a model asked to help you will mostly agree with you, and agreement is the least useful thing it can produce at this stage.

The same boundary is worth setting as a team, which is the subject of an earlier post on keeping a team's judgment while it adopts AI tools. A group that has already decided which calls stay with a person argues less about the tool and more about the decision.

Can AI tell you what the other person will say?

No. It works only from the context you give it, so the list it returns is a set of guesses to test in the room rather than a forecast.

That limit is worth taking seriously, because the same error shows up with no tool involved at all. Research on perspective getting finds that people tend to rely on egocentric projection when predicting what others feel and prefer, which is why every item on the list stays a question until the other person answers it.

So the guesses get asked rather than assumed. Three written concerns become three short questions in the first few minutes, and the answers tell you which one actually matters to the person in front of you.

This is also why the ten minutes are enough. You are not building a model of another person, you are building a short list of things to check, and checking is fast once the list exists.

The checking itself can be one plain question near the start. Something close to what would make this a problem on your side gives the other person room to put their real concern on the table while there is still time to deal with it.

A hard meeting about a decision and a hard meeting about a relationship need the same preparation in different proportions. The decision needs the ranked concerns. The relationship needs the one question you would rather not be asked, and an honest answer ready for it.

Where this preparation stops working

The method fails in two ways. The first is overpreparation, where someone has an answer ready for nine objections and stops listening for the tenth, and the meeting turns into a recital.

The second failure is quieter. One MIT Sloan professor describes AI gravity as the constant pull and push to outsource more of our thinking to AI in order to become more efficient, and preparation is an easy thing to hand over.

A tool that writes the list, ranks it, drafts the answers and suggests the opening has not prepared you for anything. The work that makes the meeting go well is reading the list and disagreeing with parts of it, which takes a few minutes and cannot be delegated.

A model is also confidently wrong at a steady rate. It will invent a concern the other person has never had, and the only correction is the meeting itself, which is a good reason to hold the list loosely once the conversation starts.

And ten minutes is the wrong budget for some conversations. A long history between two people, or a decision that affects somebody's pay or role, deserves a longer preparation and probably a second person to think it through with.

There is a simple test for whether the ten minutes worked. Close the tool, look away from the screen, and say the other person's main concern out loud in one line. If that line comes easily, the preparation is in you rather than in a document.

Hard meetings rarely turn on eloquence. They turn on whether the person across the table believes you understood their problem before you walked in, and ten minutes spent writing down their concerns is the cheapest way to earn that. The first few minutes of a conversation usually decide the rest of it.

Trevor L Taylor is an investor, entrepreneur and lifelong athlete committed to supporting education.

Photos by Ocean Ng and engin akyurt on Unsplash

Sources

This AI tool helps you ask better questions, get more authentic results
The Benefits and Obstacles to Perspective Getting
AI gravity is pulling you toward dependency. Here’s how to push back