How to Use ChatGPT at Work Without Getting in Trouble

Half your team already uses ChatGPT to fix emails and summarize long threads. Nobody has said out loud whether that's allowed. You want to use ChatGPT at work too, because it saves real time, but you also don't want to be the person who pasted a client contract into a chatbot or sent your manager a report with a number the AI made up.

That worry is reasonable. The trouble people get into with AI at work almost always comes from two places: what they paste in and what they send out without checking. Fix those two and you can use ChatGPT, Gemini, Claude or Microsoft Copilot every day without drama.

This guide gives you the rules, a worked example you can copy, and a short script for asking your manager what's allowed.

Why the rules feel so unclear right now#

AI use at work has grown fast, and company rules haven't kept up everywhere.

A national US survey summarized by the Federal Reserve Bank of St. Louis found that "the share of workers using it for their jobs rose from 33% to 45%" between August 2024 and May 2026 (St. Louis Fed, What Work Does Generative AI Do?). The researchers call adoption "widespread but shallow": lots of people use it a little.

The New York Fed found something similar inside companies. In its 2026 survey of firms in the New York region, 61% of service firms used AI, but among those firms "the median share of workers using it was just 17 percent" (New York Fed, Liberty Street Economics).

So your company may officially "use AI" while most of your colleagues are still figuring it out alone. That's exactly the gap where mistakes happen. It's also your chance to be the person on the team who uses it well.

The two ways people get in trouble#

Before the rules, get clear on what you're protecting against.

1. Data going where it shouldn't. When you paste text into a public chatbot, it leaves your company's systems. Depending on the tool, your plan and your settings, the provider may store it, staff may review some conversations for safety, and it may be used to train future models. If that text had a customer's phone number, a salary or unreleased numbers in it, you've shared it with a third party.

2. Wrong output going out under your name. Chat tools write fluent, confident text. They don't check facts unless they're searching the web or reading a file you gave them. They can invent a statistic, a quote, a policy detail or a link that looks right and goes nowhere. If that lands in a client email, the client blames you, not the tool.

A Harvard Business School experiment with 758 consultants at Boston Consulting Group shows both sides of this. On tasks AI handles well, people using it finished "12.2% more tasks" and worked "25.1% more quickly". On a task picked to sit outside what the AI could do, people using AI were "19% less likely to produce correct solutions" than people working without it (HBS Working Paper 24-013).

Read that twice. AI made smart people faster on the right tasks and worse on the wrong ones. Your job is knowing which is which, and checking before you trust.

How to use ChatGPT at work: 6 rules that keep you safe#

Rule 1: Find your company's AI policy first

Your employer's rules beat anything a tool's settings page says. A decent policy usually has three parts:

• An approved tools list (for example, "Copilot with your work login, nothing else")
• Data rules: what kinds of information can go into AI and what can't
• A named person to ask: your manager, IT or the security team

If you can't find it, ask. Here's a message you can copy:

Hi [name], I'd like to use [tool] for [task, e.g. drafting first versions of routine emails].
Could you point me to our AI policy, or tell me which tools and types of data are approved?
Thanks, [your name]

Nobody gets in trouble for asking this. People get in trouble for not asking.

Shadow AI is the name for using a tool your company hasn't approved, with work data. It can break policy and client contracts even when nothing leaks. If a free app writes better summaries than the approved one, ask IT to review it before it touches client data.

Rule 2: Know the five things that never go in

If a detail fits one row, take it out before you paste. Don't assume one small detail will go unnoticed. And if your policy is silent on a type of data, treat it as data to keep out.

Rule 3: Redact with placeholders, keep the shape

You don't need the real details for AI to help. It needs the shape of the problem: the structure, the tone, the ask.

Swap real details for placeholders:

The AI can fix the tone of an email just as well with [AMOUNT] as with $18,500. If a placeholder breaks the flow (say, a salary negotiation email), use made-up numbers with the same shape and tell the AI they're samples.

Watch out for half-redaction. "Ms. F. at O. Retail, about $18.5k" still identifies the client and the deal.

Rule 4: Check the settings in every tool you use

Most chat tools have the same few privacy controls. Learn where they live in yours.

On many free and personal plans, your chats can be used for training unless you switch it off. Business and enterprise plans from the big providers generally don't train on your data by default and give IT controls over retention. But terms change often, so open the policy page for the exact tool and plan you use and read what it says today.

Memory deserves a second look. It's handy when the tool remembers "I work in customer support for a software company". It's a problem when it quietly stores a customer complaint or a salary figure and brings it up three weeks later in an unrelated chat.

Rule 5: Check the output before it leaves your desk

This is the rule most people skip. Treat ChatGPT like a fast, well-read intern who never says "I don't know" unless you ask. Great for drafts. Never the final word.

Before anything AI touched goes out, check:

1. Facts and numbers against a real source (the report, the system, the policy page)
2. Names, dates and amounts against the original record
3. Links: open every one before you share it
4. Promises: delete any commitment, deadline or refund you didn't actually agree to
5. Tone: read it as the person receiving it

Two prompt habits make checking easier. Ask it to flag what it's unsure about:

Answer the question above. At the end, list anything you're not sure about
or couldn't check. If you don't know, say so.

