Colour theme
Book a consultation

Setting ground rules for ChatGPT at work

Pick the account, set what staff can paste in, check the output before it's sent and measure whether ChatGPT saves time on one real task.

A 3D rendering of a glowing chip labelled GPT-4 on a circuit board.

ChatGPT can help staff draft, organize or explain material. Before it becomes part of ordinary work, decide which account they’ll use, what information can go into it and who checks the result.

Start with one task that ends in something a person can check, such as a draft customer reply that the employee handling the request reviews before it’s sent.

Sign in to the company workspace

Have staff sign in to the company’s ChatGPT workspace. A personal account has its own history and training settings, and the business can’t manage them.

OpenAI’s workspace security guidance explains that covered Business, Enterprise and Edu workspace data isn’t used for model training by default. It also describes boundaries for local access and connected services.

Training is only one question. Also check how long chats are kept, which connected apps the workspace allows and whether your privacy notice covers using customer information this way.

Give the task an information boundary

List what staff may paste in and what must stay out. Practise with a made-up or redacted example until you’ve decided which real data the task can use.

Don’t enter passwords, support credentials or unnecessary customer records. Check uploaded files as well as the words typed into the prompt.

ChatGPT has no setting that stops collecting data once a conversation reaches some limit. Decide what goes in before anyone types it.

Explain the expected deliverable

Tell the tool what the draft is for, who will read it and which facts it should use. Give it the format you want and ask it to flag anything those facts don’t support.

For a customer reply, paste in the relevant policy or product details and leave out account numbers and other identifiers it doesn’t need. Say which promises, such as refunds or delivery dates, the draft may make.

Keep the source document handy so the reviewer can check against it.

Review before using the output

Check names, figures, dates and factual claims against your own records. Open the links and check quotations against their sources.

Look for invented promises, the wrong tone and missing context. A polished response can still be wrong. Legal, health and employment material needs review by someone qualified in that area.

Watch for biased or stereotyped wording as well. The model learned from a huge amount of public text and can repeat its assumptions. Take extra care when a draft describes people, or feeds a decision about a customer, applicant or employee.

The employee who owns the communication should approve it before sending. Don’t forward output just because it sounds confident.

Tell customers where you use it

If AI helps draft replies to customers, say so in your privacy notice or customer service policy. One line is enough, for example: “We use AI tools to help draft some replies, and a member of our team checks each one before it’s sent.” Keep the statement in line with what staff do, and update it if the workflow changes.

Check each connected app separately

Uploading a document, connecting an app such as Outlook or SharePoint, and letting the tool send or edit something carry different risks. Check what each connection asks permission to do and whether the task needs it.

Files and web pages can contain instructions aimed at the AI (known as prompt injection). Treat what the tool reads as information to check. It shouldn’t change the task you gave it.

Putting ChatGPT in front of customers on your website or support channels is a separate project. It needs its own access design, testing and a way to hand the conversation to a person.

Measure the finished work

Compare the time and quality of the reviewed result with the previous workflow. Include editing, fact-checking and correction effort.

Give staff a simple way to report wrong output or information shared by mistake. Keep the chats and drafts you need under your normal records retention rules.

Use the staff AI rules guide to keep this workflow consistent with wider business policy.


Featured Image Credit