itsbrill.aiAll chapters

Appendix

Worked examples by the job you're actually doing

These are made-up, real-shaped examples. They're not anyone's private inbox or client data, and they're not proof claims. Each one shows the same pattern: messy input, the prompt, the shape of the output, the human edit, time saved, and one caveat. Time saved is a rough estimate for the example, not a promise.

Here's the thing we learned from a panel of 45 readers: almost everyone's asking for the same handful of jobs. Email and inbox. Scheduling. Meetings. Documents and summaries. Chasing people. Customer questions. Your industry barely changes the job, it just changes the words. So we've built this around six universal jobs first, then bolted on three vertical extras at the end for the bits the core set misses.

Each job has one worked example plus a couple of quick swaps so you can see your own version without reading the whole thing again. Find the job you keep doing by hand, copy the prompt, swap in your words.

One note on tools before you start. Where you need to actually open something, we name a recommended pick and a couple of alternatives. ChatGPT, Claude and Gemini all do this stuff well. If you're not sure, open the free version of ChatGPT and start there. Don't overthink it.

1. Inbox and email

This was the runaway winner: 24 of 45 readers asked for it first. Same job whether it's your work inbox or the family group chat.

Messy input: Twelve unread emails. One supplier wants a decision today, one customer's annoyed about a delay, three are newsletters, two need a quick yes, one's a meeting request, and the rest can wait.

The prompt:

Here are my unread emails (pasted below).
Sort them into: reply today, reply this week, and ignore.
For the "reply today" ones, draft a short, friendly reply I can edit.
Don't promise anything I haven't told you. Flag anything I should double-check.

Output shape: Three buckets. The "today" pile comes with short draft replies. The "ignore" pile is named so you can trust it's safe to skip.

Human edit: You fix the supplier reply because the AI guessed at a date you hadn't given it, and you delete one draft that sounded too formal for a regular.

Time saved: 20 to 30 minutes of inbox dread, most mornings.

One caveat: It drafts, you send. Never let it fire off replies on its own, especially where money or promises are involved.

Quick swaps:

Which tool: ChatGPT's free version is the easiest start. Claude and Gemini work just as well. If your email lives in Gmail or Outlook, their built-in AI can read the inbox directly, but copy-paste works everywhere and leaks nothing extra.

2. Scheduling and calendar

16 of 45 wanted help here. Turning a mess of "can you do Tuesday?" into an actual plan.

Messy input: Four people trying to find a time. One can only do mornings, one's away Thursday, one's in a different timezone, and you've got a dentist appointment blocking Wednesday at 2.

The prompt:

Here's everyone's availability (below) and my own constraints.
Find the two or three times that work for the most people.
Tell me who can't make each option and why.
Then draft a short message I can send to lock it in.

Output shape: A short ranked list of options, each one showing who's in and who's out, plus a ready-to-send "let's go with Tuesday 10am" message.

Human edit: You bump the top option because you know one person hates early starts, even though they technically said yes.

Time saved: 15 minutes and three rounds of back-and-forth.

One caveat: It can't see your real calendar unless you connect it. Double-check it isn't booking over something it doesn't know about.

Quick swaps:

Which tool: Any of the big three will do the thinking. If you want it to actually read and write to your calendar, Google Gemini ties into Google Calendar neatly and Microsoft Copilot into Outlook. Start with copy-paste, connect it later if you trust it.

3. Meetings

Turning notes, a recording or a wall of chat into something you can act on.

Messy input: A 40-minute meeting. You've got rough notes that trail off halfway, and you can't remember who agreed to do what.

The prompt:

Here are my rough meeting notes (below).
Give me: a short summary, the decisions we made, and a list of who-owns-what with any deadlines.
If something's unclear or an owner is missing, flag it rather than guessing.

Output shape: A tight summary up top, then decisions, then an action list with names and dates. Anything fuzzy is flagged, not invented.

Human edit: You add two actions the notes missed entirely because you remember them, and correct one owner the AI guessed wrong.

Time saved: 20 minutes of writing it up, and fewer "wait, who was doing that?" moments later.

One caveat: It only knows what's in the notes. If your notes are thin, the summary will be thin too. It can't read minds, just text.

Quick swaps:

Which tool: For typed or pasted notes, ChatGPT or Claude. For live meetings, a dedicated notetaker like Otter, Fireflies or your video app's own AI summary will record and write it up for you. Always tell people in the room they're being recorded.

4. Documents and summaries

6 of 45 asked for this directly. Long thing in, short useful thing out.

Messy input: A 14-page PDF. A contract, a council letter, a report, a school policy, whatever. You need the gist and you need to know if anything's going to bite you.

The prompt:

Read this document (attached/pasted).
Give me: a plain-English summary in five lines, the three things that matter most to me, and anything that looks like a catch, a cost or a deadline.
Quote the exact line for anything important so I can find it.

