Using ChatGPT at Work: Is It Actually Obvious?

Short answer: probably not as obvious as you think, and that is not really the question worth asking. Most of your colleagues are already doing it. Microsoft and LinkedIn's 2024 Work Trend Index found that 78% of people who use generative AI at work bring their own tools rather than ones their employer gave them — and that share is higher, 80%, at small and medium-sized companies. Using ChatGPT at your desk is not the outlier behaviour it might feel like.

What is more interesting is that people use it and still don't want to say so. The same Work Trend Index survey found 52% of AI users are reluctant to admit using it for their most important tasks, and 53% worry that using it makes them look replaceable. That is the actual tension behind "is this obvious" — not a fear of being caught, but a fear of what being caught would imply. Worth separating those two, because the fix for each is different.

What actually gives it away (and what doesn't)

Writing style is the thing people worry about and the thing that matters least. Tone is easy to adjust — ask for shorter sentences, your usual phrases, less of the summary-paragraph-at-the-end habit, and most of the "obviously AI" feel goes away in one more prompt. AI Human Prompt: How to Make AI Text Sound Like You covers exactly that adjustment if a draft keeps coming back sounding like nobody.

The real tell is a fact nobody in the room can confirm. A language model predicts the next plausible piece of text; it does not have a separate step where it checks whether something is actually true before it says it. A survey of hallucination in large language models documents this directly: fluent, confident answers whether or not the facts behind them are real. A client call summary that states a contract renewal date nobody mentioned, or a number that is close but not the one from the actual spreadsheet, is what gets noticed in a meeting — not the sentence rhythm.

A worked example

Say you ask ChatGPT to turn rough notes from a client call into a follow-up email. The first draft reads smoothly and includes a line like "as agreed, the renewal will proceed in Q2." If your notes never actually said that — if the call ended with "we'll confirm the renewal timing next week" — that sentence is invented, not summarised. It will read exactly as confidently as the parts that are true, which is the whole problem.

The fix is not a better prompt alone, though OpenAI's own prompting guidance does help by having you state the facts you actually have rather than asking the model to "write a follow-up" and hoping it fills gaps sensibly. The fix is reading the draft against your notes, line by line, before it goes anywhere — the same habit How to Check an AI Answer When You Are Not the Expert walks through for a case where you cannot independently verify the subject at all.

Check the policy before you check the tone

Before worrying about whether a draft sounds like a machine, find out what your employer actually says about pasting work content into a public AI tool. That is the real risk, and it is boring compliance detail rather than a writing problem. If you use ChatGPT, read what happens to the text you type in — OpenAI's data controls FAQ sets out the actual settings, including what you can turn off. A client's contract details or a colleague's performance review are exactly the kind of content worth checking your company's stance on first, not after.

There is a wider trust question here too. Pew Research on how Americans view AI use to monitor and evaluate workers found real unease about AI systems making judgements about employees without a clear, disclosed process — the same discomfort, from the other direction, that makes a colleague uneasy discovering your draft was AI-written and undisclosed. Disclosure is not an admission of weakness. It is the thing that keeps the trust question from becoming a bigger one later.

Use it openly instead of hiding it well

Automate Work Tasks: Draft First, Approve Second is the right shape for most of this: ChatGPT produces a first pass, a person checks it against the actual facts, and the final version goes out under your name because you stand behind it, not because nobody can tell how it was drafted. That is a defensible way to work with AI at your desk. Trying to make the output undetectable is not — it optimises for the wrong thing and leaves the actual risk, an unchecked fact, exactly where it was.

For recurring work communication, ChatGPT Prompts for Business: A Formula, Not a List is worth reading for the structure rather than any single prompt — naming the audience, the facts you actually have, and the format you need, so the first draft already matches how you write instead of needing a tone pass afterwards. ChatGPT Cheat Sheet covers the adjacent case: the handful of tasks ChatGPT is reliably strong at, which is mostly transforming text and data you already supplied, not recalling a fact from nowhere.

What to do Monday

1. 01Find your company's actual policy on AI tools and work data, in writing, once — not a guess about what is probably fine.
2. 02Use ChatGPT to draft from notes, data or a document you already have, never to supply a fact you have not sourced yourself.
3. 03Read the draft against your own notes before it goes out, specifically checking names, dates, numbers and anything stated as agreed.
4. 04If the research behind a task is multi-step rather than one draft, Deep Research Prompts has a template built for that, with the same checking habit built in.

Where this actually pays off

Hiding AI use well is a skill with no future value. Using it accurately and being able to show your work is one that is already being rewarded: PwC's 2026 Global AI Jobs Barometer found a 62% wage premium for workers with AI skills, up from 57% the year before, across an analysis of close to a billion job ads. The people earning that premium are not the ones making their drafts undetectable — they are the ones who can use the tool well and know what to check before anyone else sees the result.

Coursium teaches that part directly — writing a clear prompt from what you actually have, and checking an AI answer before you act on it. Stay ahead of AI by learning the tools on your phone.