What Should I Not Use AI For in My Business?

What should I not use AI for in my business?

Straight answer

Avoid AI for anything needing guaranteed accuracy, real judgement, or accountability: final legal, tax and medical decisions, unchecked handling of personal or payment data, and any output sent to customers without review. AI predicts plausible answers, not correct ones, so wherever a confident error would cause real harm, keep a person firmly in charge.

Information current as at 5 July 2026

Knowing what not to use AI for is as valuable as knowing what to use it for, and it is where a lot of avoidable damage is prevented. The line is not arbitrary. It follows directly from what AI actually is: a system that produces plausible answers, not guaranteed correct ones. Wherever that gap matters, AI belongs on a short leash.

Plain English

High-stakes task: Work where a mistake causes serious harm, cost or legal consequence.
Confabulation: AI producing a confident, fluent answer that is simply untrue.
Regulated advice: Guidance, like legal, tax or medical, that carries professional and legal duties.
Final decision: A choice acted on directly, with no human review before consequences follow.

The principle behind the line

AI does not know things; it predicts plausible output based on patterns. Most of the time the plausible answer is also correct, which is what makes it useful, but it is not reliable in the way a fact is reliable, and it will state a wrong answer with exactly the same confidence as a right one. The tasks to avoid, then, are the ones where you need a guarantee of accuracy, a real exercise of judgement, or a person who can be held accountable, because those are precisely the things a plausibility engine cannot provide. Everything below is an application of this single idea.

Decisions that carry professional or legal weight

Do not use AI to make final legal, tax, financial or medical decisions for your business. It can help you prepare, draft a question, summarise a document, explain a concept in plain terms, but the actual decision or advice must come from a qualified person who carries the professional responsibility. AI will confidently state incorrect law, invent a case, or misjudge a tax rule, and acting on that can cost you dearly with no one accountable but you. Use it as a research assistant to make the qualified person faster, never as a replacement for their judgement and their duty of care.

No pressure

Show us what you built.

If you have made something and it needs to become real, send it over. We will tell you honestly what it needs to be live, safe and yours, whether that is a quick fix you can do or a proper build. No obligation.

Anything with sensitive data or real consequences unchecked

Do not hand AI unchecked control over personal or payment data, or over actions with real consequences. Automatically emailing customers, issuing refunds, changing records, or making commitments without a human check invites a confident error straight into your business and your customer relationships. The same caution applies to putting confidential or regulated data into tools whose terms you have not verified. The rule across all of these is the human check: AI may prepare and propose, but a person approves before anything consequential actually happens, especially where money, data or a customer is involved.

The grey areas and how to judge them

Plenty of tasks sit between clearly fine and clearly forbidden, and a simple test resolves most of them. Ask: if the AI were confidently wrong here and nobody caught it, how bad would it be? If the answer is trivial, an internal draft, a first-pass summary, use it freely. If the answer is serious, a customer harmed, a legal exposure, money lost, a reputation dented, then either keep a firm human check or do not use AI at all. The strength of the leash should match the size of the consequence. This one question turns a vague worry into a clear, defensible decision for almost any task you are unsure about.

Questions, answered

Can I use AI for legal or accounting work?

Only to assist, never to decide. It can help you draft a question, summarise a document or understand a concept, but final legal, tax or financial decisions must come from a qualified professional who carries the responsibility. AI will state wrong law or misjudge a rule with full confidence, and acting on that leaves no one accountable but you.

Is it safe to let AI reply to customers automatically?

Not without a human check, in most cases. Unreviewed automatic replies risk sending a confident error straight to a customer, damaging trust or making a commitment you did not intend. Let AI draft the reply and have a person approve it, at least until you have strong evidence it is safe for a given, low-stakes type of message.

How do I decide about a task that is not clearly off-limits?

Ask one question: if the AI were confidently wrong here and nobody noticed, how bad would it be? Trivial consequences mean you can use it freely; serious ones mean keep a firm human check or avoid it. Matching the strength of oversight to the size of the potential harm resolves almost every grey-area case sensibly.

Why is a confident wrong answer such a problem?

Because AI states falsehoods with the same fluency and assurance as truths, so a wrong answer does not look wrong. This is exactly why it cannot be trusted unchecked on anything that matters: the errors are easy to miss precisely when they are most costly. Keeping a human check on consequential work is how you catch them before they cause harm.