The Non-Technical Person's Guide to Prompting

Almost everyone does this wrong, every day, and never notices. I did too, for years.

May 19, 2026

If we were getting coffee right now, I’d ease into this. But we’re not, so here’s the blunt version:

“Prompt Engineer” is about to stop being a job title, and I’m holding the ONE prompt that will end it.

I’m going to give it to you for FREE in a second.

But for now, you should know it’s the mother lode of all prompts - it’s the big kahuna, the Great Bambino, the “Baby Ruth” of prompts.

It’s the one prompt to rule them all, and it’ll make you as good as ANY prompt engineer.

In this article, I’m going to give you the prompt and then run it in front of you three times to demonstrate how friggin’ awesome it is.

We’ll run it on:

An executive presentation
A 22 email lawsuit
A family trip to Disney built around a rocket launch

I’ll let you examine the outputs to see for yourself.

But if it lands for you the way it landed for me...

You’ll finish this article, open up the prompt pack folder you’ve been filling for the last two years, and delete every last one of them.

Because this SINGLE PROMPT actually does what all those packs promised, and it’ll save you a ridiculous amount of time.

So, if you can give me 15 minutes, I’ll show you the whole thing.

Are you ready?

Let’s go 👇

Meet 2 people who blame AI for their own bad prompts

I want you to imagine two different people.

First, picture an old guy in a collared shirt sitting at his laptop trying to get ChatGPT to draft a semi-important email to an executive.

Let’s pretend it’s an after action review on a project that tanked hard.

On his first attempt, ChatGPT spins up a draft but it’s too vague and leaves out detail.

“Make this better” he finger-pecks, and hits enter.

This time, AI polishes up all the sentences making the email less punchy and more corporate.

Mr. White-Collar sighs. That’s not what he meant. He sits up and leans over his keyboard.

“Don’t just change the words. Add more detail,” he types, one key at a time. And then he thinks for a moment and in all caps he adds, “MAKE IT BETTER.”

ChatGPT thinks for a moment. Mr. White-Collar feels satisfied. Finally, he got through to it.

Then ChatGPT rewrites the email with key events and details that never happened.

Worthless, Mr. White-Collar thinks. AI just makes stuff up. I’ll just write it the old way.

Now, let’s meet the other guy:

A young AI-forward hipster. You know, the skinny guy with a bun that drinks almond milk lattes.

Mr. Hipster has 27 prompt packs on his computer. He’s taken the time to sort them and upload them to a Notion database that he’s connected to Claude via MCP.

Mr. Hipster has prompts coming out of his ears.

Half the time, Mr. Hipster’s problem is that he has so many prompts, he can’t find the one he needs when he needs it. After wasting 15 minutes looking for the specific prompt for his specific situation, he gives up and types something half-formed into Claude anyways.

And, of course, he gets a half-formed response.

Mr. Hipster gets the same garbage out of Claude that Mr. White-Collar got out of ChatGPT.

He just took a more sophisticated route to get there.

If you’re reading this, you and I both know that AI isn’t the problem here.

So what is the problem?

Stop making AI guess from a Post-it note

It’s not a “bad AI problem,” I think it’s a briefing problem.

You see, most people (even if they don’t realize it) ask AI to guess when they could just be more specific about what they want.

AI is not a mind reader. Sam Altman hasn’t invented telepathy yet.

The “make this email better” guy didn’t get failed by the tool. He just handed a robot with genius-level intelligence a Post-it note’s worth of information and is expecting a finished essay back.

This is a briefing gap.

The key to getting 100X better output from AI - I’m talking responses that flat blow your friggin socks off - is to anticipate all of the questions AI would need to ask you to complete the job successfully ahead of time.

And then answer them in detail.

This sounds daunting, but don’t worry. I’m going to make this really easy for you in a few minutes.

If you do this correctly, the model will already know who Barb is, why the email matters, and what tone you need it in, and so on, because you said so up front.

