AI for meeting prep: walking into a decision meeting already argued with

What are you actually preparing for?

Not the presentation. The presentation is the part you already rehearse. What decides a contested meeting is the eight minutes after you stop talking, when someone raises the thing you had not considered and you answer it badly in real time because you are hearing it for the first time.

Meeting prep with a model is worth doing because that specific failure is now cheap to avoid. You can ask a model to argue against you in the voice of the person most likely to block you, get a ranked list in under a minute, and spend the rest of your prep writing one-sentence answers to the three that would have hurt. The skill is not prompting. It is knowing which of the generated objections belong to someone in the room.

The prep sheet has six fields, and one of them is a table

Write it in whatever tool you already use. The shape matters more than the medium.

1. The decision sentence. A verb and a date. "Approve moving two engineers to the admin console until 14 November." If you cannot write it, you are calling a discussion, not a decision meeting, and the invite should say so.
2. The room, sorted. Who decides, who can block, who is there to be informed. Three names in the first group at most.
3. Three numbers, with definitions. Whatever you will say when someone asks how that is measured.
4. The objection table. Four columns: the objection, who raises it, your one-sentence answer, and the evidence you will point at.
5. The fallback. The smaller version you will accept if the full ask fails, decided now.
6. The first two minutes, written out. Not memorised, written. The opening is the only part of the meeting you fully control.

Field 4 is the artefact. Fields 1 to 3 you could write on a train. Field 4 is what the model is for.

The red team prompt

Give the model the one-pager or memo, the decision sentence, and the roles in the room. Then ask it to stop helping.

You are our VP of Engineering. I am proposing that two engineers move to the admin console until 14 November, which delays the billing revamp. Here is my memo. List the eight objections you would raise, ordered by how likely each is to stop this from being approved. For each, say in one line what evidence would change your mind.

Then a second pass, which is where most of the useful material comes from:

Now re-rank those for a room that also contains the CFO and the head of sales. Which objections would each of them raise instead, and which of the eight would nobody in that room actually bring up?

The second prompt is doing something a single pass cannot: it forces the model to attribute objections to people, and attribution is what makes a list actionable. An objection nobody in the room holds costs you nothing to drop. An objection three people hold is the meeting.

Expect roughly eight objections and three or four hits. That ratio is worth stating plainly rather than overselling: more than half of what the model produces will not be raised. The hit rate is still a good trade, because the hits are disproportionately the ones you had not written down yourself, and each miss costs a minute to discard.

What a filled objection table looks like

Objection	Raised by	Your answer	Evidence
Billing revamp slips past the Q1 renewal cycle	Head of sales	It slips six weeks, and the three accounts renewing in Q1 are on legacy billing anyway	The renewal list with contract types
Two engineers is not enough to finish in six weeks	VP Engineering	Scope is the three unreplaced reports, not the whole console	The scoped ticket list
We said no to this in June	CFO	In June we had four exit interviews naming it. We now have 29	The exit interview tags

Three rows, each with a named holder, a one-sentence answer and one artefact you can put on a screen. A table with nine rows and no names is a list of worries, and reading it before a meeting makes you more anxious rather than better prepared.

The edits a model will not make on its own

• Delete the objections nobody in the room holds, even the clever ones. Preparing for an argument that will not happen is how you run out of time for the one that will.
• Rank by who holds it, not by how strong it sounds. A weak objection from the person who signs off outranks a strong one from an observer.
• Compress each answer to a sentence you can say out loud. If it takes a paragraph, you do not have an answer yet.
• Pick the concession in advance. Choose the cheapest thing you can give that lets someone say yes without losing face, and decide it while you are calm.

What this does not protect you against

Everything political, which is most of what decides contested meetings. A model does not know that your VP already promised those two engineers to another team last week, that the CFO quietly stopped trusting the churn metric after it was redefined in March, or that the head of sales is defending a commitment they made to a customer. None of that is in the memo you handed over, and all of it outranks the quality of your argument.

The honest framing: a red team pass upgrades you from unprepared to prepared on the technical objections, which is the category where being caught out is cheapest to prevent and most embarrassing. The political layer needs a different preparation entirely, which is one conversation with each blocker before the meeting, and a model cannot have those for you.

Where prep ends and the rest of the work starts

Prep is the front half of a pair. The back half, turning what was said into a record with owners and reversal conditions, is Claude Code meeting notes to decisions, and the two artefacts should match: the decision sentence you wrote in field 1 is the thing that record should confirm or contradict. If the meeting needs slides rather than a pre-read, AI for product presentations covers the nine slide structure the objection table feeds directly into, and how to write a one-pager with AI covers the pre-read itself.

On the Builders Camp side, the workflow discipline sits in 10x Productivity with AI, a 1 week bootcamp with 2 live sessions and 8 microlessons covering meeting and notes automation, research synthesis and building a system you run daily rather than rebuild each time. The room itself is covered elsewhere: Negotiation for Product Managers, 1 week with 2 live sessions and 9 microlessons, opens on interests, constraints and fallbacks and covers handling scope pressure and difficult conversations, while Product Storytelling covers the framing that makes an answer land when you only get one sentence.

See the 10x Productivity with AI bootcamp

After the meeting, go back to your objection table and mark which rows were actually raised. Do that four or five times and you will have a calibrated sense of which of your colleagues the model models well and which it does not, which is more useful than any prompt improvement you could make.