AI for Everyday Work · 12-day planAll days
You're back from three days off and twelve emails are waiting. Use your own AI to sort them, find what's urgent and draft the easy replies, then check the result against the traps hidden in the inbox. Most quick AI summaries miss at least three of them.
By the end of today you can
- Sort a real-looking inbox with AI into needs-me-today, this-week, just-reply and no-action
- Write a triage brief that makes the AI use the newest version of every date and number
- Catch the mistakes a quick AI summary makes, using follow-up questions and quotes
- Draft replies where you decide what to say and the AI only shapes the words
- Recognise what you should never paste into an AI tool from a real inbox
Monday, 9:04am. You've just come back from three days off. You're the Operations Lead at Fernway, a company that runs corporate events, and twelve emails are waiting for you. Your biggest client has a conference in June, your boss has a call with Finance this afternoon, and somewhere in this inbox is a deadline that moved while you were away.
Today you'll use AI to get through it in a fraction of the usual time. You'll also see exactly where AI lets you down, because the inbox has traps planted in it: the kind of thing a quick AI summary misses, and the kind of thing that causes real trouble at work.
The one rule for this course
AI is a fast, tireless assistant who has never met you, doesn't know your company, and is confidently wrong now and then. So the rule for every day of this course is:
The AI drafts. You decide.
You'll need any AI chat tool open in another tab: ChatGPT, Claude, Gemini or Microsoft Copilot. The free versions are fine. If you've never used one, our AI Assistants course covers the basics in a day.
Your inbox
Here are the twelve emails, newest first, the way your inbox shows them. Fernway and everyone in it are made up, so it's safe to paste into any AI tool. Don't read it carefully yet: the point is to see what the AI does with it.
Work pack
Alex's inbox, Monday 9 March, 9:04am
Step 1: Try the lazy way first
Most people start like this. Do it on purpose, so you have something to compare with. Copy the work pack into your AI, then add:
The lazy prompt
Summarise these emails.
Keep the answer. You'll grade it in a minute. It probably looks tidy and confident, and that's exactly the problem: a tidy summary of the wrong dates is worse than no summary, because you trust it.
Step 2: Brief it like a new colleague
Imagine handing this inbox to a smart temp on their first morning. You wouldn't say "summarise these". You'd tell them who you are, what you need out of it, and what to watch for. Start a new chat (so the lazy answer doesn't influence it) and paste this, with the work pack underneath:
The triage brief
I'm Alex, Operations Lead at Fernway, an events company. I've been off for three days. Today is Monday 9 March, 9:04am. Below are my 12 emails, newest first. Sort them into four groups: 1. Needs me today: a deadline today or tomorrow, or someone is blocked until I act. 2. This week: needs action, but not today. Give the exact deadline. 3. Just reply: needs a short answer, no decision. 4. No action: FYI, newsletters, notifications. Rules: - For every deadline, give the exact date and time and quote the line it comes from. - If a later email changes a date, number or decision from an earlier one, say so and use the newest version. - Point out anything in one email that conflicts with another email. - Flag any email that looks suspicious (an unexpected link, an odd sender address, a request for a password or payment). - Flag any email with personal or sensitive information. - Don't write any replies yet. If something is unclear, ask me instead of guessing. [paste the emails here]
Why each part is there:
| Part of the brief | What it prevents |
|---|---|
| Who you are, and today's date | The AI can't tell what "tomorrow" or "end of Tuesday" means without it |
| Four groups, each defined | "Urgent" means something different to everyone. Definitions make the sort consistent |
| Quote the line for each deadline | You can check any date in two seconds, and invented dates have nothing to quote |
| Newest version wins, and say so | Old figures in quoted text are the most common way summaries go wrong |
| Conflicts between emails | Problems that only show when you put two emails side by side |
| Suspicious and sensitive emails | The AI won't think about safety unless you ask |
| No replies yet, ask if unclear | One job at a time. Guessing is how confident nonsense gets in |
Step 3: Check your result
Now grade both answers, the lazy one and the briefed one. Tick only what the answer actually said, not what you noticed yourself.
Check your result
Did the answer catch these?
0 of 8 caught
If your briefed answer missed a few, that's normal, and it's the most useful lesson of the day. The brief makes AI much better, but not perfect. That's why the next step exists.
Step 4: Question the answer
You wouldn't take a new colleague's first summary as final. Ask follow-up questions in the same chat. These three catch most of what slips through:
Follow-up 1: what changed?
Go through the emails again. List every place where one email changes or contradicts another: dates, numbers, decisions, people. For each, quote both lines and tell me which one is current.
Follow-up 2: what could go wrong?
Is there anything in these emails that could stop the Kestrel conference on 18 June from going ahead as planned? Think about the venue, the guests and the suppliers together, not one email at a time.
Follow-up 3: what did you assume?
Which parts of your summary are you least sure about, and what did you assume? Mark anything you said was "approved", "agreed" or "confirmed", and quote the words that show it.
The second one usually finds the lift problem. The third one is the habit that matters most: it makes the AI show its reasoning, so you can see where it filled gaps with guesses.
Spot the error
Here's a real-looking answer to the lazy prompt. It reads well. Four of its lines are wrong. Can you find them without going back to the inbox?
Spot the error
An AI summary of the inbox. Four lines are wrong.