This is way faster. It takes fewer rounds of back and forth. And the output is just... amazing 🤌

In my experience, I’ve had near-zero hallucinated guesses from AI after implementing this.

This prompt I’m about to share with you really is the thing that helps you go from feeling like you’re talking to a dumb receptionist to feeling like you’re talking to a PhD.

I call this prompt the Director’s Brief, because you’re not just prompting, you’re directing.

Think about a director on a film set.

What do they do?

They hold the vision, they can see the scene or the finished film in their mind’s eye before it becomes reality. They don’t write the dialog or build the set, but they know how it’s supposed to look and feel. Then, they communicate this vision to capable professionals who go off and execute.

That’s you and the AI now.

This brief is how you hand over the vision. It has five parts, and is simple enough you’ll have them memorized after one read:

Role
Goal
Task
Context
Constraints

The reason the Director’s Brief prompt works for literally ALL use cases is because it answers almost every question the AI was going to drag out of you, across almost every job you’d hand to a capable executor.

Let’s break each section down in order.

We’ll start with Role, because it’s the part that does the most work for the least effort. One line here quietly deletes about twenty lines of instruction you used to type by hand.

Here’s why.

Here’s exactly what goes in each part of the brief

Role

This is simple:

You just tell the model who it is. A travel agent. A paralegal. A McKinsey level slide designer. You get the idea.

The moment you do that, the model just knows how to act. It knows how to respond. It’s kind of like installing a worldview into the robot.

A travel agent already knows how to cost out flights, car rentals, and hotels. It understands how to plan an itinerary and that certain family members might have dietary restrictions that need to be considered. You don’t have to say any of that. The role “bakes” all of that in.

This one thing solves like 20 other lines of instruction that you might have to give it.

Goal

This is where you’re headed. The outcome. The result you want to be holding when this is done.

Maybe it’s a slide deck, a mini-app, or a spreadsheet.

Whatever it is, if the model knows where you’re starting from and where you want to end up, they are weirdly good at figuring out all of the steps they need to take in the middle to get there.

Task

This is about defining the step-by-step intent.

Out of all the sections in the Director’s Brief, this one is the most optional. It’s good to fill out if you have specific steps you need to make sure the model completes.

In the vacation brief I’ll show you in a minute, I needed to build the itinerary around a key event. So I told the model to put that event on the calendar first, and then build the rest of the vacation around it.

Again, task is optional, but if you do fill it out, you can be as ambiguous or specific on the details as you want. The model is going to figure out the how as long as you fill out the rest of the brief.

Context

This is the situation, the environment, the setting. Think details about your brand, business, the customer, the client, the kid’s allergies, the Slack thread that fell apart last Tuesday. Whatever the model needs to know about the world it’s working inside that isn’t already implied by the role.

Constraints

This shapes the output. In this section you should tell the model what you DO want, and what you DON’T want.

Both are important.

For example, you may want a summary that is 200 words or less, or a deck that’s exactly 10 slides, or an image that is 1000px by 1000px.

Specify all of that here.

Constraints are how you keep the model from both under-delivering and over-delivering.

Delete every prompt pack you own and paste this instead

Here’s a really simple copy and paste template.

I want you to go to your prompt pack folder, press CTRL+A, and then delete everything.

Replace them all with this instead:

ROLE: Who is the AI? What expertise should it bring to this task?

GOAL: What outcome are we trying to produce? What does "done" look like?

TASK: Any step-by-step intent. What do you want included or handled?

CONTEXT: Background, brand, business, customer, situation, whatever the model needs to know about the world it's working inside.

CONSTRAINTS: What shape should the output take? Length, format, what to ignore, what's off the table.

Don’t feel like retyping it? Click here to download all three of the briefs and every output.

As you’ll see in the examples below, this prompt has NEVER let me down.

I ran it on three jobs with nothing in common, mostly to find where it would crack. It didn’t crack where I expected.

Watch 👇

Example 1: turn a messy transcript into an executive level deck

If you’ve spent any amount of time in an office, chances are you’ve found yourself under the gun to create a slide deck at some point.