Output shape: A five-line plain summary, your top three, and a "watch out for this" list with the real lines quoted so you can check them yourself.

Human edit: You read the two quoted lines it flagged, decide one's fine, and circle the other to ask someone about.

Time saved: 30 to 45 minutes of reading you were dreading.

One caveat: It's a reading assistant, not a lawyer or an accountant. For anything binding, use it to find the scary bits faster, then get a real person to confirm.

Quick swaps:

Which tool: Claude is especially good with long documents and quoting accurately. ChatGPT and Gemini both handle PDFs too. Upload the file rather than retyping it, and never upload anything confidential to a tool your workplace hasn't approved.

5. Follow-ups and chasing

6 of 45 wanted help with the awkward jobs: nudging people without sounding rude or letting things slip.

Messy input: A list in your head. Three quotes you sent that never got a reply, an invoice that's two weeks overdue, and a contact who said "let's catch up" a month ago.

The prompt:

Here's a list of people I need to follow up and why (below).
For each one, draft a short, polite chase message in a friendly tone.
Make the overdue invoice firmer but still warm.
Keep them all short. Give me a "first nudge" and a "second nudge" version.

Output shape: One short message per person, in the right tone, with a softer first version and a firmer second one ready for when they ghost you again.

Human edit: You soften the invoice chase because you know that customer's a good one going through a rough patch.

Time saved: 20 minutes, and the deeper win of actually sending the awkward ones instead of putting them off.

One caveat: Tone is yours to own. Read each one in your own voice before it goes, a chase that lands wrong costs more than the time it saved.

Quick swaps:

Which tool: Any of ChatGPT, Claude or Gemini. Keep a saved note of your favourite chase wording so you're not rewriting the prompt every time.

6. Customer and support requests

11 of 45 named this. Answering the same kinds of questions, faster and kinder.

Messy input: A customer message that's half a complaint, half a question, and a bit cranky. "I ordered last Tuesday, still nothing, and your website said two days, is it even coming or should I just get a refund??"

The prompt:

Here's a customer message (below).
Draft a calm, friendly reply that acknowledges the frustration, answers the actual question, and offers a clear next step.
Don't promise a delivery date or a refund I haven't approved. Leave a [gap] where I need to add the real status.

Output shape: A warm reply that names the frustration, answers plainly, and leaves a clear [gap] for the real facts so you don't accidentally promise something untrue.

Human edit: You fill the [gap] with the actual tracking status and decide yourself whether to offer the refund.

Time saved: 5 to 10 minutes per tricky message, and a lot less emotional load.

One caveat: Never let it promise refunds, dates or compensation on its own. It writes the tone, you own the facts and the decisions.

Quick swaps:

Which tool: ChatGPT or Claude for one-off replies. If you're handling lots of support, tools like Intercom or Zendesk have AI built in that can read your help docs, but start with copy-paste and a saved prompt before paying for anything.

Vertical add-ons

The six jobs above cover most of what most people need. These three are for the bits a particular line of work runs into that the universal set doesn't. Same compact shape.

A. Supplier or stock order

8 of 45 (mostly shops, cafes, trades) wanted this.

Messy input: A half-empty stockroom and a memory. Oat milk's nearly out, you're down to two boxes of the medium gloves, the good coffee's getting low, and you think you owe the produce supplier a reorder.

The prompt:

Here's what's running low (below) and who I usually order each from.
Group it by supplier, draft a short order message for each one, and flag anything where I should check the quantity before sending.
Don't guess prices or order numbers.

Output shape: One short order message per supplier, grouped so you're not sending five separate scrappy texts, with a flag on anything you should eyeball first.

Human edit: You count the gloves properly before sending because the AI only had "down to two boxes" to go on.

Time saved: 15 minutes and one fewer "we're out of oat milk" morning.

One caveat: It can't see your real stock. Confirm quantities yourself before anything gets ordered.

Which tool: ChatGPT or Claude. If you keep stock in a spreadsheet, paste the low-stock rows straight in.

B. Working across languages

6 of 45 work with customers, staff or suppliers in another language.

Messy input: A supplier's reply in Mandarin, and you need to answer in clear English-then-Mandarin without sounding like a robot or causing offence.

The prompt:

Translate this message (below) into English so I understand it.
Then help me reply: I want to say [your point].
Give me the reply in both English and Mandarin, in a polite, professional tone.
Flag anything that might not translate cleanly or could be taken the wrong way.

Output shape: A plain-English translation, then your reply in both languages, with a flag on any phrase that might land oddly.

Human edit: You ask a bilingual mate to glance at the final version before it goes to an important supplier.

Time saved: 20 to 30 minutes and a lot of guesswork.

