Most small business owners have tried ChatGPT or Claude once, got a bland answer, and gone back to doing things by hand. The problem is rarely the tool. It’s the request. AI does well when you give it a specific job, the context a new employee would need, and an example of what good looks like.
Below are ten jobs that work reliably today, each with a prompt you can copy. Replace anything in square brackets.
First, the one rule
Never paste anything into a public AI tool that you wouldn’t email to a stranger. That means no client names with financial details, no ID numbers, no bank details and no passwords. Replace real names with placeholders (“Client A”) or use a business plan that states your data isn’t used for training. Everything below works fine with anonymised details.
1. Turn rough notes into a clear client email
You know what you want to say; you just don’t want to spend twenty minutes wording it.
Rewrite these notes as a short, friendly email to a client. Plain English, no jargon, under 120 words, with one clear request at the end. Notes: [paste notes]
Good output is shorter than you’d write yourself. If it isn’t, ask for “half the length”.
2. Write the reply you keep putting off
Difficult replies (a price increase, a missed deadline, a complaint) are where AI saves the most stress.
A client is unhappy that [situation]. Draft a calm, professional reply that acknowledges the problem, explains what happens next, and doesn’t over-apologise. Keep it under 150 words.
3. Summarise a long document
Contracts, policies, long email threads and meeting transcripts.
Summarise this in five bullet points for a busy business owner. Then list anything that needs a decision or has a deadline, with the date. [paste text]
Always check dates and numbers against the original. Summaries are usually right; specific figures are where mistakes hide.
4. Build a checklist for a process you repeat
Onboarding a client, closing the month, opening the shop, publishing a product.
I run a [type of business]. Write a step-by-step checklist for [process], in the order it should be done. Include who usually does each step and what can go wrong.
Edit the result to match how you actually work, then keep it. A checklist written once saves the same mistakes being made every month.
5. Turn a checklist into a spreadsheet
AI is good at structure. Ask for the columns and formulas, not just the idea.
Design a Google Sheets tracker for [purpose]. Give me the column headings, which columns should be dropdowns and their options, and the formulas for any calculated columns.
6. Draft product or service descriptions
Write a description for [product or service] for [who buys it]. Lead with the problem it solves, then what’s included, then who it’s not for. Under 150 words, no hype words like “revolutionary” or “seamless”.
Banning hype words in the prompt does more for quality than any other single instruction.
7. Plan a month of social posts
I sell [product/service] to [audience]. Suggest 12 post ideas for this month: 4 that teach something useful, 4 that show behind the scenes, 4 that answer common customer questions. One line each.
Write the posts yourself from the best ideas, or ask for drafts of two or three at a time so you can steer the tone.
8. Research a question before you call an expert
I’m a small business owner in [country] trying to understand [topic]. Explain the basics in plain English, list the questions I should ask an accountant or lawyer, and tell me what you’re unsure about.
Asking the AI to name what it’s unsure about is the habit that separates useful research from confident nonsense. Treat the answer as preparation for the expert conversation, not a replacement for it.
9. Clean up messy data
Lists of customers with inconsistent capitals, addresses in one cell, dates in three formats.
Here is a messy list. Clean it into a table with columns [A, B, C]. Standardise names to Title Case and dates to YYYY-MM-DD. Flag any rows you weren’t sure about. [paste anonymised data]
10. Prepare for a meeting
I’m meeting [type of person] about [topic]. Give me a one-page brief: what they probably care about, three questions to ask, two objections they might raise and how to answer them.
Making answers better
If an answer is generic, the fix is almost always one of these:
| Problem | Add to your prompt |
|---|---|
| Too long or wordy | “Under 100 words” or “half the length” |
| Too generic | Who it’s for, your business type, a real example |
| Wrong tone | “Write like a friendly accountant, not a salesperson” |
| Made-up details | “Only use the information I gave you. Say if something is missing.” |
From one-off prompts to automation
Once a prompt works well, you can stop copying and pasting. Tools like n8n and Zapier can send a new form submission, email or spreadsheet row to an AI model and put the answer where you need it: a drafted reply in your inbox, a summary in your CRM, a tidy row in a sheet. That’s where AI stops being a chat window and starts saving hours every week. Automation guides are coming next in this hub.
If you run an accounting or tax practice, the tax and accounting firms hub shows how these same ideas fit into onboarding and monthly document collection.
