Why this lesson exists. Lesson 1 showed that the answer depends on what the model can see. The fastest way to control that is the message you type. Most disappointing output comes from a message that would confuse a human too, so this lesson gives you a structure that briefs the model the way you would brief a capable new colleague.
By the end of this lesson you will be able to write a brief that gets usable output the first time, because you will have seen a weak brief and a strong one side by side and completed a strong one yourself.
Here is what most people type:
"Write a follow-up email to a client."
And here is the same request, briefed properly:
"You are writing as me, Sam, founder of a small training-platform company. Draft a follow-up email to Priya, Head of L&D at a 300-person housing association. We had a demo call on Tuesday (notes below). She liked the branded learner space but was worried about migrating existing courses from their current platform. Goal: get a second call booked with her IT lead. Under 150 words, warm but not salesy, no bullet points, end with one specific ask. Notes:...."
The first produces an email you would never send. The second produces one you could send after a thirty-second read. It took about a minute longer to write, and saved ten minutes of editing. Notice what the second version contains that the first does not.
Look back at the strong brief. It has five ingredients. You will not always need all five, but when output disappoints, one of them is usually missing.
Role and audience. "Writing as me, Sam… to Priya, Head of L&D." Who the model is speaking as, and who will read the result. Without this it defaults to a neutral voice for a generic reader, which is why so much AI writing sounds the same.
Task. "Draft a follow-up email." The specific verb. "Summarise", "analyse", "rewrite" and "draft" produce different things, and "help me with" produces a question back.
Context. "Demo call Tuesday… liked the branded space… worried about migration… notes below." Everything the model needs and does not have. This is Lesson 1 applied to a single message, and it is the part people skip most and the part that matters most.
Format. "Under 150 words, no bullet points, end with one specific ask." Length, structure, tone. Without it the model picks, and it tends to pick long.
Constraints and examples. "Warm but not salesy." What to avoid, and what good looks like. Pasting one email that worked and saying "match this tone" does more than three adjectives.
The reason this structure works is not magic. It is the information a good human colleague would need before doing the job well, written down instead of assumed. The model cannot ask you for it unless you invite it to, so it guesses at whatever you leave out.
Add this line to any brief that matters: "If anything is unclear or you need more information, ask me before answering." It turns the model from a guesser into a colleague who checks. You will be surprised how often the question it asks is one you should have answered.
