How to use AI at work without getting caught out

My new book, AI at Work: From Your First Prompt to Your Own AI Assistants, is out on Amazon. It’s a plain-English handbook for people using AI in a UK workplace, or about to start, and it covers the whole distance. Chapter 3 is a free account and an email that needs fixing. Chapter 27 is an automation that sorts a plumbing firm’s enquiries, sends gas emergencies to the on-call engineer by a plain rule rather than by AI, and drafts replies that a person still reads and sends.

I build web apps and look after Linux servers for small businesses, and I work with AI most days. The book isn’t really about the technology. It’s about three habits that decide whether AI saves a business time or quietly costs it something.

Habit one: brief it like a new colleague

The tool knows nothing about your business. Ask it to “write a job advert for an office assistant” and you’ll get a competent advert for no company in particular, padded with “team player” and “growing business”. Tell it who the advert is for, what the job involves, what matters to you and what to leave out, and you get something you could post.

The quickest version of this is to let the tool ask. Prompt 4.5 in the book does exactly that:

I need help with [THE JOB]. Before you write anything, ask me up to five questions about what you’d need to know to do it well. Ask them all at once, numbered, so I can answer them in one go.

Most people never try it, and it’s the single cheapest improvement there is.

Habit two: decide what goes in before you paste

Chapter 8 gives you three questions to ask before you paste anything. Whose information is it? What happens if it got out? And is this tool, on this account, allowed to hold it?

The answers sort material into three colours. Green goes into any account. Amber goes only into a work account, under your employer’s rules. Red, which is anything like a customer’s health details, a staff disciplinary file or a client’s bank statements, goes in only after the business has deliberately decided it can. Passwords never go in at all.

None of that needs a lawyer. It needs a minute’s thought before the paste, and the book gives you a one-page AI policy (Appendix B) you can adapt for your own team.

Habit three: check the specific things

AI tools are fluent, and fluent is not the same as right. In December 2023 a tax tribunal found that nine earlier decisions a taxpayer had relied on in her appeal didn’t exist. I’ve written up the full story, and the two that followed it, on my books site.

The rule the book keeps coming back to is short: the more specific and checkable a claim is, the more you should check it. Names, numbers, dates, prices, quotes and anything presented as a law or a rule. You don’t need to check every sentence. You do need to check those.

The book holds itself to the same rule. Every prompt in it was tested before it went in, and where a test found a problem, the book says so. One example from chapter 25: asking a tool to explain what a short script would do turned up a bug in the script printed in the book. A scanned file called “Scan [1].pdf” was being skipped without a word of warning. The fix is in the chapter, and so is the story of how it was found.

What’s in it

It climbs in four steps, and you can stop at any of them.

• Beginner. What these tools are, how to choose one (ChatGPT, Claude, Gemini or Copilot, including what the free plans won’t do), how to ask well, and how to use them safely.
• Everyday tasks. Email, meetings, spreadsheets, research, reports and proposals, images and video.
• AI in your job. Sales and customer service, marketing, HR, finance and admin, trades and site work, and running a small business.
• Advanced. Building your own assistants, splitting big jobs into steps, getting formulas and short scripts written and checked, agents, automations without code, AI that runs on your own computer, and working out with real numbers whether any of it is saving you time.

Four invented people go through it with you: an office manager at a plumbing firm in Leicester, a letting agent in Swindon, a dental receptionist and a one-man bookkeeper. Their wins are in it, and so are their mistakes. The law, the prices and the menus are UK ones, as of October 2026.

The prompts come with it

The book has 246 numbered prompts. Buy it and you also get every one of them on a web page, with a copy button, a search box and the safety notes, so you don’t have to type them out from the paperback. The address is printed in the book.

Where to get it

AI at Work is on Amazon UK as a Kindle eBook at £5.99, free to read with Kindle+, a paperback at £14.99 and a hardcover at £22.99.

If what you’re weighing up is a chatbot on your own website rather than your staff using AI, that’s a different job with different risks. I’ve written about that separately: is an AI chatbot safe for my business?