How to Use AI at Work: A No-BS Starter Guide

28% of employed American adults are already using ChatGPT at work. That's not a prediction. That's OpenAI's own data, and it means roughly one in four of your coworkers is already using AI at work while you're still reading articles about whether you should start.

So let's skip the philosophical debate and get to the actual question: how do you use AI at work without it being a whole thing?

The answer is simpler than LinkedIn wants you to believe. No $997 course. No "prompt engineering certification." No six-week bootcamp. You need two things: a tool and a place to start. This guide gives you both.

Where most people get stuck when they try to use AI at work

They start in the wrong place.

Someone tells them "AI can do everything," so they sit down and stare at a blank chat window trying to think of something profound to ask. Nothing comes. They type something weird. The answer is generic. They close the tab and conclude AI is overhyped.

Or they go the other direction. They download 11 different tools because a newsletter said these were the "top AI apps for productivity." Now they have subscriptions they don't use and zero clarity on which one does what.

Both of these are tool problems masquerading as motivation problems. The real issue is that nobody told them where to actually start.

Start with the work you hate (this is the whole strategy)

Here's the only framework you need: find the 40% of your job that makes you want to check your phone every four minutes and do that part with AI first.

Every knowledge worker has it. The part of the job that's technically necessary and entirely soul-draining. The weekly status report nobody reads. The first draft of the deck you'll rewrite six times anyway. The email explaining, again, why the timeline slipped. The summary of a meeting that didn't need to happen.

This is not a coincidence. The McKinsey Global Institute has been saying for years that roughly 60-70% of tasks that eat time at work are automatable in some form. The tedious stuff is tedious precisely because it's repetitive and structured, which is exactly what AI is good at.

So don't start with "what can AI do?" Start with "what do I hate doing?" Make a short list. Then we'll get to work.

Here's what that looks like by role:

Pick one task from your version of this table. That's your starting point.

How to use AI at work without getting garbage back

You get garbage out when you put garbage in. This is not a metaphor. It's exactly how these tools work.

The most common mistake: typing a vague one-liner and expecting a finished product. "Write me a marketing email." Okay, for what product, to what audience, with what tone, trying to get them to do what? The tool doesn't know. So it guesses, and the guess is usually beige.

Treat it like a smart intern. Not a genius. Not an idiot. A smart, capable person who just started this week and doesn't know your industry, your company, your audience, or your context. You'd never hand a new intern a task with no background. Give it the same briefing you'd give them.

The format that actually works:

Role: Tell it what it's acting as. "You're a copywriter for a B2B SaaS company."

Context: Tell it the situation. "We're launching a new feature for finance teams. The audience is CFOs who are skeptical of new software."

Task: Be specific. "Write a 200-word email announcing the launch. Lead with the problem it solves, not the feature."

Format: Tell it what you want back. "Plain text. Short paragraphs. No bullet points."

That's it. Four things. Role, context, task, format. You don't need to "master prompting." You need to brief the intern properly.

If the first answer isn't right, don't start over. Reply and tell it what's wrong. "Too formal. Make it more direct." "The opening is weak. Try again." This is a conversation, not a vending machine.

Which tools to actually use (stop downloading things)

There are exactly two tools you need: ChatGPT and Claude. Pick one. Maybe both. That's the list.

Everything else, and there are hundreds of them now, is mostly a wrapper around one of these two. A wrapper is when someone builds a slightly different interface on top of an existing AI model and charges you for the interface. Sometimes the interface is good. Usually it's not worth the subscription when you can just talk to the source directly.

The author of Don't Replace Me put it plainly: "You don't need anything but Claude or ChatGPT. Everything else is mostly wrappers." When someone who builds AI tools for a living says stop collecting tools, it's worth listening.

ChatGPT is slightly more familiar to most people and has a massive ecosystem of plugins and integrations if you want them later. Claude tends to write better prose and handles longer documents more gracefully. Try both for a week on the same tasks. Stick with whichever one feels more natural.

The free versions of both are genuinely useful. The paid tiers ($20/month each) unlock longer context windows and faster responses, which matters if you're working with big documents. Start free, upgrade if you hit the limits.

That's the whole tool decision. Two choices. No spreadsheet needed.

A real example: what this actually looks like

A marketing manager at a mid-size software company was spending three hours every Monday writing a newsletter nobody seemed to read. Not because the newsletter was bad, but because she was starting from scratch every single time. Pulling stats, finding an angle, writing, rewriting, getting the formatting right.