And regenerate when a fact matters. If the second answer contradicts the first on a number or a date, that fact needs checking in a real source.

Rule 6: Own the result, and say you used AI when it matters

"The AI said so" is never an excuse. Your name is on the email, the slide or the form.

Say you used AI in three situations: when company, client or school rules require it, when the work is presented as your own analysis or creative work, and when someone asks you directly. One plain sentence is enough:

I drafted this with an AI assistant and checked the figures against [source report].

Fixing the spelling in a chat reply usually doesn't need a note. A market summary you're presenting as your analysis does.

Worked example: fix the tone of a client email safely#

Say you're an account coordinator. You've written this and it sounds stiff:

Hi Ms. Fernandes,

Following our call, the renewal for Orbit Retail is confirmed at $18,500.
Please call me on +44 7700 900123 if anything looks wrong. The invoice
will reference Project HALO.

Regards,
Daniel

Step 1: Redact. Four details must stay inside your company: the client name, the amount, the phone number and the project code.

Step 2: Paste the redacted version with a tight prompt:

Rewrite this email in a warmer, friendly tone. Keep the facts, the length
and the placeholders exactly as they are. Don't add new dates, new
commitments or new offers.

Hi [CLIENT],

Following our call, the renewal is confirmed at [AMOUNT]. Please call me
on [PHONE] if anything looks wrong. The invoice will reference [PROJECT].

Regards,
[YOUR NAME]

Step 3: Check the output before you put the real details back. Here's what to look for:

• Every placeholder survived. If [AMOUNT] turned into a number, the AI invented it.
• No new promises. Phrases like "we'll also include a free month" or "I'll call you tomorrow" are common additions. Cut them.
• Same facts, same order. The renewal, the call-back offer and the invoice reference should all still be there.
• Length is close. A three-line email shouldn't come back as eight lines of "I hope this message finds you well".
• It sounds like you. If it doesn't, paste an email you wrote last month and ask it to match that style.

Step 4: Put the real details back yourself, in your own email client, and send. The chatbot never saw the client's name, the price or the phone number.

That whole loop takes about two minutes once you've done it a few times. Save the prompt so you don't rewrite it every time.

Usually safe vs needs care#

What "getting in trouble" usually looks like#

It's rarely dramatic. It looks like ordinary work that went one step too far:

• Pasting a whole email thread to "get context", including a colleague's private comments
• Using a free app because it's faster, when your company approved a different one
• Copying a confident statistic into a client deck without finding the source
• Letting a draft reply promise a refund or a date nobody approved
• Using AI during an online assessment or interview that said "no external tools"

Every one of these is avoidable with the six rules above. None needs you to stop using AI.

Write your three personal rules#

Long policies get ignored. Three short rules you actually remember get followed. Here's a set you can adapt:

1. I only paste work data into tools on my company's approved list.
2. I check the data and training setting before my first real task in any new tool.
3. I check every number, name and promise before anything AI touched leaves my desk.

Write yours, keep them short, and pin them somewhere you'll see them.

Job hunting? Use the same rules in your search#

Interviewers now ask "How do you use AI in your work?" all the time. The six rules above are your answer. A strong answer names a real task, a real habit and your own quality check:

I use AI to draft first versions of routine emails and to summarize long
documents. I treat it as a first draft only. I never paste client or personal
data into public tools, and I check every number against the source before
anything goes out.

Practice saying it out loud. If it sounds stiff, the free Answer Polish tool helps you tighten it, and the Mock Interview tool lets you rehearse the follow-up questions.

Reading job posts that ask for "AI literacy" or "Copilot experience" and not sure what they mean? Paste the post into the free JD Decoder to see what the employer is really asking for. Then run your resume through the free ATS checker to see whether those keywords show up. When you're ready to apply, browse remote jobs on JobRise.

FAQ#

Can I get fired for using ChatGPT at work? It depends on your employer's policy and your contract, so check those first. The common problems are using an unapproved tool with work data, pasting confidential or personal information, and sending AI output that turned out to be wrong. Follow your policy, redact, and check output, and you've covered the usual risks.

Is it okay to use ChatGPT for work emails? Usually yes, if your company allows the tool and you redact names, amounts, contact details and anything confidential first. Read the draft before sending and remove anything you didn't actually agree to.

Does ChatGPT keep what I paste? Assume the provider stores it for some time. Whether it's used for training depends on your plan and settings. Check the tool's data settings and its current policy page, and use a temporary chat for anything sensitive.

Should I tell my manager I used AI? Yes when your company requires it, when the work is presented as your own analysis, or when someone asks. One sentence covers it: what AI did and what you checked.

Is Microsoft Copilot safer than ChatGPT for work? Not automatically. What matters is whether you're signed in with a work account your company manages, and what your company's settings and policy say. The same tool can treat your data very differently on a personal login.

We're building short, hands-on courses on using AI at work. Get an email when they open.