# How do you save time with an AI assistant? Date: 2023-07-13 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/save-time-ai-assistant/ **Answer:** Give an AI assistant like ChatGPT the first draft of routine work, such as sorting email, turning rough notes into action items and sketching a daily plan, and keep the reviewing, deciding and sending for yourself. The time savings are real for writing tasks, but they come from drafting faster, not from skipping the check. The clearest evidence so far comes from a study by two MIT economists, Shakked Noy and Whitney Zhang, released as a working paper in March. They gave 444 college-educated professionals realistic writing tasks from their own jobs, such as press releases, short reports and delicate emails. Half were shown how to use ChatGPT. That group finished about 10 minutes faster than the comparison group, who averaged 27 minutes, a drop of 37 percent, and blind graders from the same professions rated their work higher on average. The study also shows where the time went. Before ChatGPT, people spent roughly half their time on a rough draft; with it, drafting time fell by more than half and editing time more than doubled. That is the pattern to aim for in your own day: let the assistant produce the first version, and spend your attention on reviewing it. ## A worked daily routine Here is a simple routine built around three common time sinks. Before you start, read the privacy section below; it changes what you paste. **1. Morning email triage.** Copy the subject line and first lines of your unread messages, with names and private details replaced by placeholders, and ask for a sort. ```prompt Here are 12 emails, numbered, with names replaced by [Client A], [Colleague B] and so on. Sort them into three groups: reply today, reply this week, read only. For each "reply today" email, write a one-line summary of what the sender needs and a three-sentence draft reply in a friendly, professional tone. Don't invent facts; put [CHECK] wherever a date, price or promise is needed. ``` The `[CHECK]` instruction matters. ChatGPT sometimes makes up details, which OpenAI calls hallucinating, and a placeholder is far easier to spot than a plausible wrong date. **2. Meeting notes into actions, right after each meeting.** ```prompt Turn these rough meeting notes into: (1) a five-line summary, (2) a table of action items with owner and due date, (3) open questions. If an owner or date is missing from my notes, write "not decided" instead of guessing. Notes: [paste your notes] ``` **3. End-of-day plan for tomorrow.** ```prompt Here is my task list for tomorrow with rough time estimates, and my fixed meetings. I have about 6 working hours. Suggest an order that puts the hardest task in the morning, groups small tasks together, and leaves 30 minutes free. Tell me which tasks don't fit and should move. ``` Across a week, the gain comes from consistency more than from any single prompt. Save the three prompts in a note so you can paste them in seconds. > **Try it:** For one week, keep a simple log: the task, the minutes it took with ChatGPT's draft, and the number of changes you made before using the result. At the end of the week, drop any task where your corrections took longer than writing it yourself would have. ## Protecting what you paste Every routine above involves pasting work content into a chat. OpenAI's FAQ says conversations may be reviewed by its AI trainers and asks users not to share sensitive information. Since April, ChatGPT's settings let you turn off chat history; conversations started that way are not used to train OpenAI's models and are deleted after 30 days. That helps, but it does not make the chat a safe place for a client's health details, passwords or your company's confidential figures. Placeholders such as [Client A] keep most tasks useful while leaving the sensitive parts out. ## What should you never delegate? **Judgment.** Deciding which client gets priority, whether a proposal is good enough, or how to answer a colleague who is upset involves values and relationships the assistant cannot see. Ask it for options and drafts, then decide yourself. **Private and confidential data.** Anything you would not forward to a stranger should not go into the chat box in identifiable form: personal information about others, health or financial records, unreleased plans, login details. When in doubt, summarize in general terms or leave it out. **Final sign-off.** The Noy and Zhang study found that 68 percent of the people using ChatGPT submitted its first output without editing it. In a timed experiment that was rational; in real work it is how errors reach clients. OpenAI itself recommends checking whether answers are accurate. Every message sent in your name, every number in a report and every commitment in a plan needs a person to read it and approve it. **Your own thinking on what matters most.** If a piece of writing is how you work out what you believe, such as a strategy note or a difficult letter, write the first draft yourself and use the assistant to tighten it. Save the delegation for the routine, and spend the time you win on the work only you can do. ## Sources - Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence: https://economics.mit.edu/sites/default/files/inline-files/Noy_Zhang_1.pdf - New ways to manage your data in ChatGPT: https://openai.com/index/new-ways-to-manage-your-data-in-chatgpt/ - ChatGPT General FAQ: https://help.openai.com/en/articles/6783457-chatgpt-general-faq