30 Days to 5 Hours Back: A Simple AI Starter Plan

The goal isn't to "learn AI." It's to get real hours back in your week. This plan does that by fixing one task properly instead of automating everything at once (which is why most people quit). One task, done well, then the next.

Takes about 30 minutes a week for four weeks. You need: any AI chat. You end with: one task off your plate permanently, and a method you can repeat every month after.

Week 1: Find the task worth fixing

List every task you repeat weekly at work or home.
Circle the one you dread most or that eats the most time. (For a lot of people it's follow-ups, scheduling, or turning notes into to-dos.)
Do that task manually one more time, and write down every single step as you go. This step list is gold.

How to pick when everything feels equally annoying. Score each candidate out of three: does it happen at least weekly, does it follow the same shape every time, and is a rough first draft genuinely useful? A three is your task. A one is a trap — most people's first instinct is the worst task, which is usually the one that's irregular and high stakes, and it's the reason they conclude AI doesn't work for their business.

Skip for now: anything legal, medical, or where money leaves the account. Come back to those in month three when you can tell a good output from a confident wrong one.

End of week 1, you should have: a step list of six to twelve lines. If it's longer, you picked a process, not a task — take one piece of it.

Week 2: Hand it to AI

Open ChatGPT or Claude and set your custom instructions (see the Starter Guide).
Take your step list and ask: "Here's a task I do every week, step by step. Which parts can you do for me, and how should I set that up?"
Get it working for one real instance. Don't polish, just prove it works once.

Better question to ask than the obvious one:

Here's my step list. Before you tell me what you can do, tell me which of these steps aren't really tasks at all — the ones that are me hunting for something, remembering something, or fixing something that shouldn't have broken. Then handle your part on this real example: [paste one, names changed].

Half of most step lists is finding and remembering, not doing. Those don't need AI, they need a fix you make once.

End of week 2, you should have: one real instance done, in the chat, that you'd have been willing to send.

Week 3: Make it repeatable

Save the prompt or setup so you're not rebuilding it each time.
Run it live for the week. Note where it breaks or needs a tweak, and fix those.
If it involves reminders or other people, connect it to your calendar, email, or team chat so it runs without you babysitting it.

This is the week people quietly lose. Week 2 feels like success, so week 3 feels optional. It isn't — a thing that works once and isn't saved is a thing you'll rebuild from memory in a month, slightly worse.

Save the prompt and one good example next to it. The example is what tells you what changed when it stops working.

End of week 3, you should have: run it three or four times, and the last run took under five minutes.

Week 4: Bank the time, then expand

Add up the time you got back. Even 30 minutes a task adds up fast.
Only now, pick your next task and repeat Week 1.
Write one sentence: "AI now handles ___ for me." That's your proof it's working.

Be honest in the maths. Count the time the task takes now, including your review, against what it took before. If it's 40 minutes down from an hour, that's a win — write down 20 minutes, not "an hour saved." Honest small numbers compound. Inflated ones fall apart the first time someone asks.

And spend the time on purpose. Time you win back and don't claim gets absorbed by other work within a fortnight. Decide what the 20 minutes is for before you start week 5.

What a finished month looks like

Task:     weekly client follow-ups
Before:   ~2.5 hrs/wk, always late, sometimes missed
Now:      ~35 min/wk - AI drafts from my notes, I edit
Saved:    ~2 hrs/wk, and none missed in 3 weeks
Saved as: "Follow-ups" project + one good example
Next:     the four repeat questions in my DMs

Six lines. That's a successful month, and it's more than most people get from a course.

When it goes sideways

"I missed a week." Then it's a five-week plan. This is the most common thing that happens and the least important. Pick up where you stopped — nothing here expires.

"Week 2 worked and week 3 fell apart." Almost always because the task was slightly different each time and the saved prompt only fit the first version. Add to the prompt: "If anything I've given you is missing or looks different from the example, ask me before you draft."

"I got to week 4 and couldn't measure anything." You didn't time the before. Don't relitigate it — just time this month's task properly in week 1 of the next round, once, with a phone timer.

"I finished and I'm not sure it was worth it." Ask yourself whether the task is off your mind, not just off your clock. The mental weight of "I still owe six people a follow-up" costs more than the 40 minutes did.

Make it yours

Client service: follow-ups in month one, proposals in month two. Both are shaped the same every time and both are visible to the person paying you.
Studio, shop, scheduled business: the weekly schedule in month one. It's the thing that eats a whole evening and follows identical rules each week.
Solo: whatever you do on Sunday night. Take that one back first, for reasons that have nothing to do with productivity.
Small team: month one, the thing your team interrupts you about most. You get your time back and so do they.

Do this next

Open a note right now and list five tasks you repeat every week. That's week 1, day 1, and it takes four minutes.

The whole secret: one task at a time. People who try to automate their entire life in a weekend burn out and quit. People who fix one annoying thing a month are unrecognizable in six.

Next guide: The Time Audit if you can't decide which task to pick — one prompt that ranks them for you, including the option of reading it straight off your calendar.