An AI teammate needs a brief, not a prompt. The brief covers four parts: goal, context, constraints, output. Skip one and the work drifts. Below are five templates we use every day at relly: research brief, meeting prep, doc draft, decision check, and follow-up. Copy them, fill the blanks, ship.
Why a brief beats a prompt
A prompt is a single instruction. A brief is a handoff. The difference shows up the second your task has any real shape: a deliverable, a stakeholder, a deadline, anything more than a quick lookup.
If you've ever opened a chat window, typed a request, gotten back a confident-sounding paragraph that's almost but not quite useful, and then spent ten minutes re-prompting, you've felt the gap. The fix isn't a longer prompt. It's a brief.
Treat the AI like a smart contractor on day one. They're capable, but they don't know what your team already decided, who's reading the output, what tone to use, or what "done" looks like. A brief tells them.
The four parts of any good brief
Every template below is a variation on the same shape. Memorize the shape and you can write a brief on the back of a napkin.
- Goal. One sentence on what the output is for. Not what to do, but why it matters. "We need to decide if X is worth building this quarter."
- Context. The 3 to 5 facts the AI would otherwise have to guess. Audience, prior decisions, related links, who's reading this.
- Constraints. Hard rules: length, tone, format, things to avoid, source requirements.
- Output. What you want back. A doc with H2s? A 5-bullet list? A Notion page with three sections? Be specific.
That's it. Four parts. Below is what they look like in practice.
Template 1: the research brief
Use this when you want a real synthesis, not a list of links. Best for competitive analysis, market sizing, or any "what's actually going on with X" question.
Goal: Help me decide whether to ship a free tier for our voice AI product.
Context: We're a 7-person SaaS startup, $30/seat plan, 200 paying users, mostly product teams. Competitors with free tiers: Granola, Otter. Competitors without: Fathom, Circleback.
Constraints: Use only sources from the last 12 months. Cite every claim. Don't reference our internal pricing doc; treat it as confidential. Max 600 words.
Output: A 4-section memo: 1) what comparable startups did, 2) outcomes (paid conversion, churn, support load), 3) the strongest argument for, 4) the strongest argument against. End with a one-line recommendation.
Why it works: the goal is a decision, not "research." The context tells the AI which comparisons are real. The constraints stop it from quoting a 5-year-old blog post. The output makes the work usable in a 1:1 without re-formatting.
Template 2: the meeting prep brief
Use this 30 minutes before any meeting that matters. Sales calls, board prep, design reviews. The output goes in your notes app, not your inbox.
Goal: Get me ready to lead a 30-minute design review for the new onboarding flow.
Context: Attendees: 2 designers, 1 PM, 1 eng. We're choosing between two layouts. Last week's review surfaced concerns about mobile density and accessibility contrast. PRD is in Linear ticket OPS-411.
Constraints: Don't include best practices I already know. Focus on the open questions. Skip anything not mentioned in OPS-411 or the most recent design file.
Output: 1) the 3 decisions we have to leave with, 2) the 5 hardest questions someone could ask, 3) a one-line position I should take into the room.
The constraint "don't include best practices I already know" is the one most people miss. Without it, the AI writes a competent intro to design review that wastes your prep time.
Template 3: the doc draft brief
Use this when you need a real first draft, not a paragraph generator. Internal memos, customer emails, launch announcements.
Goal: Draft an internal memo announcing we're sunsetting the legacy export feature in Q3.
Context: Audience: 40-person company, mostly non-engineers. Decision was made by the leadership team last week. The new export is shipping June 15. Three customers actively use the legacy version.
Constraints: Tone: warm but direct. Don't apologize. Don't oversell the new version. Acknowledge the three customers without naming them. Max 250 words.
Output: A Slack-ready memo with a one-line subject, a short opener, the decision, the timeline, and what to tell customers if asked.
The "don't apologize" line saves the most time. Most AI drafts of internal news read like a customer service script. A brief that names the tone you want produces a draft you can ship after one pass.
Template 4: the decision check brief
Use this when you've already made a call and want a fast sanity check before you commit. Faster than a 1:1, more useful than rubber-ducking.