She started using ChatGPT. Not for the whole newsletter, because AI writing newsletters from scratch produces newsletters that feel like they were written by a committee. But for the scaffolding.

She'd paste in three or four links from the week, plus her rough notes on what mattered, and ask for a structured outline with a suggested angle. Five minutes instead of 45. Then she'd write the actual thing herself, faster, because she wasn't staring at a blank page.

Two hours back every week. Not because AI replaced her job. Because she stopped doing the part she hated manually.

That's the pattern. You're not outsourcing the judgment. You're outsourcing the scaffolding.

The disclosure question (yes, you should think about this)

At some point you'll wonder: do I need to tell anyone I'm doing this?

Short answer: it depends, but the answer matters less than you think right now, and more than you'll think later.

About 68% of people who use AI at work don't tell their managers. Some because the company culture would punish it. Some because they're not sure if it's "allowed." Most because they're using it for drafts and summaries and it never comes up.

The question gets more complicated as you rely on it more, or when the output matters more, like a legal document or a medical summary or anything where accuracy really counts. For a first draft of internal communications? Nobody's checking the provenance of your meeting summary.

If you want to think through the workplace politics of this more carefully, there's a full breakdown at /should-i-tell-my-boss-i-use-ai/. For now, just start.

What "getting good at AI" actually means

This is where everyone overthinks it.

Getting good at using AI at work doesn't mean becoming a technical expert. It doesn't mean learning to code. It doesn't mean taking a course or getting a certification or reading a book on prompt engineering (which is, for the record, largely a dead discipline).

It means developing decent instincts. What kinds of tasks does it handle well? What does it hallucinate? When should you trust it and when should you double-check? How do you brief it quickly enough that using it is actually faster than not using it?

Those instincts come from use. From trying it on your actual tasks, seeing what comes back, correcting it, trying again. After about two weeks of regular use, you'll have a feel for it that no amount of reading about it could give you. That's the thing people keep skipping. They're still reading articles about AI when they should be typing into a chat window.

If you want to go deeper on the skills side of this without falling for the "learn to code" trap, ai-skills-for-non-technical-people is worth a read. And for practical prompts you can steal for emails, reports, and presentations, /how-to-use-chatgpt-at-work/ has a bunch of them broken down by task type.

Your first 72 hours with AI at work

Don't try to change your whole workflow at once. Pick one task. The most tedious one on your list.

For the next three days, every time that task comes up, do it with AI first. Give it the four-part brief. See what comes back. Adjust. Reuse the brief next time. Refine it slightly. By the end of the week, you'll have a working template you can reuse in 30 seconds.

That's the whole move. One task. Three days. One template.

Then pick the next one.

Most people who get good at using AI at work didn't do a course. They just started small and kept going. The tools got easier because they used them. The instincts developed because they paid attention to what worked.

"AI is just your keyboard now," as one builder of AI tools put it. You don't study the keyboard. You just start typing.

Frequently asked questions

How do I start using AI at work if I've never used it before?

Start with the task you find most tedious, not the most impressive one. Pick either ChatGPT or Claude (both have free tiers), and type a brief that includes your role, the context, the specific task, and the format you want. Don't expect perfection on the first try. Treat it like a conversation.

Do I need to pay for AI tools to use them effectively at work?

Not to start. The free versions of ChatGPT and Claude handle most common work tasks: drafts, summaries, outlines, editing, brainstorming. The paid tiers ($20/month) unlock longer context windows and faster responses, which matters if you're regularly processing long documents or hitting rate limits.

Will using AI at work get me in trouble?

It depends on your workplace. About 68% of people who use AI at work don't tell their managers. Most companies don't have clear policies yet. The main risks are confidentiality (don't paste sensitive client data into public AI tools) and accuracy (always check factual claims). See should-i-tell-my-boss-i-use-ai for a fuller breakdown.

What's the difference between ChatGPT and Claude?

Both are large language models that can help with writing, analysis, and research. ChatGPT has a larger ecosystem of integrations. Claude tends to produce better long-form writing and handles large documents well. Try both on the same task and pick whichever feels more natural to you.

Do I need to learn prompt engineering to use AI at work?

No. Prompt engineering as a discipline is largely dead as a specialized skill. What you need is the ability to give clear, specific instructions: role, context, task, format. That's it. You already know how to brief a colleague. This is the same thing.

How long does it take to actually get good at using AI?

Two weeks of regular daily use. Not studying. Not taking a course. Just using it on real tasks and paying attention to what works. Most people who become comfortable with AI at work didn't spend any time formally learning it. They just started and kept going.