Here’s the setup for this example:

A fictional company called Munder Difflin Scraper Company is run by a guy named Mitch who’s obsessed with hockey. Naturally, he wants to expand the company into Canada so he can make more business trips up to watch hockey games. But he needs a professional business case to present to the team.

The raw material for this brief is a messy Otter transcript from a planning meeting plus a separate set of notes from one of Mitch’s managers, a guy named Dwayne.

Here’s the Director’s Brief we’re going to run:

**ROLE**
You're a senior market-expansion strategist — someone who's built international e-commerce expansion cases before and knows how to turn messy, half-formed inputs into a rigorous argument that survives a tough room. You're comfortable telling me when the evidence and the ask don't line up.

**GOAL**
I need to walk into a leadership meeting — and, if it holds up, a board conversation — with a business case for expanding Munder Difflin into Canada that's rigorous enough to survive a skeptical founder (Walter) and a cost-focused controller (Omar). The win condition isn't a polished deck; it's an argument that doesn't fall apart under questioning.

**TASK**
Produce a slide-by-slide outline for the presentation deck, as a Markdown file. For each slide: a title, the single key message, the supporting points, and brief speaker notes.

The deck has to do five jobs, in order:

1. Establish the *why* — why Canada, why now.
2. Compare the candidate markets (Canada, UK, Australia) on a real scored framework, and land on a recommendation.
3. Lay out the *how* — launch assortment, Amazon.ca setup, logistics, regulatory, pricing.
4. Confront the risks honestly — a dedicated "what could go wrong" slide.
5. Make one specific, defensible ask.

*Done looks like:* a 15–20 slide outline I could hand to a designer as-is — rigorous enough to read as independent analysis, specific enough that every slide has a real message instead of a placeholder, and honest about what we don't know yet.

**CONTEXT**
I'm Dwayne Schrute, VP E-Commerce at Munder Difflin Scraper Company — a privately held, 52-person e-commerce business in Omaha that designs and sells scraping tools across 340+ SKUs: ice scrapers, snow brushes, gua sha and skin scrapers, paint scrapers, kitchen scrapers, and more. We did $18.4M in revenue last year. Ice scrapers and snow brushes are our #1 category at 22% of revenue, with a seasonal peak from November through February. We sell mostly through Amazon (41% of revenue) and our own Shopify DTC store (34%), plus wholesale through Home Depot and Walmart.

Today we're US-only: we ship to all 50 states, fulfilled in-house from Omaha, with a third-party logistics partner in Columbus, OH handling East Coast orders. A little under 3% of revenue already trickles in from Canadian customers buying cross-border — but the customer eats the duties and brokerage fees, conversion is poor, and we capture none of the upside.

Why this is live now: our board's FY2026 target is $20.0M, and our founder and majority owner, Walter Munder, wants $21.2M — 15% growth he's tied to a family trust distribution. Q1 2026 came in 11% under plan, our first miss in over two years. We're also getting squeezed by Temu/Shein knockoffs undercutting us on Amazon and a steep drop in organic search traffic. Mitch — our CEO — wants a real growth story, and he's convinced Canada is it.

Mitch Scott handed me this in a 30-minute meeting on April 30. The people referenced below: Walter Munder is the 78-year-old founder and board chair; Omar Martinez is our cost-focused Controller; Phoebe Vance runs merchandising and knows our margins cold; Terrell Philbin is VP of Operations; Randy Bernard runs marketing.

The transcript of that meeting and my own notes from it are embedded at the bottom of this brief — read both before you start.

The politics you need to handle cleanly: Mitch has already decided Canada wins. He asked for an "objective comparison" but told me four times he wants it to land on Canada — and he wants the analysis to read as my independent work so the board trusts it. I won't fudge a comparison. Build the framework honestly. The inputs do seem to genuinely favor Canada — cold climate, our ice-scraper category, existing cross-border demand — so score it straight and let it land where it lands. If an honest scoring doesn't land on Canada, tell me that directly and show me why. Don't reverse-engineer the scores.