Goal: Pressure-test my decision to move our blog from a custom static site to a hosted CMS.
Context: We publish 5 posts a week. Current site is well-tuned for SEO but takes 20 minutes per post to deploy. Team is 3 people, 1 engineer. Budget is $200/month.
Constraints: Don't list pros and cons; I've done that. Don't suggest alternatives I haven't named. Be willing to disagree.
Output: 1) the strongest objection to my plan, 2) the cheapest way to test the objection in 2 weeks, 3) a one-line gut check (ship, wait, kill).
The line "be willing to disagree" matters. Default AI behavior is to agree pleasantly. Naming the disagreement lane is the only way to get a real second opinion.
Template 5: the follow-up brief
Use this right after any meeting where work was assigned but no one wrote it down cleanly. The brief turns a transcript or rough notes into shippable artifacts.
Goal: Turn the attached meeting notes into the follow-up artifacts we promised.
Context: 45-minute weekly sync, 6 people, 4 decisions, 7 action items. Owners are named in the notes. Two threads went unresolved.
Constraints: Don't invent owners. Don't merge separate decisions. If something is ambiguous, flag it instead of guessing.
Output: 1) a Linear-ready list of action items with owner and due date, 2) a Slack recap (under 80 words), 3) a list of the unresolved threads with one suggested next step each.
This is the brief that benefits most from voice AI. relly writes it for you because it was already in the room. With a chatbot, you have to paste the transcript and explain the meeting twice.
Where briefs go wrong
Three patterns to watch for.
Stuffing context into goal. "Help me figure out the best way to think about whether or not we should consider building a free tier given that we have competitors who have done it..." That's not a goal, it's an avoidance. Cut to the decision: "Help me decide if we should ship a free tier."
Skipping constraints because they feel obvious. Length, tone, and "what to skip" are never obvious. They are the first things to write down.
Vague output. "Give me some thoughts" is not an output. "5 bullets, one sentence each, no preamble" is. Specificity at the end of the brief saves cleanup at the end of the work.
Briefs and voice AI
The whole structure above assumes you're typing. Voice AI changes one thing: you can speak the brief in 20 seconds instead of typing it in 90. The four parts are still the four parts.
"relly, research brief: I'm deciding whether to ship a free tier. Compare Granola and Otter, last 12 months only, cite sources, give me a memo with the strongest arguments either way and a one-line rec." Done. The AI does the work while the meeting keeps moving.
That's the unlock. The skill of briefing transfers; the medium gets faster. (For a deeper take on why typing slows teams down, see Typing is the new bottleneck.)
How to start using these tomorrow
Pick one template. Pin it to your notes app. Use it for every task in that category for one week. By Friday you'll know which lines you keep, which you cut, and which you wish you had. Then move to the next template.
Good briefs aren't a writing exercise. They're a thinking habit. The AI is just the first reader who'll never let you skip a step.
Common questions
How do you brief an AI teammate?
Brief an AI teammate the same way you'd brief a new hire: state the goal, share the context it would otherwise miss, name the constraints, and describe the output you want. Skip any of those four and the result drifts. Templates work because they make the four parts impossible to forget.
What is the difference between a prompt and an AI brief?
A prompt is a single request typed into a chat box. A brief is a structured handoff that includes goal, context, constraints, and output format, which is what an AI teammate needs to act independently. Prompts work for one-off questions. Briefs work for tasks you want done well without follow-up.
Should every AI task use a template?
No. Quick lookups, light edits, and exploratory questions don't need a template. Anything with a real output, a deadline, or a stakeholder waiting on the other side benefits from a brief. Use a template when the cost of getting it wrong is higher than the cost of writing the brief.
How long should an AI brief be?
Aim for 4 to 10 lines. A good brief is short enough to read in 20 seconds and specific enough that two people would produce the same output. If your brief is longer than half a page, you're explaining instead of directing, and the AI will spend its budget on the wrong thing.
Want an AI teammate that already knows the brief?
relly listens in your meetings, reads context, and acts before you ask. Early access is open with 50% off for your first year.
Claim early access →