For a small business, AI is a fast helper for the writing and admin that eats your day: customer emails, social posts, product descriptions, supplier letters and notes. It is not a safe place for client data, and it does not know your customers, your prices or your promises, so a person checks everything before it goes out.
- Use it for: first drafts of emails, posts, descriptions and routine admin, which you then edit.
- Do not paste: customer details, staff records, passwords or anything you promised to keep private.
- Never fake it: no AI-written reviews and no generated photos of your product or premises. In the US, fake reviews are banned by a federal rule.
- Write it down: a five-line team policy saves daily guesswork. A template is linked below.
- Check before sending: names, prices, dates and promises. If the AI gets it wrong, your business still owns the mistake.
What AI is good and bad at in a small business
The plain point: AI is quick at turning a few notes into a tidy draft. It is poor at knowing whether that draft is true for your business.
The comparison: think of a capable temp on their first morning. They type fast and write politely, but they have never met your customers and do not know your prices. They will guess rather than ask, unless you tell them not to.
The US Small Business Administration (SBA), the government agency for small firms, lists similar uses: repeat tasks, business content, brainstorming and customer service.1
| Job | How AI does | What you do |
|---|---|---|
| Draftingemails, posts, descriptions | Good at: a clean first version in seconds. | You: add the real details and your own voice. |
| Rewordingshorter, politer, plainer | Good at: changing how something is said. | You: check it did not add facts. |
| Factsprices, laws, dates | Bad at: it can state wrong facts with full confidence. | You: check every one against your own records. |
| Judgementrefunds, complaints, hiring | Bad at: it does not know the customer or the history. | You: decide, then let it help with the wording. |
Why does AI hallucinate? explains the confident mistakes.
Hand it the typing. Keep the facts and the decisions for yourself.
Five everyday jobs, with a prompt to start
The plain point: the best AI jobs are dull, frequent and easy to check. Each one below comes with a starter prompt. Change the parts in [brackets].
The comparison: a prompt (the request you type) is like a note for that temp. A note saying “reply to this” gets a vague reply. A note with the facts, the tone and the length gets something you can send after a quick edit.
1. Customer emails. Give it the facts of the case, never the customer’s personal details.
2. Replies to reviews. Draft as the owner, then edit and post it yourself. Answer the specific thing the reviewer said.
3. Social posts. Ask for options, then pick and edit one. Use your own photos.
4. Product or service descriptions. Give it the real list and forbid additions.
5. Admin letters. Supplier emails, opening-hours notices, rota messages: the work that is tedious and low-risk.
Running a shop, salon, garage or café? AI for a local business is the detailed page on this: why you should photograph your product instead of generating it, and the before-you-post checklist. How to ask AI well covers briefing habits.
Give it your real facts and tell it not to add any. Most bad AI drafts come from a request that left gaps for it to fill.
What never goes into a prompt
The plain point: customer details, staff records and passwords stay out. Once text is sent, you cannot take it back.
The comparison: typing into a chatbot is like posting a letter to another company. Write nothing you would not be happy for that company to keep on file.
The SBA gives the same advice in its own words: “Try not to feed any sensitive data or proprietary information.”2
- Customer personal data: names with addresses, phone numbers, order histories, card details.
- Staff matters: pay, health, performance, a CV someone sent you.
- Logins: passwords, card numbers, account keys.
- Anything confidential: a contract under a confidentiality agreement, a client’s plans.
The account matters, not just the tool. A personal account and a business account from the same company can have very different terms. Anthropic’s consumer terms, for example, say they do not cover its business products.3 What happens to what you type explains storage, training and human review in four simple questions.
You can still use AI with customer work. Take the names and details out first. “A customer whose parcel is three days late” gets just as good a reply as the real name and address would.
A short AI policy for your team
The plain point: if more than one person uses AI in your business, write down a few rules. It takes ten minutes and stops everyone guessing.
The comparison: it is like the note by the till that says how refunds work. No manual, just a clear answer.
A workable small-business version answers five questions:
- Which tools, on which account? Name them. Say whether personal accounts are allowed for work.
- What may go in? Use the list in the section above. Removing names and details first is encouraged.
- What must be checked? Prices, dates, names and promises, before anything reaches a customer.
- Who owns the result? The person who sends it. Using a tool does not move the blame to the tool.
- Who do I ask? One named person, so questions get a quick answer.
A team AI policy you can actually adopt has a copyable one-page template built on the same questions, and explains why it says “do not enforce with AI detectors”. Using AI at work covers the same rules for one person.
Put this sentence in your policy word for word: the person who sends it owns it.
Check before it goes out
The plain point: anything a customer will read gets a human check. If an AI tells a customer something wrong, your business answers for it.
The comparison: AI is a borrowed van. It does the carrying, but you are still the driver, and the parking ticket comes to you.
A Canadian case shows it. When Air Canada argued that its website chatbot was a separate legal entity, a tribunal disagreed.4 The SBA suggests a second pair of eyes.2
Four things to check every time:
- Facts: prices, opening hours, dates, delivery times, ingredients.
- Promises: did it offer a refund, a discount or a deadline you never agreed?
- Names: the customer’s, your product’s, your town’s.
Two lines never to cross. Do not let AI write reviews of your own business, and do not post generated pictures of your product or premises as if they were real. In the US, the Federal Trade Commission (FTC), the consumer-protection agency, has a rule against fake reviews that names AI-generated ones.5 That is a US rule; other countries handle fake reviews under their own consumer law. AI for a local business explains both lines in detail.