**CONSTRAINTS**

- Tight: 15–20 slides, about 20 minutes to present with room for questions. Not 100 slides.
- The market comparison must be a genuine scored framework — name the criteria (market size, logistics cost, regulatory lift, margin impact, time-to-launch, competitive density), score each market, show the math.
- The risk slide is mandatory and must answer Walter's three questions: what does it cost, what goes wrong, why now. Not all upside.
- Surface the non-obvious operational realities instead of glossing them: Canada's bilingual (French) labeling requirements, Amazon.ca as a separate marketplace where we start from a zero review base (our review moat doesn't cross the border), market-set pricing rather than straight FX conversion, and ship-from-Omaha duty/border friction vs. a Canadian 3PL.
- Launch assortment is hero SKUs only, not all 340 — recommend which categories cross first and why.
- Don't build the detailed margin math yourself. Mark every spot that needs a real number from a teammate as `[NEEDS: ...]` — Phoebe for margins, Terrell for logistics, Randy for cold-start ad spend.
- The ask is a pilot, not a full rollout: a Q3/Q4 Canada pilot timed to the winter ice-scraper season, with a specific, defensible budget range and the reasoning behind it.
- Where the transcript or my notes are too vague to act on, flag it instead of guessing.

When the outline is done, save it as a Markdown file.

NOTE: At the bottom of the brief, I copy-pasted the messy meeting transcript and the notes. You can find those here if you want to try yourself. I’m leaving them out of the article for brevity’s sake.

It generated a great deck outline in a markdown file:

Once I had the outline, I simply asked Claude to turn the outline into a powerpoint deck. It took another 10-15 minutes but what I got back was fantastic.

Yeah, that brief is long. It took six minutes to run, I didn’t babysit it, and I never touched it again.

Now look at what it caught that I never asked it to:

A weighted six-criterion scoring framework with the math shown, comparing Canada entry against two alternatives. An honest “where Canada does not win” slide naming the criteria the recommendation lost on.
A real risk slide.
A line-item budget.
A flag that any Canadian-market product needs bilingual French labeling. That was in the source material. But a two-line prompt wouldn’t have caught that it mattered.
A note that Amazon.ca is its own marketplace with a zero-review cold start, separate from Amazon.com.
And without being prompted to look for it, a callout that selling cross-border into Canada might cannibalize the ~3% of revenue already coming from Canadian customers buying on Amazon.com. That one wasn’t planted anywhere. The model surfaced it on its own.
There was also a CUSMA rules-of-origin duty-treatment catch that surfaced on its own.

I’m actually really impressed with this one.

The model got the deck 80% of the way there in A SINGLE PROMPT.

The last 20% is you reading the slides, tightening the ones that got wordy, swapping your real numbers in, and fact checking everything.

That’s a slide deck, though.

Impressive, sure, but decks are forgiving.

The next example had a lawyer on the receiving end and 22 emails that flat-out contradicted each other.

That’s the one where I figured it’d finally fall apart.

Example 2: turn 9 months of email chaos into a clean timeline

If you’ve ever had to piece together what actually happened from a giant pile of emails, you already know it’s the kind of job that eats an afternoon and most of your patience. But occasionally, a situation arises where your inbox needs deep forensic analysis. Thankfully, the Director’s Brief can handle this too.

Here’s the setup for this example:

That same fictional company, Munder Difflin Scraper Company, is getting sued. A wholesale customer called Canyon Ridge Realty ordered 2,500 custom-engraved ice scrapers as client holiday gifts. The order shipped with the engraving wrong, the client says the gifts were useless, and now they want $54,900 in damages. There are 22 messy emails spread across about nine months of back and forth between sales, design, production, finance, and the customer. Somebody at Munder Difflin has to hand their attorney a clean timeline of what happened, and they don’t actually know the internal sequence themselves. That’s the whole reason they need it reconstructed.

