The Unwritten Rules of Using AI at Work

A colleague forwards you a report. It reads beautifully — crisp, confident, well-structured. Halfway down you notice a statistic that cannot possibly be true, a source that does not exist, and a paragraph that repeats itself in slightly different words. You now know two things: they used AI, and they did not check its work. The second fact quietly changes how you read everything they send you from now on.

That moment — the quiet erosion of trust — is the real risk of AI at work. Not the technology itself, but the social and professional habits people build around it. Most organisations have written policies about which tools are approved and what data is off-limits. Almost none have addressed the unwritten rules: when to say you used AI, how to use it without offloading your judgement, and how to keep it from making you look careless in front of the people whose opinion of you matters most.

This is a guide to that softer, more human layer. It assumes you already use AI or want to. The question is how to do it in a way that earns respect rather than quietly spending it.

Rule 1: You Are Still the Author

The single most important principle is also the simplest. When you send something, you are vouching for it. It does not matter that a model drafted the first version. Once your name is on it, every claim, every number, and every recommendation is yours.

This changes how you should treat AI output. It is a draft from an unusually fast, occasionally unreliable junior colleague — never a finished product. The people who get into trouble are the ones who treat generated text as done. The people who thrive treat it as raw material they shape, question, and correct.

In practice this means three habits become non-negotiable:

• Verify every factual claim. Names, dates, figures, legal points, and citations are exactly where models confidently invent things. If you cannot confirm it independently, cut it or flag it as unverified.
• Read the whole thing as yourself. Not skimming for typos — reading for meaning. Does this actually reflect what you think? Would you say it out loud in a meeting?
• Rewrite anything that does not sound like you. AI has recognisable tics: hollow enthusiasm, tidy lists of three, phrases like "in today's fast-paced world." Colleagues notice. Strip them out.

If you would not put your name on it after a proper read, it is not ready — no matter how polished it looks.

Rule 2: Know What Never Goes Into a Chatbot

This is where casual AI use becomes a genuine liability. Every prompt you type is data leaving your control. Depending on the tool and settings, it may be stored, reviewed by humans, or used to train future models. Treat the input box like a postcard, not a private diary.

Some categories should never be pasted into a general consumer AI tool without explicit, informed approval:

• Client or customer personal data — names tied to sensitive circumstances, health information, financial details, anything covered by data protection law.
• Confidential commercial information — unreleased financials, deal terms, pricing strategy, unannounced products, board material.
• Credentials and secrets — passwords, API keys, internal system details, security configurations.
• Anything under NDA — if a contract restricts who can see it, an AI vendor counts as a third party.
• Colleagues' private information — salary data, performance notes, disciplinary matters, personal circumstances shared in confidence.

The safe test is a single question: would I be comfortable if this text appeared in a screenshot shared outside the company? If the answer is no, do not paste it. When you genuinely need AI on sensitive material, use an enterprise-grade tool with a data-processing agreement and confirmed retention settings — not the free version you happened to have open. This is one of the first things we help organisations lock down when we design enterprise rollouts.

If you would not email it to an unknown third party, do not paste it into a chatbot. The convenience is never worth the exposure.

Rule 3: Disclose When It Matters — and It Often Does

The disclosure question causes the most anxiety, because there is no universal rule. Nobody expects you to announce that AI helped you rephrase an email or brainstorm a subject line, any more than you would disclose using a spellchecker or a search engine. But there is a spectrum, and knowing where the line sits protects your reputation.

When disclosure is unnecessary

Routine assistance on work you fully own and have reviewed — drafting, summarising, reformatting, ideation, code you understand and tested. The AI was a tool, the judgement was yours, and you stand behind the result. No disclosure needed.

When disclosure is courteous

When someone might reasonably assume more human effort went in than did. If a colleague thanks you for a "thoughtful" analysis that a model largely produced, a light "I used AI to get the first draft together, then worked through it" is honest and costs you nothing. It also models good practice.

