Learn it on your own work, 20 to 30 minutes a day for 30 days: week one, daily use on real admin; week two, briefing it and checking its answers; week three, data, privacy and costs; week four, automating one task end to end. Add one free course, such as Anthropic's AI Fluency or OpenAI Academy, for structure.
At day 30 you'll have a handful of prompts that work for your business, a costed shortlist of the uses worth paying for, one working automation or a clear reason not to build one, and enough understanding to judge a vendor's pitch. You won't need to write code, and the whole month can cost between nothing and $20.
Before day one: one assistant, one list, one log
Spend 20 minutes setting up. It makes the next 30 days far more useful.
- Pick one assistant and stick with it. If your business runs on Microsoft 365, Copilot Chat is included in the business plans at no extra cost. If you use Google Workspace, Gemini is now built into the business plans. Otherwise, the free tiers of ChatGPT or Claude are enough to start. Don't pay for more than one during the month.
- Write your task list. Ten recurring jobs that eat your week: quotes, customer emails, chasing invoices, social posts, supplier orders, job sheets. This is your syllabus.
- Open a learning log. A simple document with three columns: what I asked, what came back, what I'd change. Two minutes at the end of each session.
- Check two settings. On a personal plan, switch off the model-training option in the privacy or data settings. And decide now that no customer details go in during the month unless they're anonymised; whether it's safe to put customer data into ChatGPT explains why.
Week 1 (days 1 to 7): daily use on real work
The only rule this week: every session uses a task you'd have done anyway. Practice on made-up examples teaches you very little about your own business.
| Day | Task | What you're learning |
|---|---|---|
| 1 | Paste an email you've already written and ask for a clearer, shorter version | How it handles tone, and when its version is worse than yours |
| 2 | Summarise a long email thread or document, then check it against the original | Summaries are fast and sometimes miss the one detail that matters |
| 3 | Draft a reply to a real enquiry from your rough notes | How much context it needs to sound like you |
| 4 | Ask for 20 ideas for a promotion or a month of posts, then pick three | It's a generous brainstorming partner; the choosing is still yours |
| 5 | Turn rough notes or a voice memo transcript into a checklist or job sheet | It's good at putting mess into structure |
| 6 | Ask it something about your trade that you already know well | How confidently it can be wrong |
| 7 | Read your log. Which two tasks saved real time? | Where your own quick wins are |
The log is where the week's learning actually sits. For an illustrative dog-grooming salon, the first three entries might read:
| Day | What I asked | What came back | What I'd change |
|---|---|---|---|
| 1 | Shorten my reply to a customer asking why de-matting costs extra | Half the length, but it dropped the line saying we charge by time, not coat size | Tell it which sentence has to stay |
| 2 | Summarise the 14-email thread with the van leasing company | Accurate overall; missed that the return date moved to the 28th | Ask for every date and figure as a separate list under the summary |
| 3 | Draft a reply to a puppy enquiry from my notes | Warm and well written, but it offered a "puppy package" we don't sell | Paste the real price list in with the notes |
Notice the pattern in the last column: each fix is about giving it facts instead of letting it fill gaps. By day 7 that lesson usually shows up three or four times, which is exactly what week two builds on.
Week 2 (days 8 to 14): briefing it well and checking what comes back
Most disappointing AI answers come from thin instructions. This week you learn to brief it like a capable new assistant who knows nothing about your business. Day 8 starts with a four-part brief you'll reuse all month:
ROLE: You're helping the owner of a [type of business] with
[number] staff. Our customers are [who they are].
CONTEXT: [What the reader needs to know: the job, the history,
the constraint. Paste the relevant email or notes.]
TASK: [Exactly what you want: draft, summarise, list, compare.]
FORMAT: [Length, tone, structure. For example: "Under 150 words,
friendly but not chatty, ending with one clear next step."]Filled in for an illustrative three-person joinery workshop, it looks like this:
ROLE: You're helping the owner of a joinery workshop with
3 staff. Our customers are homeowners who want fitted
wardrobes, alcove units and staircases.
CONTEXT: A customer wants painted alcove shelving either side
of a chimney breast. Each alcove is about 95 cm wide. We can
survey next Tuesday or Thursday. A pair usually costs $1,800
to $2,600 depending on doors and drawers. Lead time 5 weeks.
TASK: Draft a reply that gives the price range, offers both
survey dates and explains what moves the price.
