Use your own facts first

The one idea

Remember the golden rule from lesson 1: give it the facts, let it do the language. This lesson turns that into a habit.

When the facts come from you — your notes, your numbers, the pasted document — the model is on solid ground. It is doing what it is best at: shaping language. When you ask it to recall facts on its own, it is guessing from patterns, and that is where it invents things.

The move

Before you ask for the output, paste the raw material and add a boundary:

"Using only the notes below, write the update. Do not add any facts, names, or numbers that are not in my notes. If something important is missing, ask me."

That one constraint — do not add facts — is doing real work. It tells the model to stay inside your material.

Why it is safer

• Lower hallucination risk: the model has nothing to invent because you supplied the substance.
• Easier to check: every claim should trace back to something you wrote.
• Language beats research: "rewrite these notes" is far safer than "tell me the facts about X", because the first is transformation and the second is recall.

Try it now

Take some rough notes or a messy paragraph from your work. Paste it and ask for a polished version with the rule: no new facts. Then scan the result — every fact should be one of yours.

Your win today

You can produce polished, trustworthy work fast by feeding the facts yourself and letting the model do only the part it is reliable at: the words.