How can ChatGPT improve your workflow?

Most office work is not one big task but dozens of small ones: answering the same kinds of emails, turning notes into summaries, reformatting the same report every Friday. These are exactly where a chat assistant helps, because it is fast at producing a first version of text and you are fast at judging whether that version is right.

There is early evidence that the gains are real for this kind of work. In an experiment by MIT researchers Shakked Noy and Whitney Zhang, published as a working paper in March, 444 professionals did realistic writing tasks such as press releases, short reports and delicate emails. Those allowed to use ChatGPT finished in about 37% less time, and graders rated their work higher on average. The study is not yet peer reviewed and used short, one-off tasks, so treat the figures as a sign of what is possible rather than a promise.

Why your workflow, not just your prompts

Microsoft's 2023 Work Trend Index, a survey of 31,000 workers, found that 68% say they lack enough uninterrupted focus time and 62% spend too much of the day searching for information. Its analysis of Microsoft 365 usage puts meetings, email and chat at 57% of a typical employee's time in those apps. A tool that trims the small repeated pieces of that load gives back time for work that needs thought.

But using ChatGPT at random, a question here and there, rarely changes much. The bigger gains come from deciding in advance which tasks it handles, writing a reusable prompt for each, and building a quick check into the routine.

A worked audit of one week

Here is an audit for a fictional marketing coordinator. Each task is scored from 1 to 5 for repetitiveness (how often it recurs in a similar form) and risk (how much harm a mistake would do). The tasks, times and scores are illustrative.

The top three rows total 8.5 hours of high-repetition, low-risk work. Even if ChatGPT cut that time by a third, the coordinator would win back close to three hours a week, while the risky items stay under close control.

Three starter prompts

These follow a simple structure: give ChatGPT a role, the context it needs, the task and the output format, then refine the result in follow-up messages.

1. Routine replies

You are a friendly customer-service writer for a small outdoor-gear shop. Our policies: free returns within 30 days with receipt; shipping takes 3 to 5 business days; we do not ship outside Canada.
Draft a reply to the customer email below. Keep it under 120 words, warm and clear. If the email asks something our policies don't cover, write [CHECK] instead of guessing.
Email: """[paste email]"""

2. Repurposing content

You are a social media editor. From the blog post below, write three LinkedIn posts of 60 to 90 words each, each built around a different key point, with no hashtags and no exclamation marks. Use only facts stated in the post.
Post: """[paste blog post]"""

3. Meeting summaries

Summarize these meeting notes for team members who missed the meeting. Use three headings: Decisions, Action items (with owner and due date), Open questions. If an owner or date is missing, write "not stated". Do not add anything that is not in the notes.
Notes: """[paste notes]"""

Each prompt includes a guard against invented content: a [CHECK] marker, "use only facts stated" or "not stated." ChatGPT can make errors and invent details, so these lines make gaps visible instead of quietly filling them. The shop policies in the first prompt are examples; replace them with your own.

Which repetitive tasks should you hand over first?

Start where three conditions meet: the task recurs often, a mistake is cheap, and you can check the output in less time than it would take to write it. Routine replies, meeting summaries, reformatting, first drafts of internal updates and turning long text into short versions usually qualify.

Keep for later, or for yourself, tasks where errors are costly or hard to spot: anything with figures that go to clients or regulators, legal or medical wording, personnel decisions, and messages that carry your organization's name in a sensitive situation. ChatGPT can still help there with outlines or brainstorming, but the final text needs careful human review.

Also mind what you paste. Don't put confidential client or employee data into ChatGPT unless your employer has approved it, and remove names when a task does not need them. Then keep a simple log of each handed-over task, its prompt and what you had to fix. After a few weeks, the log shows which tasks are safe to rely on and which prompts need work.