Say which facts are locked.

The model invents every blank you leave open. Say which facts are locked in the prompt so it only fills what you meant to leave free.

Prompt Fluency is communicating so the model doesn't have to guess. Locked facts are the ones you already decided. Dates, names, prices, hours, rules. You know them. The model doesn't. Leave them out and it will invent a clean, confident version that sounds right and is wrong.

Try this ask:

"Write a confirmation for a birthday party at the bowling alley."

You get a polished note with "Saturday at 2pm," "lane 4," and "pizza and soda included." None of that was in your head. The model needed a birthday party, so it built one out of averages.

The model didn't ignore you. It wrote a confirmation. It just invented the facts you forgot to lock.

Same break shows up everywhere.

"Draft the morning bake list for the bakery."

"Write the check-in note for the climbing gym front desk."

"Make the arrival text for the campground."

Bake list. Check-in note. Arrival text. Every one of those hangs on fixed facts: which loaves, which waiver rule, which site number. Leave those locked facts out and the model fills them with whatever usually shows up in text like that.

Name what is locked. Plain English. Right in the prompt.

"Write a confirmation for a birthday party at the bowling alley. Locked facts: party name is Maya's 10th, Saturday Oct 18, 3 to 5pm, lanes 7 and 8, $180 already paid, cake from outside is allowed, no outside pizza. You can invent the friendly wording. Do not invent times, lanes, or prices."

"Draft the morning bake list for the bakery. Locked facts: 24 sourdough, 12 focaccia, 40 chocolate chip cookies, oven 1 for bread, oven 2 for cookies, dough already proofed in fridge 2. You can invent the checklist order. Do not invent quantities or which oven."

"Write the check-in note for the climbing gym front desk. Locked facts: first-time climbers need a waiver before they touch a wall, day pass is $22, shoes rent for $5, kids under 14 need an adult on the floor. You can invent how it reads on a sticky card. Do not invent prices or the waiver rule."

"Make the arrival text for the campground. Locked facts: site B12, gate code 4819, quiet hours 10pm to 7am, firewood is $8 a bundle at the office. You can invent a warm welcome line. Do not invent the site, code, or quiet hours."

Look at what changed. Same job. Same need for helpful wording. But now the model knows which numbers and rules are finished decisions. It can still write. It just can't rewrite reality.

No shame here. I do this too. Once a fact is settled in your world, it stops feeling like something you need to type. It feels like background. The model has no background. It only has the words.

A quick check before you hit send:

Facts that need a "locked" line in the prompt

• Numbers: prices, counts, times, dates, site or lane numbers
• Rules: waivers, age limits, quiet hours, what is allowed in
• Names and IDs: party name, oven assignment, gate code, site number
• Money already handled: paid, unpaid, deposit held
• If you would correct a coworker for getting it wrong, lock it for the model

When the answer comes back looking sharp but false, ask yourself: did I say which facts were locked? Odds are the model filled a blank you thought was obvious. Lock the facts. Leave only the wording free. Ask again.

You're not being controlling. You're removing the need to guess. That's Applied Intelligence Mastery in practice: say which facts are locked, so the model invents the wording instead of inventing your world.