Auditing AI-generated work goes deeper on checking.
Simple automation and AI agents
The plain point: some AI tools can now do things, not just write: send emails, fill in forms, run a task every morning. Start with jobs that produce a draft for you to approve.
The comparison: a chatbot is a helper who writes a note for you to send. An agent (an AI that takes actions with your accounts) is a helper you have handed your keys. Give a new helper one key, not the whole bunch.
The risk is that an agent reading emails or web pages can be tricked by hidden instructions in them. This is called prompt injection, and even the companies say it is not solved.6
- Good first jobs: a morning summary of new enquiries, drafted replies waiting for you, a weekly list of overdue invoices.
- Keep for later: anything that pays, sends to customers or deletes without asking you first.
- Keep out: your bank account and your main business email, while you learn.
AI agents for beginners explains agents and their risks. Permission tiers for agents shows how to decide what an agent may do alone, and Meta Muse vs Gemini Spark vs Grok Bot compares the best-known ones.
An agent that cannot pay cannot be tricked into paying. Limit what it can touch, and keep the final yes for yourself.
Where to go next
Pick the line that sounds like you.
- “I run a shop, salon or café.” Read AI for a local business.
- “I need rules for my staff.” Use the team AI policy template.
- “Which AI tool should we use?” Start with choosing an AI model and free vs paid AI chat.
- “What happens to what we type?” See what happens to what you type.
- “I want the bigger picture on AI at work.” Read AI at work for beginners.
- “I sell property.” See AI images in property listings.
- “Show me everything on this.” Open the Living with AI hub.
Questions people ask
What business information should I never paste into AI?
Customer personal details, staff records, passwords and card numbers, and anything confidential. The US Small Business Administration says: “Try not to feed any sensitive data or proprietary information.”2 Take names and details out first, and the reply is just as good.
Can I use AI to write reviews for my business?
No. Do not let AI write reviews of your own business. In the US, the Federal Trade Commission has a rule against fake reviews that names AI-generated ones.5
Who is responsible if AI tells a customer something wrong?
Your business is. When Air Canada argued that its website chatbot was a separate legal entity, a Canadian tribunal disagreed.4 Anything a customer will read should get a human check first.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. Which of these should stay out of a chatbot prompt?
- The tone you want, such as warm and plain
- A customer’s name, address and order history
- The length you want the reply to be
Show the answer
Answer: A customer’s name, address and order history. The page says customer details, staff records and passwords stay out; take names and details out first and the reply is just as good.
2. What are the two lines the page says never to cross?
- Using AI before 9am or after 5pm
- Using more than one AI tool or account
- AI-written reviews of your business and generated photos posted as real
Show the answer
Answer: AI-written reviews of your business and generated photos posted as real. The page says never let AI write reviews of your own business and never post generated pictures of your product or premises as if they were real.
3. If an AI draft tells a customer something wrong, who answers for it?
- Your business: the person who sends it owns it
- The company that made the AI tool
- No one, since a computer wrote it
Show the answer
Answer: Your business: the person who sends it owns it. The page says anything a customer reads gets a human check, because your business still owns the mistake.
Related on this site: How to Write Excel Formulas With AI
- 1US Small Business Administration, AI for small business (business guide, no date shown), read at source 1 Oct 2026AI can help your business “Take on repeat tasks”, “Create business content”, “Collaborate and brainstorm” and “Improve customer service.” ↑
- 2US Small Business Administration, AI for small business, read at source 1 Oct 2026“Try not to feed any sensitive data or proprietary information.” ↑a“have another person review all AI products.” ↑b“Try not to feed any sensitive data or proprietary information.” ↑c
- 3Anthropic, Updates to Consumer Terms and Privacy Policy, 28 Aug 2025, read at source 16 Sep 2026 (as cited on this site’s linked pages): the updates “do not apply to services under our Commercial Terms, including Claude for Work”.↑
- 4McCarthy Tétrault, Moffatt v. Air Canada, 19 Feb 2024, read at source 16 Sep 2026 (as cited on this site’s linked pages), quoting the tribunal: “It makes no difference whether the information comes from a static page or a chatbot”.↑a↑b
- 5US Federal Trade Commission, Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials, 14 Aug 2024, read at source 1 Oct 2026the rule covers “reviews and testimonials that misrepresent that they are by someone who does not exist, such as AI-generated fake reviews”. ↑athe rule covers “reviews and testimonials that misrepresent that they are by someone who does not exist, such as AI-generated fake reviews”. ↑b
- 6Meta AI Research, How We Built Safety Into Muse, 8 Sep 2026, read at source 1 Oct 2026 (as cited on this site’s linked pages): “Prompt injection remains an open problem in the industry”.↑
US Small Business Administration, AI for small business (business guide, no date shown), read at source 1 Oct 2026.
Anthropic, Updates to Consumer Terms and Privacy Policy, 28 Aug 2025, read at source 16 Sep 2026 · McCarthy Tétrault, Moffatt v. Air Canada, 19 Feb 2024, read at source 16 Sep 2026 · Meta AI Research, How We Built Safety Into Muse, 8 Sep 2026, read at source 1 Oct 2026.
Read dates for the last three are those recorded on this site’s linked detail pages. The comparisons on this page (first-morning temp, note for the temp, posted letter, note by the till, borrowed van, handing over keys) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK