Why Your AI Prompts Don't Work at Work (The Fix Big Labs Recommend)

You’ve seen the headlines: “150 AI prompts to save 3 hours a day.” You copy a few, paste them into ChatGPT, and the answers come back… fine. Generic. Not wrong, but not the thing you’d have written either. So you rewrite half of it, and the time you were supposed to save is gone.

The problem usually isn’t ChatGPT, and it isn’t you. It’s that a prompt copied from a list has no idea who you are, what you’re working on, or what you actually need back. The big AI labs that build these tools all publish the fix, and it’s the same fix: add four things — your role, your context, the real task, and the output you want.

Below, pick a task you have this week. Flip between the prompt most people type and the lab version, see why the second one lands, then fill in the blanks and open it straight in ChatGPT.

You are me, a office manager, writing to the 8-person team.

Write a short email telling them: the client deadline moved to next Friday

Context they need:
the client pushed our review call to Thursday, so we get two extra days. Nothing else on the timeline changes.

Keep it calm and brief, not alarming. End with the one action I want them to take. If anything is unclear, ask me before writing.

The rest of this post is short. It explains the one pattern under every “after” prompt above, so you can stop collecting lists and write your own.

The problem isn’t ChatGPT. It’s the prompt.

Here’s a thing the prompt lists won’t tell you: the same model that gave you a generic answer can give you a great one. Nothing changed about the AI. What changed is how much it had to guess.

When you type “write an email telling the team the deadline moved,” ChatGPT has to invent everything you left out. Who’s writing? A manager or a peer? Why did it move? Is this calm news or a fire drill? What should people do now? It fills those gaps with the blandest safe defaults, because that’s the average of everything it has seen. You get the average. That’s the “generic” feeling.

A copied prompt from a list has the same problem, one level up. It was written for a generic person doing a generic version of the task. Your task isn’t generic. You have an actual team, an actual reason, an actual tone your company uses. None of that is in the list. (If your answers feel generic for a different reason — vague, hedged, no real opinion — that’s a separate fixable problem.)

Even people who use AI all day fall into this. The one-liner is faster to type, so you type it, get something mushy back, and rewrite it by hand — when two sentences of context up front would have gotten it right the first time. So if your prompts feel like this, it isn’t a sign you’re bad at AI. It’s a missing habit, and habits are easy to fix.

The one pattern behind every good work prompt

Look back at any “after” prompt above and you’ll see the same four parts. This is the whole trick.

1. Role — who the AI is being, or who you are. “You are me, an office manager.” This sets the voice and the assumptions.
2. Context — the real details. The actual notes, the actual numbers, why the thing is happening. This is the single biggest lever. Most “bad” answers are just missing context.
3. Task — what you want done, stated plainly and specifically. Not “help with,” but “turn these notes into a one-hour agenda and list the decisions we need to make.”
4. Output — the shape you want back. A table. One page. Two subject-line options. A summary plus a chart. If you don’t say, you get prose.

That’s it. Role, context, task, output. You don’t need to memorize a single prompt if you have this, because you can build any prompt you need on the spot.

If you only ever add one of the four, make it context. A copied list can already give you a role and a task. What it can’t know is your actual situation — your team, your numbers, your reason. That’s the part that turns a generic answer into yours.

A quick way to feel the difference: take your last weak prompt and ask yourself which of the four you skipped. It’s almost always context, output, or both.

What OpenAI, Google, and Anthropic actually recommend

This isn’t my framework. It’s what the companies building these models tell their own business users, and they say it independently.

OpenAI’s small-business guide walks through tasks like analyzing a spreadsheet or drafting a campaign, and every example prompt names the role, hands over the real data, states the task, and asks for a specific format — a chart with an executive summary, a campaign kit with five named pieces (OpenAI, “Think Bigger: How Small Teams Win with ChatGPT”, 2026).

Google’s prompting guidance for Workspace says the same in plainer words: write in full sentences like you’re talking to a person, be specific, and give as much context as you can (Google Workspace, “Writing Effective AI Prompts”, checked June 2026).

Anthropic ships its small-business product with ready-to-run workflows built the same way — each one grounded in your actual tools and data rather than a generic request (Anthropic, “Introducing Claude for Small Business”, 2026).

Three competitors, three guides, one answer. When OpenAI, Google, and Anthropic all tell their users to do the same thing, it’s a safe bet it works regardless of which AI you use.

Copy these, then make them yours

The builder above isn’t a list to hoard. It’s a way to feel the pattern a few times until you can do it without thinking. Here’s how to make it stick:

1. Next time you’re about to type a one-liner, stop. Add the four parts: role, context, task, output. Ten extra seconds.
2. Front-load context. Paste the real notes, the real numbers, the real past email. Context is the part the AI can’t guess and the part that fixes most bad answers.
3. Always name the output. “Give it back as a table / one page / two options.” This alone removes most of the rewriting.

You don’t need to learn prompt engineering. You need these four parts and the habit of using them. That’s the whole difference between a generic answer and one you can actually send.

If your week is mostly marketing work, the same four parts turn into a task-by-task prompt builder for email, social, ads, and campaigns.

If you build something useful with this — or hit a task where it falls apart — I’d like to hear about it. Get in touch.