How to Write a Clear Request to an AI Assistant
Maker's note: 기준일 2026-10-03. Written by the AZET team (AZET 운영팀) — we build and operate azet.io. Any status statement about AZET below is current as of October 3, 2026.
Most disappointing answers from an AI assistant trace back to a request that left too much unsaid. That isn't a user failing; it's a skill nobody was taught. We spend a lot of time looking at how people phrase requests, because the assistant we're building at azet.io is organized around one line — "Ask anything. Get it done." — and "get it done" depends heavily on the ask. This guide is the short version of what we've learned, written for people who don't want to study prompt engineering but do want better answers.
The four parts of a good request
A useful request usually carries four things. You don't need all four every time, but when an answer misses, one of them is almost always missing.
| Part | What it answers | Example |
|---|---|---|
| Goal | What are you actually trying to do? | "I'm choosing between two phone plans" |
| Context | What does the assistant need to know about you or the situation? | "I mostly use data, rarely call, and travel twice a year" |
| Constraints | What must the answer respect? | "Keep it under 150 words; no plans that require a two-year contract" |
| Format | What shape should the output take? | "A short table, then one sentence of recommendation" |
Goal first, request second
"Summarize this article" is a request. "I need to decide whether this article is worth reading in full before a meeting" is a goal. The second version lets the assistant tell you what matters for your decision instead of compressing everything evenly. When you lead with the goal, the assistant can make sensible judgment calls you didn't have to spell out.
Context is the part people skip
People tend to assume the assistant knows what they know. It doesn't know your budget, your city, your deadline, or that you already tried the obvious fix. A sentence or two of context usually does more than any clever phrasing. If you'd have to explain it to a new colleague, explain it to the assistant.
Constraints prevent the long, generic answer
Without limits, many assistants default to thorough and broad. If you want three options, say three. If something is off the table, name it. Constraints are also where you state your tolerance for uncertainty: "If you're not sure, say so rather than guessing" is one of the most useful sentences you can add.
Format saves you the second request
If you know you'll paste the answer into an email, ask for email-ready text. If you'll compare options, ask for a table. Asking for format up front avoids the round trip of "now make that shorter."
Before and after
A vague request: "Help me plan a dinner."
A clear request: "I'm hosting four people on Saturday; one is vegetarian and one avoids nuts. I have about two hours to cook and a normal home kitchen. Suggest one menu with a main, a side, and a dessert, plus a shopping list grouped by store section."
Nothing in the second version is technical. It's just the four parts written as plain sentences. The second answer will be more useful because the assistant has less to guess.
Iterating without starting over
A first answer is a draft, not a verdict. The most efficient follow-ups are specific:
- "Keep the structure, but make it more casual."
- "The second option doesn't work because of X; replace it."
- "Which part of this are you least confident about?"
That last question is worth building into your habits. A good assistant should be able to point to the weakest part of its own answer, and that tells you where to check.
Checking what comes back
Clear requests reduce errors; they don't eliminate them. For anything that matters — dates, prices, quantities, names, rules — verify against the original source before you act. A practical habit: when an answer contains a specific fact you plan to rely on, ask where it comes from, then confirm it yourself. The assistant does the drafting; you remain the editor.
What we're building, and where it stands
We're building azet, an AI assistant meant to handle the small, practical requests that fill an ordinary day. As of October 2026 it is not generally available: it's in an early-access waitlist, joined by registering an email at azet.io, and pricing has not been published. Nothing in this guide depends on our product — the four-part habit works with any assistant you use today.
FAQ
Do I need special keywords or "magic phrases" to get good answers? No. Plain sentences that state the goal, context, constraints, and format do most of the work. Tricks matter much less than clarity.
How long should a request be? As long as it needs to be for a new colleague to understand the task. For simple questions, one line is fine; for decisions, a short paragraph usually helps.
What if the answer is still wrong after a clear request? Say exactly what's wrong and why, and ask for a revision. If a fact matters, check it against an original source rather than asking the assistant to confirm itself.
Should I tell the assistant when I'm unsure what I want? Yes. "I'm not sure what I need — ask me three questions first" is a perfectly good request.
Is the AZET assistant available to try? Not generally yet. As of October 2026 it's in an early-access waitlist at azet.io, and pricing hasn't been published.
Join the waitlist
If you'd like to follow what we're building, the waitlist at https://azet.io is open. Registering an email is free and obligates you to nothing — and in the meantime, the four-part request will serve you well with whatever assistant you already use.