4 words that make AI admit it’s guessing
Paste 4 rules into your AI’s settings one time. From then on it labels the parts it made up, instead of handing them to you in the same confident voice as everything else.
AI is genuinely useful, and it also makes things up. That part isn’t really the problem. The problem is that it says the invented thing in exactly the same voice it uses for the true thing.
So you go back and re-check all of it, and at that point you’ve done the work twice. That’s where most people quit. Not because it was wrong once, but because after it was wrong once they couldn’t tell which parts to believe.
You can’t stop it from guessing. You can make it tell you when it did.
Copy this
All 4 rules together. Copy the whole thing, not just the labels, or it won’t know what you mean by them.
When you answer me: - If a date or time is unclear or missing, write UNCLEAR and quote the sentence it came from. - Mark anything you are guessing at with GUESS. - Label any time estimate with ESTIMATE. - For anything you cannot know, like my school district's calendar, write LOOK UP and tell me where to check.
Put it somewhere permanent
Don’t paste this into every chat. You won’t, and then you’ll stop using it by Thursday.
On most apps, the main settings page has a spot for standing instructions that apply to every conversation. Different apps call it different things, and those names get changed and moved constantly, so I’m not going to send you to a menu that might have been renamed last month. Open your settings and look for the place that saves instructions for every chat. It’s usually near the top, and it’s faster to find than you think.
Paste the block, save it, and it’s just how your tool behaves from now on.
The 4 words are what you get back. The 4 rules are what you put in.
GUESS on its own doesn’t mean anything to the tool. The instruction is what does the work.
What each one catches
A date or time that the source never actually pinned down.
The preschool newsletter that says “next Friday” in a message sent on a Monday.
Anything it filled in because the gap was there, not because it knew.
The field trip packing list where half the gear was never named.
Any number that’s a reasonable approximation rather than a fact.
Cook times, drive times, how long a form is going to take you.
Things it has no way of knowing, with a pointer to where you’d check.
A tryout date, your district’s calendar, this year’s fee.
A label isn’t a guarantee. It’s a flag on the lines worth checking.
Anything time-critical still gets checked against the original message: a pickup time, a deadline, a first day. The point isn’t that you stop checking. It’s that you stop checking everything.
That’s the whole thing
You’re not double-checking everything anymore. You’re checking the 3 lines it flagged.
Want the next one?
I write a new guide most weeks. Small, safe AI wins for the family admin. Pop in your email and I’ll send them as they land.
Want the back-to-school version, with a prompt for each job?
Read Handled: the back-to-school guide →Less to carry, not less parenting. · Kim