Zavmo Blog
Tip Sheet: How to Actually Talk to AI at Work
A one-page cheat sheet for getting real work out of AI. The habits that turn a vague answer into something you can use, and the lines to never cross.
Most people get weak answers out of AI for one reason. They ask it like a search engine. Talk to it like a sharp new colleague instead, one who is fast and eager but needs proper briefing, and the quality jumps.
Here is the one-page version. Save it. You will come back to it.
The Habits That Get Better Answers
Give it the brief, not just the ask. A vague prompt gets a vague answer. Tell it four things: what you want, who it is for, the format you need, and the tone. “Write something about our new policy” gets you mush. “Draft a short, warm email to the sales team explaining the new expenses policy, five sentences, no jargon” gets you something you can send.
Tell it who to be. Start with a role. “You are a careful finance manager reviewing this for errors.” It focuses the answer more than any clever wording.
Show it one example. If you want a certain style, paste one example of it. One good sample beats a paragraph of description.
Give it your rough version. Do not ask it to start from nothing. Hand it your messy notes and ask it to shape them. You get your thinking, done faster, instead of a stranger’s.
Push back. The first answer is a draft, not a verdict. “Too formal, make it plainer.” “You missed the deadline point, add it.” The second and third replies are where the good work is.
Ask it to show its reasoning. For anything that matters, add “explain how you got there”. It makes weak logic easy to spot.
The Habit That Matters Most
Check the answer. Always. This is the whole skill, and it is the half most people skip. AI will hand you a confident answer that is quietly wrong. A made-up figure. A stat with no source. A rule that does not exist. Treat every answer as a draft from a bright colleague who sometimes gets things wrong with a straight face. Your judgement is the point, not the tool’s.
If you cannot check it, do not send it.
The Lines Not to Cross
Do not paste anything you would not email outside the company. The UK’s National Cyber Security Centre puts it plainly: do not include sensitive or confidential information in queries to public AI tools. Client data, personal details, anything you would not want made public. If in doubt, leave it out.
Check your workplace’s AI rules first. Most organisations have them now. Knowing yours takes five minutes and saves a bad afternoon.
Own the output. When it goes out under your name, it is yours. “The AI wrote it” is not a defence anyone accepts.
Three Prompts to Steal Today
Sharpen your own writing. “Here is a message I am about to send. Point out anything unclear, too long, or easy to misread. Do not rewrite it, just flag it.”
Prep for a hard conversation. “I need to give a colleague difficult feedback about missed deadlines. Play them, and push back the way a defensive person would, so I can practise.”
Understand something fast. “Explain this in plain English as if I have ten minutes and no background. Then tell me the one thing most people get wrong about it.”
Where This Leads
Getting good at this is not a party trick. It is quickly becoming the baseline for staying sharp as your field changes. PwC found that jobs asking for AI skills now carry a real pay premium, and they keep growing. The people who point these tools well, and catch what they get wrong, are the ones who get ahead.
Want to build this properly, on your own real work?
→ Start Where You Are Today → More on getting confident with AI
Sources
- UK National Cyber Security Centre, guidance on ChatGPT and large language models (do not put sensitive or confidential information into public LLMs): ncsc.gov.uk
- PwC, 2025 Global AI Jobs Barometer (wage premium for AI skills): pwc.com
Zavmo is a learning companion that gets to know how you learn, not just what you learn. It builds these skills on your actual work, so what you learn is useful the same day.