When disclosure is essential

Whenever the human origin of the work is part of its value or its integrity. Client deliverables where personal expertise is what they are paying for. Anything with legal, safety, or compliance weight. Creative work presented as original. Contributions to decisions where people deserve to know a machine shaped the recommendation. Here, silence is not neutral — it is misleading.

The reliable instinct: if someone would feel deceived on learning AI was involved, tell them beforehand. Trust survives disclosure. It rarely survives discovery.

Rule 4: Do Not Outsource the Thinking You Are Paid For

There is a subtler etiquette failure than a hallucinated statistic: using AI to skip the exact thinking your role exists to do. A strategist who lets a model set the strategy. A manager who has AI write the feedback they should have shaped themselves. A specialist who stops forming their own view because the machine offers one instantly.

Colleagues can feel this even when they cannot name it. Contributions become generic. Meetings fill with plausible, characterless input that could have come from anyone — because, in a sense, it did. Your distinctive value fades.

The discipline is to use AI on the parts that genuinely benefit from speed — first drafts, research scaffolding, formatting, exploring options — while keeping the judgement calls firmly human. Before you send AI output onward, ask: what did I add that a prompt alone could not? If the honest answer is "nothing," you have become a relay, not a professional. Building that instinct for where to lean in and where to stay in the driver's seat is exactly what strong foundational AI training teaches.

Rule 5: Respect the Room's Comfort Level

AI adoption is uneven, and etiquette means reading that. Some colleagues are enthusiastic; others are anxious about their jobs, sceptical of the quality, or quietly worried they are falling behind. Bulldozing that with evangelism damages relationships and, ironically, slows adoption.

• Do not weaponise speed. Producing ten times the output and making others look slow breeds resentment, not admiration. Share the method, not just the results.
• Do not shame the cautious. "You still do that manually?" lands badly and shuts people down. Curiosity spreads far faster than superiority.
• Do offer to help. The person who quietly shows a nervous colleague one useful workflow earns more goodwill than the one who broadcasts their productivity.

This is also where AI champions genuinely earn the title — not by being the flashiest user, but by making the tools feel accessible and safe for everyone else. If that is the role you want to grow into, our Become the AI Champion programme is built around exactly this kind of influence.

Rule 6: Be Honest About AI in Meetings and Collaboration

A few situational norms are worth naming, because they come up constantly:

• Live note-taking and transcription tools. Tell people if a bot is recording or transcribing a call. Consent is not optional, and in many places it is a legal requirement. A silent recorder in a meeting is a fast way to lose trust.
• AI-generated meeting summaries. Useful, but check them before circulating. A summary that misattributes a decision or invents an action item creates real confusion.
• Shared documents. If you drop AI-generated content into a collaborative doc, say so, so others know to scrutinise rather than assume it is settled.
• Responding to people. A fully AI-written reply to a heartfelt message from a colleague or client will usually be sensed, and it stings. For anything relational, let the machine help you think and then write it yourself.

Rule 7: Build Team Norms Before You Need Them

Most AI etiquette problems are really the absence of shared expectations. When nobody has agreed what "good" looks like, everyone improvises, and the gaps show. Teams that get ahead of this tend to agree a few things explicitly:

• Which tools are approved for which kinds of data.
• What must always be human-reviewed before it leaves the team.
• When disclosure is expected internally and with clients.
• How to share prompts and workflows so good practice spreads instead of staying siloed.

These do not need to be a heavy policy document. A one-page shared understanding, revisited as tools change, does the job. The point is that etiquette is far easier when it is a collective agreement rather than a series of private guesses. Leaders who want help drawing these lines will find it a core part of how we advise business leaders on rollout.

The Quiet Payoff

None of these rules are about slowing you down or dampening your enthusiasm. They exist because the people who use AI well — visibly, honestly, with their judgement intact — end up more trusted, not less. They become the person others ask, "how did you approach that?" rather than the person whose work quietly gets double-checked.

The technology is the easy part. Anyone can learn to prompt. The rarer, more valuable skill is using these tools in a way that strengthens your professional reputation instead of undermining it — and that is entirely a matter of habit and etiquette. Get those right, and AI stops being a risk to your credibility and becomes one of the clearest signals of it.