Can you explain the spreadsheet AI helped you make?

The move Before you present an AI-assisted analysis, close the chat and answer three questions about one output: what it represents, which inputs determine it, and which direction it should move if one input changes. If you get stuck, check the formula or ask someone who knows.

You asked AI to build a revenue-by-region breakdown. The output looks clean. Then your manager asks what's in the Southwest total, and you realize you need to reopen the chat.

OpenAI research on how people prompt AI for cross-occupation tasks found they supply more context and request more verification, but ask for explanations less often. The gap shows up when someone asks you to walk them through the file.

If you can't explain it without reopening the chat, you can't maintain it.

The check

Before presenting an AI-assisted analysis, close the chat and answer three questions about one output you expect to discuss:

1. What does this number represent? Write it in plain language.
2. Which inputs determine it? Name the actual cells or data.
3. If [one specific input] changed, which direction should this number move?

If you get stuck, check the actual formula in the current file or ask a colleague who knows the domain.

Filled in, it looks like this

Output: Average order value, Southwest = $312

1. Total revenue from Southwest accounts divided by the number of orders in that region.
2. Cells B14 (revenue) and C14 (order count), Southwest row.
3. If we include three new Q3 accounts, the order count goes up. The average should drop unless those accounts ordered above $312.

Why this helps: You walk in knowing where the number comes from and what moves it. A follow-up question becomes a formula check, not a guess.

One catch

This check tells you whether you can explain the file. It doesn't tell you whether the analysis is right. A correct formula built on the wrong assumption still gives a wrong answer. That question belongs to someone who knows the subject.