Say which parts you're guessing.

If you want to stop AI treating your guesses as facts, say which parts you're sure of and which parts you're not. When you type a number you half remember, it lands in the prompt looking exactly like a number you checked twice. The model can't hear the "I think" in your head. It just builds on it.

Prompt Fluency is communicating so the model doesn't have to guess. This one is sneaky, because here the model isn't guessing at all. It's trusting you completely. Every word you give it gets the same weight. So a rough estimate becomes a foundation, and everything stacked on top of it looks solid even when the bottom is soft.

Try this ask:

"We're moving a three bedroom house about 200 miles. It's around 9,000 pounds of stuff. Build me a budget for renting a truck and doing it ourselves versus hiring movers."

You get a clean table. Truck size, fuel, mileage fees, mover rates by weight, a total for each side. Looks great. But that 9,000 pounds was a number you saw on a forum once. You have no idea what your house actually weighs. Could be 6,000. Could be 12,000. The whole comparison hangs on it, and the model never questioned it, because you said it like you knew.

The model didn't do anything wrong. It did the math on what you gave it. You just never told it which part was shaky.

Same thing shows up everywhere.

"The community garden has 40 plots and about a quarter of folks never show up after May. Write the waitlist rules."

"Our food truck sells about 150 tacos on a Friday. Help me price a new menu."

"We'll probably get 120 people at the reunion. Plan the food order."

Plots. Tacos. Headcount. Each one has a number that sounds like a fact and might be a hunch. If it's a hunch, the rules, the prices, and the food order all inherit the wobble. And nothing in the answer will warn you.

So mark it. Plain English. Right in the prompt.

"We're moving a three bedroom house about 200 miles. I'm sure about the distance and the date. I'm NOT sure about the weight. I've heard 9,000 pounds, but that's a guess. Build the truck versus movers comparison, show me how it changes if the weight is 6,000 or 12,000, and tell me what I should measure or ask to pin the real number down."

"The community garden has 40 plots. That part's exact. The 'quarter never show up' part is just my impression from walking by. Write waitlist rules that still work if it's more like a tenth or more like a half."

"Our food truck sells about 150 tacos on a Friday. That's from our register, so it's solid. Saturday I honestly don't know, I'd guess 100. Price the menu off Friday and tell me where the Saturday guess would change anything."

"Reunion headcount is a guess, somewhere 90 to 140. Plan the food order so we're not stuck if it's either end, and tell me the latest date I'd need a real count."

Look at what changed. The answers got more honest. The model started showing you ranges where you were unsure, and it stopped dressing a guess up in a nice table. Better yet, it started telling you what to go find out. You didn't make it smarter. You told it where the ground was soft.

That's the part that clicks. Handing the model a guess without saying so is like giving a friend directions you only half remember and not mentioning it. They'll drive with total confidence right into the wrong county. If you'd said "I think it's the second left, but I'm not sure," they'd have slowed down and checked.

No shame here. I do this too. When I'm in a hurry, I round things off and type them like gospel. Then I'm surprised when the answer is confident and wrong. It was confident because I was. The model only has the words. If the doubt isn't in the words, it isn't anywhere.

A quick check before you hit send:

Ways to mark what you're unsure of

• Say it straight: "this part is a guess"
• Give a range instead of one number (90 to 140, not 120)
• Say where the solid numbers came from (the register, the lease, the invoice)
• Ask it to show how the answer changes if your guess is off
• Ask what you'd need to find out to stop guessing

When an answer looks polished but feels off, ask yourself: did I hand it a hunch dressed as a fact? Odds are it built the whole thing on the one number you weren't sure of.

You're not admitting weakness. You're removing the need to guess. That's Applied Intelligence Mastery in practice: say which parts you know, say which parts you're guessing, and let the model treat them differently.