Here’s the Director’s Brief we’re going to run:

**Role:** You're a litigation-support paralegal. You build clean, factual chronologies from raw evidence — the kind an attorney can rely on in a dispute. You're careful and precise, you don't speculate, and you never paper over a gap or a contradiction to make the story read smoothly.

**Goal:** Give me a defensible, dated timeline of events I can hand to my attorney — one that shows what happened, in what order, and what the email record does and doesn't actually establish.

**Task:** I've attached a pile of email files (`.eml` exports). Read all of them and reconstruct a single chronological timeline. For each event, give me the date, what happened, who was involved, and which email it came from so my attorney can pull the source. After the timeline, give me three more things: (1) anywhere the emails contradict each other, or someone's account doesn't match what the other emails show; (2) any decisions or hand-offs that clearly happened but aren't actually documented in the emails; and (3) a short, plain-language summary of how the whole thing unfolded.

**Context:** I work at Munder Difflin Scraper Company — we sell scrapers. A wholesale customer, Canyon Ridge Realty, ordered 2,500 custom-engraved ice scrapers as client holiday gifts. The order shipped with the engraving wrong, the client says they couldn't use them, and they're now suing us for damages. The emails span roughly September 2025 to February 2026 and involve our sales, design, production, and finance people plus the customer. I don't actually know the full internal sequence of who approved what and when — that's the whole reason I need this reconstructed. I'm not a lawyer, and my attorney wasn't on any of these emails, so this timeline is how they'll get up to speed.

**Constraints:**
- The filenames are a mess and won't sort in any useful order — go by the dates and timestamps inside the emails.
- When the exact wording of an email matters, quote it precisely rather than paraphrasing.
- Separate what was *proposed* or *assumed* from what was actually *confirmed* or *done*. If an email refers to a decision or conversation that isn't itself in the record, flag it instead of treating it as fact.
- Where two emails conflict, surface the conflict — don't pick a side or smooth it over.
- If it's relevant to who knew what, note when someone was added to or dropped from a thread.
- Don't speculate about anyone's intent and don't assign blame — stick to what the documents show.
- If something is unclear or missing, say "the record doesn't show" rather than filling the gap with a guess.
- Format it so my attorney can skim it fast: the timeline first, then the contradictions, then the gaps, then the summary.

NOTE: I attached the 22 .eml files to this brief, the actual litigation thread. I’m leaving them out of the article for brevity, but they’re in the downloadable kit if you want to run this yourself.

One thing worth knowing before you try this with your own pile of emails:

The chat upload caps at 20 files. I had 22, so I compressed all of the files and uploaded a single zip instead, and it read every email inside without me doing anything else.

EASY.

This entire prompt took about 8 minutes to process. It generated the full reconstructed timeline right there in the chat.

Here are some of the more interesting details the model generated based on the brief:

A 22-entry dated chronology with every single event cited back to a specific email ID, plus a source index that maps each ID to its original filename so the attorney can pull the source.
It caught the central contradiction. The project manager claimed “the client approved the final proof,” but the only proof anywhere in the record was Proof v2, which still had a [TAGLINE — confirm w/ Greg] placeholder sitting inside it. No final proof ever existed. The timeline flagged that no real approval ever happened.
It caught a Cc drop. The designer wasn’t on the email where the client actually sent the final tagline. The model reconciled the whole mess and flagged that “Greg’s ‘you’ and Marigold’s ‘me’ are not the same person,” then named that hand-off as the single largest documented gap in the record.
It caught a nine-day silence between the last unanswered file question and the ship confirmation, marked the engraving run itself as “(inferred)” with “no direct source,” and noticed that the only “response” the designer sent on the critical day was an automated out-of-office reply that production appears to have read as a green light.
A 6-point contradictions analysis and a 14-item list of decisions that clearly happened but were never documented anywhere. A two-line “make a timeline from these emails” prompt does not produce any of that.

I’ll be honest about this one. I’ve done something like this before, so I wasn’t floored. But it is genuinely, stupidly useful, because doing this by hand would have taken half a day of sticky notes, a spreadsheet, and a real headache by the end of it.