Click every line you think is wrong, then check your answers.
Needs you today
This week
Can wait
0 selected
Notice what's not there at all: the lift. The most important problem in the inbox is missing, and nothing about the summary tells you so. Wrong lines can be checked; missing ones can only be found by asking the right follow-up question.
Step 5: Draft the replies
Now the AI can write. The trick is to decide what to say yourself, and let the AI handle how to say it.
For a simple, factual reply, a clear instruction is enough:
A simple reply
Draft a reply to Marta at Saffron Kitchen. Tell her the guest count for 18 June is now 135 (within her 140 maximum) and that I'll send dietary requirements by Friday 27 March. Under 70 words. Friendly and direct. No "I hope this email finds you well". Sign off as Alex.
For anything involving people's feelings, write your points first. Compare these two for Sam:
Weak: let the AI decide what to say
Reply to Sam's email.
Better: your points, the AI's wording
Here's what I want to say to Sam, in note form: - fair point, sorry he wasn't in the kickoff, it was arranged while I was off - he knows Kestrel well, that matters - Dana and I are talking about the Kestrel team on Wednesday, I'll come back to him after that - happy to chat before then if he wants Turn this into a short, warm reply from me. Keep my meaning exactly. Don't promise him a place on the account, and don't add anything I haven't said. Under 100 words.
The weak version often promises Sam a place on the account, or apologises for things that weren't your decision. The better version can't, because you set the content. Read every draft before sending: the AI drafts, you decide.
Before you try this on your real inbox
Your real inbox is not a made-up work pack. Before you paste anything:
- Check your company's rules. Many workplaces allow only approved tools (often Copilot in Outlook or Gemini in Gmail) for work email, because those keep the data inside the company's account.
- Leave out personal and sensitive details: health, salaries, performance issues, customers' personal data, passwords. For Jo's email, "a team member is back on 16 March, mornings only for two weeks, no set-up days" gives the AI everything it needs and nothing it shouldn't have.
- Never paste passwords, security codes or bank details, wherever they appear.
Adjust the brief to your job:
| If you're… | Add this to the brief |
|---|---|
| A manager | "Also list anything where someone on my team is waiting for a decision from me." |
| In sales or customer service | "Put anything from a customer above anything internal, and flag any customer who sounds unhappy." |
| In operations or admin | "List every supplier date and number in one table, so I can check them against our plans." |
| A freelancer or small business owner | "Flag anything about money: invoices, payments, quotes and price changes." |
Keep the template
Here's the brief as a reusable template. Save it to your playbook: by the end of the course you'll have one for every situation, ready to download.
For your playbook
Inbox triage
I'm [your role] at [company], which [what it does in a few words]. Today is [day, date and time]. I [have been away since… / am catching up on…]. Below are [number] emails, newest first. Sort them into four groups: 1. Needs me today: a deadline today or tomorrow, or someone is blocked until I act. 2. This week: needs action, but not today. Give the exact deadline. 3. Just reply: needs a short answer, no decision. 4. No action: FYI, newsletters, notifications. Rules: - For every deadline, give the exact date and time and quote the line it comes from. - If a later email changes a date, number or decision from an earlier one, say so and use the newest version. - Point out anything in one email that conflicts with another. - Flag anything suspicious (unexpected links, odd sender addresses, requests for passwords or payment). - Don't write replies yet. If something is unclear, ask me instead of guessing. Then answer: what changed between emails, and what could go wrong if I only read the latest ones? [paste emails, with personal and sensitive details removed]
Is it worth it?
Think about the last time you came back to a full inbox. How long did sorting it take you? Then try today's method on your next real batch of email, and log the difference, honestly, including the time spent checking.
Time log
Sorting a full inbox: is it worth it for you?
Be honest and count the checking. Some tasks aren't faster with AI, and that's worth knowing too.
Recap
- The AI drafts, you decide. That's the rule for the whole course.
- A brief beats a bare request: who you are, today's date, defined groups, and what to watch for.
- Make the AI quote the line for every deadline, and use the newest version of every date and number.
- Ask follow-ups: what changed, what could go wrong, what did you assume. The worst mistakes are things that are missing, not things that are wrong.
- For replies, you choose the points, the AI chooses the words.
- Real inboxes contain personal and sensitive data: follow your company's rules and leave it out.
Tomorrow: teaching the AI who you are, so its drafts sound like you instead of a chatbot.
Knowledge check
Test what you learned
5 questions · pick an answer to see if you're right
- 1
An email thread says "hold until Friday 13th", but the newest reply in the same thread says "only until Wednesday 11th noon". What usually goes wrong with a quick AI summary?
- 2
Your AI lists "Reset your password using the IT link before it expires" as an urgent task. The sender is it-support@fernway-helpdesk.com, and your company is fernway.com. What should you do?
- 3
Which of these is the best way to get a reply to a colleague who is upset?
- 4
An email from HR includes a colleague's medical details. Should you paste it into a free AI chat tool to help plan their schedule?
- 5
Your AI summary says "Dana has approved adding Sam to the Kestrel account". Dana's email actually says "have a think about whether we bring Sam on, let's chat Wednesday". What kind of mistake is this?
That's the last day of AI for Everyday Work!
Was this lesson helpful?
Need a breather? Take a two-minute break