How to use AI responsibly at work

Most guidance on responsible AI use at work is written for the organisation: approved tools, data classification, governance. It matters, and it is not enough. The decisions that actually determine whether AI makes your work better or worse are made by you, alone, dozens of times a day, usually in under a minute. Should I use it for this? Do I trust this number? Do I mention that the tool wrote most of it?

These five habits are for those moments. They apply whatever tool you use and whatever your company's policy says, and the policy still applies on top of them.

Habit 1: Choose when to use it, and when not to

The strongest reason to use AI is that it saves time on something you already know how to judge. The weakest reason is that the task feels hard. When you reach for the tool because you do not know how to do something, you lose the ability to tell whether its answer is any good.

The last column is not a ban. It is a pause. Writing your first board summary yourself, badly, then comparing it with what the tool produces, teaches you far more than reading the tool's version first. An apology drafted by a tool tends to read like one, and the person receiving it usually notices.

Habit 2: Verify before your name goes on it

AI tools produce fluent, confident text whether or not it is correct, and fluency is not evidence. The more precise and convenient a claim looks, the more it deserves checking: a specific percentage, a named study, a quotation, a legal or financial figure. A practical way to verify:

1. Trace every number to a source you can open. If the tool cannot point you to one, or the source does not say what the tool claimed, the number does not go in.
2. Recompute anything you can. Totals, growth rates and conversions take a minute to check and are a common place for errors.
3. Check names, dates and quotations against the original. Plausible-sounding citations that do not exist are a known failure of these tools.
4. Ask for the opposite case. Ask the tool for the strongest argument against its own answer, and see whether it holds.
5. Read it as the recipient would. Would you be comfortable if they asked where did this come from? about every sentence?

Treat a fluent, confident answer exactly as you would a hedged one. Confidence is a writing style, not a sign of accuracy.

Habit 3: Own the result, and say how it was made

If AI-assisted work goes out with a mistake in it, the mistake is yours. The tool cannot be accountable, so the person who sent the work is. That is not a reason to avoid the tools; it is a reason to put your name only on work you have checked.

Disclosure follows from ownership. When AI did a substantial part of a piece of work, say so when you hand it over, in plain words: I used AI to draft the market section and checked every figure against the sources listed. The recommendation is mine. That sentence tells your manager what to trust and what to probe. Hiding it makes every later discovery look like deception, even when the work was fine.

Ownership also covers data. Never paste client, company or personal information into a tool your organisation has not approved for it. If you see a colleague doing it, say something, privately and early: it protects them as much as the client. This is a conversation many people avoid, and how to have difficult conversations at work applies directly.

If you want an honest read on these first three habits before going further, the free AI wisdom check puts fifteen realistic scenarios to you and scores how you actually handle them, with a written report naming your strongest and weakest habit. It takes about four minutes.

Habit 4: Think first, then ask

On work that matters, form your own answer before you ask the tool, even a rough one in two lines. Then compare. If the tool agrees, you have a check. If it disagrees, you have something to think about, which is far more useful than a conclusion you never questioned.

The opposite habit, asking first and editing the answer, is faster and quietly dangerous. You start judging the tool's framing instead of the problem, and over time you lose the ability to frame problems yourself. When the model's conclusion contradicts the view you walked in with, neither should win by default: find out why they differ. That is critical thinking applied to a new kind of colleague, and the heart of AI literacy.

Habit 5: Keep building the skill the tool is doing for you

The honest test: if the tools disappeared tomorrow, could you still do your core work well? If the answer is drifting towards no, the tools are making you faster at the cost of making you less capable, and that trade gets worse over a career.

• Do some core work unaided on a regular basis, the way a pilot still flies manually.
• When someone finds an error in your AI-assisted work, learn from the error, not just the fix. Was it a verification step you skipped? A task you should not have handed over?
• Invest in the skills AI raises the value of: judgment, clear writing, verification, and the people skills that no tool delivers on your behalf. Kolb's experiential learning cycle is a good model: do, reflect, draw the lesson, try again.

Three situations, handled well

A precise figure, due in an hour. The tool gives you a confident market size for a client deck. You cannot trace it to a source in the time available. The responsible move is not to use it as fact: give a range from a source you can open, or say the figure is being confirmed and send it later. A late, correct number costs you an hour. A wrong one costs you the client's trust in every other number in the deck.

An error you already sent. You find a mistake in AI-assisted work that went out yesterday. Correct it yourself, quickly, before anyone else finds it, and say plainly what was wrong. People forgive errors that are owned early far more readily than errors they discover.

A difficult personal message. You need to apologise to a teammate you let down. You can use a tool to help you think about what to say, but write the message yourself, in your own words, and deliver it in person or by voice if you can. The effort is part of the apology.

A pre-send checklist

Before AI-assisted work leaves your hands, five questions:

1. Did I choose to use AI here for a good reason, or because the task felt hard?
2. Has every number, name and quotation been checked against a source I opened?
3. Did any confidential or personal information go into a tool not approved for it?
4. Will the recipient know which parts AI helped with?
5. Could I explain and defend every sentence if asked?

If any answer is no, fix it before you send. The checklist takes about a minute. The conversation after a confident wrong figure reaches a client takes a great deal longer.

For managers

These habits are personal, but they spread or fail as a team. If people see a manager disclose AI use, admit an AI-assisted mistake and reward someone for catching one, they will do the same. If you are setting the ground rules for a whole team, our guide on how to prepare your team for AI covers the team-level working agreement, and the AI judgment check measures the team's habits rather than one person's.

Tour De Force runs this as live, experiential training for teams worldwide, online and in person. Talk to us, or play The Weekly Challenge from our gamified learning apps to see the method in ten minutes.