FORMAT: Under 150 words, friendly but not chatty, ending
with one clear next step.Compare the two drafts in your log. On day 3, working from bare notes, the assistant wrote "prices start from around $1,200", a figure nobody gave it. With the brief it quoted the right range and both dates, but still promised "a free 3D design visual", which the workshop doesn't offer. That second slip is what day 10 is for: one standing line saying "never offer anything that isn't in the notes" stops it recurring.
| Day | Task | What you're learning |
|---|---|---|
| 8 | Redo day 3's enquiry reply using the four-part brief | How much better a proper brief makes the first draft |
| 9 | Add two past emails you liked as examples, and ask for a third in the same style | Examples beat descriptions of tone |
| 10 | Write your standing custom instructions: business facts, tone, words to avoid | Setting context once instead of every time |
| 11 | Ask for sources on a question about your market, then check every one | Citations can be real, misquoted or invented |
| 12 | Give it a small quote calculation and check each figure | Why AI arithmetic needs checking |
| 13 | Paste one of your own proposals and ask for the three weakest points | It's a useful critic when you ask it to be one |
| 14 | Save your five best prompts in one document | The start of a reusable prompt library |
Day 10 is the one people skip and then regret. How to give AI your business context shows what to put in custom instructions. And if day 12 goes badly, you're in good company: why AI is bad at maths explains exactly what went wrong.
Here's the sort of thing day 12 turns up. An illustrative window cleaner asks: "A customer wants 18 windows cleaned monthly at $3.50 each, plus the conservatory roof at $45 every three months. What's the yearly total, and the monthly amount if spread evenly?" The answer comes back looking tidy:
Windows: 18 x $3.50 = $63 a month, $756 a year
Conservatory roof: $45 x 3 = $135 a year
Total: $891 a year, or $74.25 a monthThe multiplication is fine; the reading is wrong. "Every three months" is four visits a year, not three, so the roof is $180, the year is $936 and the monthly figure is $78. Undercharging by $45 a year on every customer with a conservatory adds up. The fix is to ask it to list each item with the number of visits before it totals anything, so the assumption is visible, and to check that line first.
Day 11 teaches a different kind of caution. Suppose an illustrative florist asks for five sources on how much couples typically spend on wedding flowers. Five links come back, each with a neat one-line summary. Checked one by one, a typical result is: two open to real pages but the figure quoted isn't on either of them; one is a trade magazine page that exists but is four years old; one returns "page not found"; and one is fine. Log the score, one good source out of five, because that's the habit that matters: open every link before you repeat a number to a customer or put it in a proposal. Asking "quote the exact sentence from each source that supports the figure" makes the weak ones easier to spot, since an invented quote won't be on the page when you look.
Week 3 (days 15 to 21): data, privacy and money
This week is less hands-on and more about judgement. It's what stops you making expensive mistakes later.
| Day | Task | What you're learning |
|---|---|---|
| 15 | Read your assistant's data and privacy settings page in full | What happens to what you type |
| 16 | Read the pricing page, including the business plans, and list every term you didn't know | Seats, limits, credits and what business plans add |
| 17 | Run days 3, 5 and 13 again on a second assistant's free tier | The differences between tools are smaller than the marketing suggests |
| 18 | Try the AI features already built into your email, spreadsheet and accounting software | You may already own what you were about to buy |
| 19 | Upload a PDF or spreadsheet and check five figures it quotes back | Where file reading slips |
| 20 | Cost your three best uses: minutes saved a week × your hourly value, against the subscription | Whether it pays, in your numbers |
| 21 | Write one paragraph: what's worth paying for, and for whom | A decision, not just an impression |
Day 16 goes faster with the AI glossary for business owners open in another tab.
Day 20 is where impressions become numbers, so do the sum on paper. For an illustrative two-person bookkeeping practice valuing the owner's time at $40 an hour, a filled-in version reads:
| Use | Minutes saved a week | Minutes spent checking | Net minutes | Value a month |
|---|---|---|---|---|
| Chasing emails for missing receipts | 50 | 10 | 40 | about $115 |
| Summarising client calls into action points | 45 | 15 | 30 | about $87 |
| Explaining a tax-deadline change to clients in plain words | 20 | 15 | 5 | about $14 |
The monthly value is net minutes times 4.33 weeks, divided by 60, times $40. The third row is the useful one: the drafts are quick, but every line needs checking against the source, so it barely pays. Two uses clearly justify a $20 plan; the third stays a free-tier job, or goes. Leave the checking column out and these three uses look worth about $330 a month instead of about $216, which is how owners end up paying for tools that never earn their keep.
Week 4 (days 22 to 30): one task, automated end to end
The last week turns a chat habit into something that runs without you starting it. Pick a task that recurs at least weekly, has clear inputs, and would do little harm if the output were wrong once, because a person will still check it.
- Day 22, choose the task. Typical first picks: enquiry forms turned into a tidy summary and draft reply; invoices due this week turned into polite reminders; new reviews summarised each Monday.
- Day 23, draw it on paper. What starts it, what information it needs, what the AI step does, who checks it, where the result goes. For an illustrative domestic cleaning firm chasing invoices, the page reads: starts every Monday at 8 am; needs invoices due in the next 7 days or up to 30 days overdue, with client name, amount and due date; AI step drafts a reminder in the firm's tone, firmer if more than 14 days late; checked by the owner before 10 am; goes to the email drafts folder, plus a row in a "chased" sheet. Writing "firmer if more than 14 days late" down exposes a decision that belongs to you: that rule goes in a filter step, and the AI only writes the words.