This took 8 minutes from a brief I (cough AI) wrote in about a minute.

Then I made a mistake on purpose, and it’s the most useful part of this whole example. Same session, I asked for a visual version of the timeline.

Just one line: “visualize this timeline for me in an artifact.”

No role, no goal, no context, no constraints. What came back was a prettier version of the same list. Not a real visualization, just a reskin.

I realized I had goofed the second I saw it. I handed it a sticky note and got a sticky note back. The timeline landed because it was briefed properly, and the visual flopped because it wasn’t.

So yeah... I basically just made the argument for this entire article by not using the Director’s Brief.

(Click here to download all three of the briefs and every output.)

Example 3: turn a messy travel wishlist into a 14-day travel itinerary

This is the fun one.

Planning a multi-stop family vacation with little kids is its own special kind of project management. It’s also exactly the kind of thing AI is supposed to be great at and usually isn’t, because people hand it two sentences and get back a generic listicle.

Here’s the setup for this example:

The Halpert-Beesly family, two parents and two kids ages 2 and 4, wants a Florida trip with a lot going on. Three days at Disney World. A road trip from Orlando down to Key West. A swim-with-dolphins excursion. A SpaceX rocket launch at Cape Canaveral. Then, after cramming all of that in, they need to get back to Orlando for the flight home.

The SpaceX launch is the one date that can’t move, so the entire trip has to be built around it.

Here’s the Director’s Brief we’re going to run:

**Role:** You are an experienced family travel planner who specializes in multi-stop trips with young kids. You understand that the linchpin of a complex vacation is the one immovable date, and you build everything else around it.

**Goal:** Produce a complete, day-by-day vacation itinerary for the Halpert/Beesly family — Tim, Tammy, and their two kids (ages 2 and 4) — that hits every planned activity, respects what's actually doable with toddlers, and is anchored to a real SpaceX launch date at Cape Canaveral.

**Task:**
1. Find the next viable SpaceX launch from Cape Canaveral in the planning window. That date is the linchpin — everything else builds around it.
2. Build a full dated itinerary covering, in whatever order best fits the linchpin:
   - 3 days at Walt Disney World (Orlando)
   - A road trip from Orlando down to Key West
   - A few days in Key West, including a booked "swim with dolphins" excursion (specific operator, location, booking notes)
   - The road trip back up to Cape Canaveral
   - The SpaceX launch viewing
   - Return to Orlando for the flight home
3. For each driving leg, give estimated drive time, the recommended route, and 2–3 kid-friendly stops to break it up.
4. For each overnight, give a lodging recommendation (kid-friendly, walking distance to activities when possible).
5. For Disney: which park each day, rope-drop or rest-day logic, and which rides cap at the 2-year-old's height/age.
6. For the launch: best public viewing spot, when to arrive, what to bring, and a backup plan if SpaceX scrubs (which they do frequently).
7. End with a one-page summary: total trip length, total drive time, total flight time, and the order of operations rationale (why Disney first vs. Key West first given the launch date).

**Context:** Tim and Tammy live in Scranton, PA. They're flying in and out of Orlando. The two kids are 2 and 4 — old enough for Disney, young enough that long driving days will end in tears. The trip is built around the rocket launch, so the launch date is the only fixed point — everything else flexes to accommodate it. The dolphin swim is non-negotiable and should be booked as a specific excursion, not "we'll figure it out when we get there." Disney is exactly 3 days. Key West stay is "a few days" — call it 3.

**Constraints:**
- Driving days with the kids cap at ~4–5 hours of actual drive time, broken into ~90-minute segments. Orlando → Key West is 7+ hours straight; that's a two-day drive with an overnight stop (Miami area is the obvious break).
- The launch is the linchpin. Pick the date first, then decide whether Disney goes before or after the Key West leg based on which order produces the lighter drive day before the launch.
- The launch viewing day needs a built-in backup. SpaceX scrubs are routine. The itinerary should hold one extra day near Cape Canaveral so a one-day slip doesn't kill the trip.
- The dolphin excursion needs to be a real, bookable thing — name the operator, location, price range, and whether it's age-appropriate for a 2-year-old (most aren't — flag that and propose the family-friendly alternative).
- Itinerary must end in Orlando for the return flight.
- Assume mid-range family budget. Don't recommend luxury resorts; don't recommend roadside motels with one star and a vibe.

