By Rakibul Hasan, Moonlab
ChatGPT and similar AI assistants can save a small business real time on writing, summarising, planning and sorting, if you know what to ask and what to check. They can also produce confident mistakes and are a risk if you paste in the wrong data.
This guide covers practical uses of ChatGPT for small business, how to write better requests, the limits to respect, and how to handle data with care. We use ChatGPT as the example, but the advice applies to other assistants such as Claude and Gemini.
What ChatGPT is good at in a small business
These tools are strongest at working with language. They are good at drafting, rewriting, summarising, explaining and structuring text, and at turning messy notes into something clean. They are weaker at facts you have not given them, exact arithmetic, and anything that needs current information they may not have.
A useful mindset is that it is a fast, well-read assistant that sometimes makes things up. Give it the material, ask for a first draft, then check and edit. You remain responsible for what goes out under your name.
Practical use cases you can try this week
These are everyday jobs where ChatGPT business use tends to pay off. Start with one or two.
- Email drafts. Paste a customer message and your key points, and ask for a short, polite reply in your tone.
- Meeting and call notes. Paste a transcript or rough notes and ask for decisions, action items and owners.
- Quotes and proposals. Turn a bullet list of scope into a clear proposal, then adjust the numbers yourself.
- Job descriptions and policies. Get a first version of a job advert or staff handbook section to edit.
- Product and service descriptions. Rewrite your own descriptions for different audiences or lengths.
- Spreadsheet help. Ask for a formula, or for an explanation of an unfamiliar one.
- Customer FAQs. Turn a pile of past questions into a clear FAQ page.
- Brainstorming. Ask for ten ideas for a campaign, a newsletter topic or a name, then pick and refine.
How to write requests that give useful answers
Vague requests give generic answers. A good request has four parts: the task, the context, the format and any limits.
For example: "You are helping a small dental clinic in Bristol. Write a friendly reminder text for patients with an appointment tomorrow. Under 300 characters. Include the clinic name and a way to reschedule. Do not give medical advice." Then follow up: shorten it, change the tone, try three versions. Keep a note of requests that work well and reuse them. If you do the same task weekly, that is a sign it could be automated, as covered in moonlab.space/guides/workflow-automation-examples.
The limits you need to respect
AI assistants can state wrong things in a convincing way. This is often called hallucination. It is more likely when you ask about specific facts, laws, prices, citations or recent events.
- Check facts, figures, names and links before you publish or send anything.
- Do not rely on it for legal, tax, medical or financial advice. Use it to prepare questions for a professional instead.
- Do not let it make promises, discounts or commitments on your behalf without a person reading them.
- Expect a generic voice. Edit so the text sounds like you and includes details only your business knows.
- Tell it what you want left out. It tends to pad with filler.
Data care: what not to paste in
Treat anything you type into an AI tool as data you are sharing with a third party. Before you paste, ask whether you would be comfortable if that data left your control.
Avoid pasting health information, payment card numbers, passwords, government ID numbers, and customer lists with personal details unless you have checked the tool's terms and have a lawful basis. Replace names and details with placeholders where you can. Read the provider's settings on whether your content can be used to train models, and whether a business or team plan changes that. Plan names and terms change, so check the vendor's current documentation. If you handle European or UK customer data, see moonlab.space/guides/gdpr-ai-automation. This is not legal advice.
Free accounts, team plans and company rules
Many small teams start with personal accounts. This quickly becomes untidy: work content sits in staff members' private histories, and nobody controls the settings. Business or team plans from the main providers generally add shared admin controls and different data-handling terms, but features and terms vary, so read the current details before choosing.
Write a one-page rule for staff: what tools are allowed, what data must never be pasted, who checks AI-written content before it goes to a customer, and where to ask questions. Short and clear beats long and ignored.
A worked example: turning notes into a customer reply
Say you run a small bathroom fitting firm in Bristol, and a customer emails asking why their quote is higher than a friend's. You do not want to sound defensive, and you have ten minutes between jobs. You paste the customer's email into the assistant with this request: "Draft a polite, short reply from a bathroom fitter. Explain that our quote includes removing the old suite, waterproofing and a 5 year workmanship guarantee. Do not mention other companies. Do not promise a discount. Offer a call to talk it through." Replace any real names or addresses with placeholders first.
The assistant returns a draft in seconds. You check it against what you actually include, delete a sentence that overstates something, add the customer's first name and send it. The tool saved the blank-page effort and you kept the decisions. That is the pattern that works: you supply facts and limits, it supplies wording, you approve.
- Mistake: pasting customer lists or sensitive details. Use placeholders and keep personal data out where you can.
- Mistake: copying the first answer. Ask for two or three versions and then edit.
- Mistake: trusting figures and legal claims. Check every number, date and rule against a reliable source.
- Mistake: letting everyone use their own account. Agree one set of approved tools and rules.
- Mistake: expecting it to know your business. Paste in the relevant price list, policy or notes each time.
- Decision rule: if a wrong answer would cost money, safety or trust, a person checks it before it is used.
From chat to automation
Chatting with an assistant is manual: someone has to open it, paste, and copy the result back. When the same task repeats, such as sorting incoming email or drafting replies, the same AI can run inside an automation that does the pasting for you and hands the output to a person to approve. Start with chat to learn what works, then automate the tasks that prove their value.
If you are not sure which tasks are worth automating, the free audit at moonlab.space/contact can help you pick. Our overview of the wider picture is in moonlab.space/guides/ai-automation-for-small-business.