- Days 24 to 26, build it. Zapier has a free tier, and Make's free plan includes 1,000 credits a month, which is plenty for a test. How to add AI steps to Zapier walks through the build.
- Day 27, test it on ten real examples and check every output against what you'd have written.
- Day 28, add the human check: drafts go to you for approval rather than straight to customers.
- Day 29, measure it. Time per item before and after, and how many drafts needed changes.
- Day 30, write one page: what you learned, what you'll pay for, what you'll try next month and what you'll stop doing.
Worked example: a sign maker's month
Take the owner of an illustrative four-person sign maker who follows the plan at about 25 minutes a day, roughly 12 to 13 hours across the month. They use the free tier for two weeks, then pay $20 for one month of a paid plan when they hit the limits.
By day 30, three things have changed. Quote cover emails, which took about 15 minutes each, now take about 5 with a saved prompt; at 12 quotes a week, that's two hours back. Weekly social posts drop from an hour to 20 minutes. And the week-four automation takes enquiries from the website form, pulls out sign size, type, fixing surface and deadline into a spreadsheet row, and drafts a reply for the owner to approve, saving around six minutes on each of ten enquiries a week.
That's roughly three and a half hours a week. At an illustrative $50 an hour, it's worth about $175 a week against $20 a month. Two lessons from the log matter as much as the time: an AI estimate of a shop front's width from a photo was 40 cm out, so quotes are never based on photos; and the owner decided not to buy the AI design tool a supplier had been pushing, because the free tools already covered the concept stage.
Free material worth your time, and what to skip
A course adds structure alongside the daily practice. Two free options are good fits for owners:
- Anthropic's AI Fluency: Framework and Foundations is free, needs no Claude account, and is built around four skills: delegation, description, discernment and diligence. That maps neatly onto weeks two and three. There's an optional assessment and certificate at the end.
- OpenAI Academy offers free courses, starting with AI Foundations. You need a ChatGPT account to track progress.
The help pages of whichever tool you picked are underrated too. They're where settings, limits and data handling are actually documented.
What to skip in month one: courses that teach coding (unless you want to build tools yourself), packs of "1,000 prompts", tool-of-the-week newsletters, and anything promising income from AI. After day 30, keeping up with AI in 30 minutes a week shows how to stay current without it taking over.
The day-30 self-test
Answer yes or no, honestly:
- Could you explain to a member of staff why AI sometimes invents facts?
- Do you know which of your business data can go into which tool?
- Do you have at least five saved prompts you've reused more than once?
- Could you estimate what AI would cost your team each month?
- Do you know what your assistant does with what you type?
- Have you caught an AI mistake yourself, before it mattered?
- Could you tell whether a vendor's "agent" is really one?
- Do you have one task that now takes less of your time?
Six or more yeses: you've learned enough to make good decisions, and the next step is rolling what works out to the team. Four or five: repeat weeks two and three with different tasks. Fewer than four: the month was probably too theoretical, so start week one again with the three tasks from your list that annoy you most.
Questions owners ask before starting
Do I need to learn to code to use AI in my business?
No. Everything in this plan, including the automation in week four, uses tools built for people who don't code. Coding helps if you later want custom tools, but most small-business gains come from using chat assistants well, adding AI steps to automation platforms, and switching on AI features in the software you run the business on.
Which AI assistant should I learn on?
Start with the one that fits the tools you already use. On Microsoft 365 business plans, Copilot Chat is included at no extra cost; Google Workspace business plans now include Gemini. Otherwise ChatGPT or Claude both have capable free tiers. The skills transfer between them, so the choice matters less than using one every day.
How much should I spend while I'm learning?
Between nothing and about $20 for the month. Free tiers cover weeks one and two for most people; a single paid plan such as ChatGPT Plus or Claude Pro at $20 a month helps if you hit limits. Don't buy team plans or specialist tools until day 30, when you'll know which uses are worth paying for.
Further reads
- How to Use AI on Your Own Admin First, Then Roll It Out — Why your own admin is the right place to practise first.
- AI Quick Wins: 12 Things a Small Business Can Set Up This Week — Twelve quick set-ups to try once the basics feel easy.
- How to Build a Shared Prompt Library for Your Team — Turn your best prompts into a library the whole team uses.
- ChatGPT Prompts for Small Business Owners: 50 Tested Examples — Fifty tested prompts to borrow and adapt during the month.
- AI for Small Business Owners: A Plain-English Beginner's Guide — The plain-English background if you want it before day one.
- Is AI Too Complicated for Non-Technical Business Owners? — An honest answer for owners who don't think of themselves as technical.
- How Much Time and Money Does AI Staff Training Take? — Training hours by role, free and paid course options with current prices, a costed plan for an eleven-person decorating firm, and how to check it paid off.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Anthropic Academy course page for AI Fluency: Framework and Foundations; OpenAI Academy; vendor pricing pages for ChatGPT, Claude, Microsoft 365, Google Workspace, Zapier and Make (checked September 2026).