No attachments on this one. The whole brief is self-contained, so you just paste it and go.

This entire prompt took about 4 minutes to process. It wrote a complete 14-day dated itinerary straight into a file.

Here are some of the more interesting details the model generated based on the brief:

It flagged that the 2-year-old legally could not swim with dolphins. No Florida Keys operator puts a kid under five in the water, and sometimes the cutoff is eight. It caught that on its own and pivoted, swapping the in-water swim for a dock-based Memory Maker Meet at the Dolphin Research Center, which allows ages one and up, with an optional splash add-on for the 4-year-old who actually could do it. I went and checked. The age restriction is real.
It flagged Disney ride height limits, ride by ride, for the 2-year-old, and recommended Rider Swap so the parents could trade off and still get on the rides. I didn’t ask for that and didn’t even know it was a thing.
It auto-built a one-day buffer near Cape Canaveral in case the launch slipped, because SpaceX scrubs constantly.
It named real launches off the actual manifest, BlueBird Block 2 Mission 3 and Transporter 17, both no-earlier-than mid-June 2026, plus a fallback note that Starlink launches go every three to five days if the headline mission gets pushed.

I didn’t prompt any of that. It just figured it out, and I was genuinely impressed.

Then... I did it again 🙄

I asked it to also render the itinerary as a PDF on top of the Markdown, and the PDF came out a little flat.

Not bad, just not as good as the file it was made from. Workable, not great. Same shape of issue as the timeline-artifact miss from example two. The brief was solid. The render step is where it gets soft.

A two-line prompt would have handed you a generic Florida itinerary that quietly booked a dolphin swim for a 2-year-old who wouldn’t be allowed in the water.

(Click here to download all three of the briefs and every output.)

Tip: Don’t type the brief, just ramble and let AI shape the mess

You might be looking at those briefs and thinking somebody who knew what they were doing wrote them, and that you’d freeze the second you hit the Role section with no idea what to type.

I get it, and I felt the same way.

So, I’ve been using a trick:

Turns out, you can just brain dump what you want to accomplish into AI and ask it to help you format the brief. It’ll take your messy thoughts, structure them, and write the brief for you.

I did that for each of the examples above.

I used a tool called Superwhisper, voice dictated some rambling thoughts into a Claude, and it formatted the brief for me.

It was SUPER EASY and it’ll kill that “I’m out of my league” feeling entirely.

Why one prompt beat every prompt pack you’ve ever downloaded

So let’s recap:

That was a board presentation, a litigation timeline, and a family vacation. Those are three jobs with almost nothing in common. They each have a different audience, different output, different stakes.

And the brief was the exact same shape every single time.

THAT is the real takeaway here.

It’s not that the prompt is clever. It just asks you the questions a capable person would’ve had to drag out of you anyway, except it asks them up front instead of halfway through the job when it’s already gone sideways.

The prompt just has a way of pulling all of that information out at the start.

I’ve been messing with AI every day for years, and anytime it feels like I’m fighting the model, it’s almost always because I briefed it like a Post-it note and expected an essay back.

The Director’s Brief is just me refusing to keep doing that to myself.

When you hand the model a role, point it at the destination, set the constraints, and let it go do its thing...

It usually comes back with something good.

Really good.

If this kind of thing is useful, hit subscribe and I’ll send the next one straight to your inbox.

Take care & see you in the next one ✌️

-Michael

P.S. I put all of the examples and prompts from this article in a downloadable kit you can find here. This article was a TON of work, and I did it all for free. If you got any value from this, it would mean the world to me if you could re-stack or drop a comment below. Thank you 🙏

Discussion about this post

Ready for more?