How to Use AI at Work Without Losing the Human Touch
The first email I ran through an AI tool at work came back grammatically perfect and completely forgettable. Every sentence was correct. None of them sounded like me. I deleted it and started over by hand, which mostly defeated the point of using the tool at all.
Looking back, the problem wasn't the tool. It was the prompt. I'd typed something close to “write a team email” and expected a specific, in-my-voice draft out of a vague ask.
That's the exact gap the checklist I hand out in my “Ask Better, Get Better” workshops was built to catch: five moves — Clear, Specific, Contextual, Accurate, Ethical — that turn a weak prompt into one that actually gets you something usable. Two of those five, Accurate and Ethical, do double duty in this post, because they're also where AI at work most often goes sideways after the draft comes back.
That's usually where people give up on AI at work: the tool works, technically, but the output reads as if nobody in particular wrote it. The fix isn't avoiding the tool. It's using it differently, one small task at a time, with your own judgment, your own voice, and a decent prompt checklist still doing the real work.
Here's a simple way to start, with two of those five moves built in along the way.
Start with one small, real task for AI
Don't hand AI your whole job in the first week. Pick one repeatable task — a meeting recap, a first-draft email, or an answer to a question your team gets asked constantly — and use AI just for that. A narrow, low-stakes task teaches you what the tool is actually good at, and where it needs help, faster than trying to overhaul how you work all at once.
Stay in the loop on accuracy
AI drafts with total confidence whether or not it has the facts right. That's the Accurate move in the five-move prompt framework I teach in my “Ask Better, Get Better” workshop: before a draft goes out, check every number, date, name, or claim in it against something you could actually prove. AI trusts whatever you feed it, and it can invent details of its own just as confidently. Verifying isn't optional. It's the part of the job that stays yours.
Edit for your own voice, not AI’s
Treat the first draft as raw material, not a finished product. Read it out loud. If a sentence doesn't sound like something you'd actually say, rewrite it, even if it's technically fine. This is where I've put the most work myself, and it's worth walking through, because it's a skill you can build on purpose.
How I taught my own AI to sound like me
I've spent years developing an oral storytelling instinct: the habit of opening on a scene instead of a claim, using one concrete detail instead of five vague ones, and letting a little personality through even in professional writing. When I started using AI regularly, the drafts kept flattening that instinct into the same handful of generic openers and hedges — “In today's fast-paced world...” “It's important to note that...” “Navigating the landscape of...” None of that is how I talk, and none of it is how I wanted to sound in print.
So I trained my AI collaborator out of it, the same way I'd coach a student. Here's the process, if you want to build the same thing for your own voice.
1. Notice your own tells first. Before you can fix the AI's voice, name yours. What do you naturally do when you're telling someone a story out loud? Mine is opening on a scene and reaching for a little dry humor. Yours might be starting with a question, a specific number or a flat one-line joke.
2. Build a short “cut list.” Keep a running list of the phrases the AI reaches for that don't sound like you. Mine included the ones above, plus “delve,” “unlock” and “seamless.” The list gets longer the more you use the tool, and that's normal.
3. Replace, don't just forbid. “Don't do that” isn't a full instruction. Tell the tool what to do instead: open with a real scene or a specific person, not a general statement; use first person; keep the wit.
4. Test it on real drafts, not hypotheticals. Correct the tool in the moment, on an actual piece of writing you need anyway, rather than running practice prompts. The correction sticks better when it's attached to something real.
5. Save it so you're not re-teaching it every time. Once you have a working list of do's and don'ts, keep it somewhere you can hand back to the tool at the start of a new project, instead of rebuilding it from scratch each time.
Do this once, on purpose, and every draft after gets a little closer to sounding like you instead of like a template.
Know where the line is
Not everything belongs in an AI draft. The Ethical move in the same five-move framework is a good gut check: if you wouldn't do it yourself, don't ask AI to do it either — impersonate someone, fabricate a firsthand account, or make a call that should carry your name and your judgment, not a tool's. Use AI for organizing and first drafts. Keep the judgment calls, the hard conversations, and anything that should sound like a real account of what happened for yourself.
Culture still runs on people
It's worth saying plainly: none of this works if the team culture around it is broken. SHRM's 2026 workplace research found that even as AI use grows, the real driver of whether people stay engaged and stick around is human leadership and culture, not the tools. Ninety-one percent of workers who feel their organization actually meets their needs report being satisfied in their jobs, and workers who don't feel that way are considerably more likely to leave within the year. AI habits should support that kind of culture. They shouldn't be asked to substitute for it.
Where to go from here
None of these five moves takes long to learn, but they're a lot easier to build as a team than to figure out alone. My “Ask Better, Get Better” workshop walks a group through the full Clear, Specific, Contextual, Accurate, and Ethical framework in about an hour, with a take-home handout so the habit sticks after the session ends. If your team is starting to use AI at work and you want a shared, simple way to do it well, that's exactly the conversation I'd love to have.
Comments