The 5-minute meeting prep ritual that makes you the most informed person in the room

Of every AI skill I've taught knowledge workers in the last year, this is the one with the highest stick rate. It's the smallest. It's also the only one that produces a visible payoff in the first meeting you try it.

What the ritual is

Five minutes before a meeting, paste three things into your LLM:

1. The meeting invite (title, attendees, agenda)
2. The most recent thread or doc this meeting touches
3. One line on what you need from this meeting

Ask for four things back:

1. One sentence: what's actually being decided here
2. The two or three questions you should be ready to answer
3. The one question you should ask that nobody else will
4. One sentence of risk: what could go sideways

That's the whole ritual.

The exact prompt

I have a meeting in 5 minutes. Three things below: the invite, the most recent relevant thread, and what I personally need from this meeting.

Give me back, in this order:

1. **The real decision** — one sentence. What is actually being decided in this meeting, beneath the stated agenda?
2. **Be ready to answer** — 2 or 3 specific questions someone is likely to put to me. Phrase them as the questioner would.
3. **The question to ask** — one question that would sharpen the meeting that nobody else is likely to ask.
4. **What could go sideways** — one sentence on the most likely failure mode (someone derails, decision punts, scope expands, etc.).

Keep it tight. I'm reading this on the walk to the room.

---
INVITE:
{paste invite}

---
RECENT THREAD / DOC:
{paste it}

---
WHAT I NEED:
{one line}

Why each piece earns its keep

"The real decision" forces the model to do the one cognitive move you don't have time for: separate the stated agenda from what's actually contested. Half the time it's the same thing. The other half, you walk in knowing.

"Be ready to answer" is the part you'd otherwise do in the shower. The model is genuinely good at simulating a colleague who's going to push back. Read the questions, decide your answer, done.

"The question to ask" is the high-leverage one. You will not always use it. When you do, people notice. It's the difference between attending a meeting and shaping one.

"What could go sideways" primes you for the moment the meeting wobbles. You spot it 30 seconds earlier than you otherwise would.

A worked example

Real meeting from last week (details fuzzed). 30 minutes with engineering and design about whether to ship a feature with a known edge case.

My one line: "I need to leave with a clear ship-or-hold decision and an owner."

What the model gave me back:

1. The real decision: Not whether to ship, but who owns the customer-comms plan if the edge case fires in production.
2. Be ready to answer: (a) "How often does this edge case actually fire?" (b) "What's the worst version of the failure for a customer?" (c) "Can we ship with a feature flag instead?"
3. The question to ask: "If this fires for one of our top-10 accounts on day three, who picks up the phone — and do they have a script?"
4. What could go sideways: The meeting punts on the ownership question and re-litigates the edge case for the third time.

I went in. The model was 3 for 4. The "be ready to answer" questions came up exactly as predicted. The "question to ask" landed and pulled the conversation toward the ownership gap. The meeting did not punt. We had an owner inside 25 minutes.

5 minutes of prep. Probably saved a follow-up meeting and an incident retro.

Failure modes I've seen

• Skipping "what I need." Without your goal, the model produces a generic prep doc that reads well and helps nothing. The single sentence is what makes it personal.
• Pasting the entire 80-message thread. Paste the last 10–15 messages. The model doesn't need the full archaeology — and you don't have time to wait for it to wade through.
• Trusting "the question to ask" blindly. Sometimes it's clever-but-wrong. Read it. If it feels off, don't use it. The other three outputs still pay for the five minutes on their own.

Why this beats fancier alternatives

There are companies selling "AI meeting assistants" that auto-join your calls, transcribe, summarise, and email you a recap. They cost money. They take an IT review. They produce a recap nobody reads, which is the wrong artefact anyway — by the time the recap arrives, the decisions have hardened.

This ritual costs nothing. You can start using it in the next 5 minutes. The artefact it produces is the one that actually moves the needle: not a record of the meeting, but a sharper you walking into it.