AI will answer any question you ask it. That is the risk.

Published 2026-08-09 by Woodruff Training

Somewhere between the quote you did not want to pay for and the accountant you did not want to bother at nine on a Sunday evening, there is now a chatbot that will answer anything, instantly, for nothing. Small business owners have noticed. Redundancy questions, contract wording, what a clause in the insurance schedule actually means: it all goes into the same box now.

That is not a foolish thing to do. It just needs one piece of judgement that nobody hands you with the tool.

Confident is not the same as correct

An AI tool writes every answer in the same assured tone. It has no sense of the difference between something it has seen a thousand times and something it is more or less guessing at, so the answer about your dismissal process arrives sounding exactly as certain as the answer about how long to boil an egg.

Four things go wrong more than owners expect:

• It answers as if you were American. A great deal of what these tools learned came from the United States. Employment advice that would be perfectly sound in Texas can walk you straight into a tribunal claim here.
• It may be working from an older version of the world. Thresholds, rates and rules change with each Budget. The tool will quote you the figure it learned without mentioning that it has moved.
• It invents specifics that sound real. Ask which regulation applies and you can get an official-sounding name with a section number attached to it that does not exist. Lawyers have been caught filing court documents citing cases that were never heard.
• It knows nothing about you. It does not know one of your staff is on a contract from 2014, that your insurer requires a particular control, or that you took a payment holiday two years ago. It answers the general question, and your question is never the general one.

The same applies to technical fixes, which is where we see it most. Somebody cannot receive email from a supplier, asks an AI tool why, and gets a workable instruction to relax a filter or add an allow rule. The email starts arriving. A protection you were relying on is now off, and nobody wrote that down anywhere.

Where it earns its place

None of this is an argument for banning it, any more than it is for staff using AI day to day. Used properly it saves real money:

• Translating jargon. Paste in the paragraph of the contract or the insurance schedule you cannot follow and ask what it means in ordinary English. Low risk, and it makes you a better reader of your own paperwork.
• First drafts. A job advert, a policy, an awkward email to a customer. You were going to rewrite it anyway.
• Preparing for the professional. Ask it what you should be asking your accountant or solicitor. Turning up with sharp questions is the cheapest way to shorten a billable hour, and almost nobody does it.
• Getting through something long you would otherwise never read at all.

Notice what those have in common. Every one of them uses AI to help you think, and not one of them lets it decide.

Ask what being wrong would cost

That is the whole test, and it takes about two seconds.

If a wrong answer costs you an afternoon, go ahead and trust it. If a wrong answer costs you money, someone's job, a client, or a letter from a regulator, then AI is where you start rather than where you stop. Employment, tax, contracts, data protection and anything a regulator oversees all sit firmly in the second group.

Worth remembering too: "the AI told me" has never once worked as a defence. HMRC, the ICO and an employment tribunal will hold you responsible for the decision, whatever helped you reach it. The tool takes none of the risk, which is rather the point of it being free.

If you would rather your managers made that call consistently, our free AI course for decision makers covers this ground, and we are happy to talk it through.