One caveat: It's good, not perfect. For anything legal, contractual or relationship-critical, get a human who speaks the language to check it.

Which tool: ChatGPT, Claude and Gemini all translate well and handle the back-and-forth. For quick one-way translation only, Google Translate or DeepL are faster.

C. Building a roster

3 of 45 (cafes, shops, anyone scheduling staff) wanted this.

Messy input: Five staff, their availability scribbled on different days, two who can't work weekends, one who's only learner-level on the coffee machine, and a Saturday that's always slammed.

The prompt:

Here's my staff list, their availability and their skills (below), plus the shifts I need covered.
Build me a draft roster for the week.
Make sure every shift has someone who can do the key tasks, respect the availability, and flag any shift I can't safely fill.

Output shape: A draft week laid out by day and shift, skills matched, with a clear flag on any gap you'll need to solve yourself.

Human edit: You move one person off the Saturday open because you know they've got an exam, something the availability note didn't capture.

Time saved: 30 to 45 minutes of the job nobody enjoys.

One caveat: It doesn't know your award rates, break rules or the human stuff. Treat it as a first draft, not a final roster, and check it against your obligations.

Which tool: ChatGPT or Claude for the draft. If rostering's a weekly grind, a proper tool like Deputy or When I Work handles pay rules and shift swaps, but the AI draft is a fine place to start.

More worked examples

Improve something you already wrote

You don't need a blank page. You need an editor. Paste what you've got and let it tighten the thing.

Messy input: An email you've rewritten four times and still hate. It's too long, a bit ranty in the middle, and you can't tell if it reads as firm or just rude.

The prompt:

Here's something I wrote (below).
Tighten it, fix the flow, and make the tone [firm but friendly].
Keep my voice. Don't make it sound corporate or stiff.
Give me the edit, then one line on what you changed and why.

Output shape: A shorter, cleaner version that still sounds like you, plus a quick note on what got cut so you can tell whether it went too far.

Human edit: You put one phrase back because it's how you actually talk, and you trim the sign-off it made too formal.

Time saved: 15 minutes of staring and rewriting, and you finally hit send.

One caveat: "Keep my voice" only works if it has your voice to start with. Give it the clunky draft, not a one-line brief, or it'll hand you back generic mush.

Which tool: Any of ChatGPT, Claude or Gemini. Paste a couple of your old emails first and tell it "match how I write" if you want it to nail your tone.

Make sense of a spreadsheet or write a formula

You don't need to be a spreadsheet wizard. You need to describe what you want. Say it in plain words and it'll hand you the formula.

Messy input: A few columns of sales. You want the total for one month, or the average per customer, and you've got no idea where to even start with the formula.

The prompt:

Here are a few rows from my spreadsheet (pasted below).
I want to [add up sales for June only].
Give me the exact formula for [Excel / Google Sheets], tell me which cell to put it in,
and explain in plain English what it's doing so I can change it myself later.

Output shape: The ready-to-paste formula, where it goes, and a plain explanation of each bit so next time you can tweak it without asking.

Human edit: You swap the cell range to match your real sheet, because the example rows you pasted weren't the whole thing.

Time saved: 20 minutes of googling formulas and getting #REF! errors.

One caveat: Pasting a few rows is plenty for a formula. But if you want it to actually crunch a whole file, that's the paid data tier on most tools, and you'll need to upload the file rather than paste it. Tell it your version (Excel and Sheets aren't identical) or the formula won't fit.

Which tool: ChatGPT or Claude are both strong on formulas. For working over an entire uploaded file, you'll want a paid plan that allows file uploads and analysis.

Tailor a CV or cover letter to a job ad

Don't send the same CV to twenty jobs. Aim each one at the ad. Paste both and let it line them up.

Messy input: Your work history in one block, copied from your old CV, and a job ad you're keen on. The ad wants things your CV mentions but buries.

The prompt:

Here's my work history (below) and the job ad I'm applying for (below that).
Rewrite my CV summary and the key bullets to match what this ad is asking for.
Use only what's in my history. Do not invent jobs, dates, skills or qualifications.
Then draft a short cover letter in a warm, confident tone.

Output shape: A tailored summary and bullets that pull your real, relevant experience to the front, plus a short cover letter aimed at that specific role.

Human edit: You check every line against the truth, fix one bullet that overstated what you actually did, and add a sentence about why you want this job.

Time saved: 30 to 45 minutes per application, and a far better hit rate than spray-and-pray.

One caveat: This is the big one. Never let it invent qualifications, dates, job titles or credentials, and it will if you let it. A made-up line on a CV gets found in the interview or the reference check, and that's worse than a gap. Read every word and make sure it's true before it goes anywhere.

Which tool: ChatGPT or Claude for the draft. Keep your full real history in a saved note so you can re-aim it at the next ad in two minutes flat.