Ask for the boring version of a forecast before the confident one

When you ask an AI tool to help write up a forecast, it's naturally going to produce something that reads as coherent and confident, because that's what a well-structured piece of writing sounds like. The problem is that confidence is a property of the prose, not of the forecast underneath it - a set of shaky assumptions can be written up just as fluently as a solid one, and the fluency makes them harder to tell apart.

Ask for the plain, assumption-by-assumption version first, before any narrative polish: every input the forecast depends on, listed separately, with your actual confidence in each one stated next to it (known fact, reasonable estimate, or genuine guess). Only once you've looked at that list and decided it's honest should you ask for it to be turned into the readable version for the pack.

This catches the specific problem where one shaky assumption near the start of a calculation quietly propagates through everything downstream - a growth rate you weren't sure about feeding into three other lines - and would otherwise disappear into a single smooth paragraph that never mentions it's carrying that uncertainty at all.

It also gives you a much better answer the first time someone in the meeting asks "what happens if that number's wrong" - you've already got the assumption isolated, rather than having to unpick it out of prose under pressure.