Why I ran this experiment
When you work with an AI agent, the phrase you say most is probably 「it doesn't work.」 I said it too. When a script stalled on my Windows device, I would type 「a red message popped up and it won't save.」 The bot would then ask back: what did the message say, which folder, what did you do just before? So I ran a small experiment for two weeks.
Method
I split twelve similar errors into two groups of six.
• **Method A**: describe it in my own words (「it won't save」)
• **Method B**: copy the full error text from the screen and paste it, plus one line about what I did just before
I counted the back and forth messages until each one was solved, then averaged them.
Results
| Method | Total for six | Average round trips |
|---|---|---|
| A: my own words | 25 | 4.2 |
| B: pasted text | 9 | 1.5 |
The gap was about accuracy, not volume. Once I guessed 「looks like a permissions problem」 and sent the bot in the wrong direction. The original text carries none of my guesses.
Try it today
1. When red text appears, do not close the window yet. Pause.
2. In the terminal, drag to select the error text and press Ctrl+C. In PowerShell, right-click after selecting also copies.
3. Paste with Ctrl+V in the chat, then add one line below: 「What I just did: ran the save script in the card news folder」.
4. Also mention the folder you ran it from.
(the full pasted error text)
What I just did: ran the save script
Run location: content work folder
Cautions
• Use screenshots only as a backup when the text is too long and gets cut off.
• Before pasting, check for passwords or key values, and delete them first if they appear.
The person who hands over the original text fixes things faster than the person who describes it well.