You will be able to choose which background details go into a request and which stay out.
Giving an assistant the right background works like briefing a new colleague who is taking on the job. Tell it who the reader is, what the writing is for, and the facts nobody outside your workplace knows. Leave out anything the job can be done without, personal details above all.
Take Priya, who works in HR at a logistics company in Tampines, Singapore (an invented example). On Monday her manager asks her to tell all staff that outpatient medical claims are moving from paper forms to the company's HR app next month. Her first attempt is a single line: "write an announcement about the new claims process". The assistant returns three cheerful paragraphs with phrases such as "an exciting new chapter". Nothing in them is specific to her company. A prompt with no background tends to produce this kind of generic output, the sort that would fit any company anywhere.
For her second attempt she adds what she would tell a new colleague handed the same job. The draft that comes back needs very little work before she can send it. Her request was the same both times, and the only difference was the background. Lesson 1.1, "The assistant only knows what you put in front of it", calls that background context.
If you only have room for one piece of context, describe the reader. That means three things: who they are, what they know already, and what they have to decide or do once they have read it. A finance director with five minutes to spare gets a very different draft from a warehouse supervisor checking a phone between shifts. Once the reader is described, the model can choose the vocabulary, the length and the level of explanation for itself, so you don't have to list every rule.
What the reader already knows matters as much as who they are. Mention it if they saw an earlier version, if they have complained before, or if the topic is new to them. If you leave that out, the assistant will explain the parts they already know and skip the parts they don't.
Most of Priya's readers are drivers and warehouse staff. Plenty of them read company messages on a phone, and some have filled in paper claim forms for ten years. They need three things from her message: what is changing, when it starts, and what to do if the app does not work for them.
Tell the assistant what the output is for. The same facts could become an email to a client, a slide for your own team meeting, talking points you will read aloud, or notes only you will see, and each of these reads differently. The topic alone gives the model no way to tell which one you want. A purpose statement can be short: a WhatsApp message for the drivers' group; a one-page note the manager will forward to the regional office; points to speak from at tomorrow's briefing.
The purpose decides the length, the tone, how much gets explained and what goes at the top. A group chat message puts the action first, while a note for a regional office usually opens with the reason.
The model has read widely about subjects such as medical claims. It has not read your company's policy, it does not know what your app is called, and it knows nothing about your history. Those details are what make the draft yours. To find them, picture a new colleague about to start the task and list what they would have to ask you first. The answers are usually your product or service, the deadline, the figures, what you have tried already, and whatever went wrong last time.
When a fact is missing, the model doesn't stop to ask. It carries on and puts something plausible in its place, and plausible guesses are often wrong in the one detail that matters. AI fundamentals lesson 6.1, "Why a fluent answer can still be false", explains why. It is part of the separate AI fundamentals course.
Priya's context runs to five sentences, and each one changes what the assistant writes. Between them, they cover these facts:
She works in HR at a logistics company in Singapore. The readers are drivers and warehouse staff who mostly read on their phones and have used paper forms for years. From 1 July, outpatient claims have to go through the HR app and paper forms will no longer be accepted. The change was announced last year and then postponed, so some staff will doubt it is really happening. Anyone who has trouble with the app can go to the HR counter at the depot on Tuesday or Thursday afternoons.
Her paragraph does not name any employee or clinic.
Good context is chosen with care. The job needs the reader, the purpose and the facts that shape the answer, and very little else. Client names, NRIC numbers, bank account details, salaries and anyone's medical history should stay out. When you type them into a chat, a copy ends up somewhere you cannot control.
You can describe the situation without saying who it is about, then add the names yourself when the message goes out. Suppose Priya later wants help answering one driver whose claim was turned down. She would describe him as "a driver whose claim was rejected because the receipt photo was unclear", then put his name in herself before sending the reply. A placeholder such as [client name] or [amount] works well for this. The model drafts around the brackets and you replace them at the end.
For any other detail, ask whether it would change the draft. If it wouldn't, leave it out of the prompt. Your workplace may also have rules about which information can go into which tool. Lesson 8.2, "Data settings and the rules at your workplace", shows how to find them.
Three to five sentences of context is enough for most everyday tasks. A sensitive message or a long report may need more. If the background is running to two pages, the job is probably too big for a single request, and module 3 covers how to break it up.
Most of the time you save comes from work you repeat. If you write the same kind of thing every week, write its context once, keep it somewhere handy, and change only the details each time.
Take three questions into every prompt: who reads it, where it ends up, and what you know that the model cannot. Pick the task you come back to most often and answer all three for it.
Pick a task you do often and write a context paragraph of three to five sentences about it. Cover the reader, the purpose and the facts only you know.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).