// the tuesday guide

The 7 mistakes small businesses make with AI (and how to avoid them)

2026 · 09 · 29 — THE TUESDAY GUIDESeven AI mistakes to avoid

For two years now, everyone has been telling small businesses to “use AI.” Almost nobody tells them how not to use it. Yet the real trouble rarely starts with the tool: it starts with bad habits, always the same ones. A text published without a second read, a contract pasted into a chat, a chatbot left free to promise refunds. This is an anti-guide: seven mistakes, with real cases, and for each one what to do instead. If you recognize one of them, you're not alone — and it can be fixed in an afternoon.

1 · Publishing without a second read

AI always writes in the same confident tone, whether it's right or making things up. And it makes things up more often than you'd think: a number, a regulation, a date, a quote. The best-known case in Italy dates from March 2025: the Court of Florence found, in a defense brief, several rulings of Italy's Supreme Court that didn't exist. ChatGPT had “found” them, and nobody at the law firm had checked them before filing.

You don't have to be a lawyer to fall into the same trap. All it takes is a quote with the wrong tax rate, a product page with a made-up measurement, a post citing a law that doesn't say what the post claims.

2 · Pasting into a chat what you shouldn't

In the spring of 2023, some Samsung engineers pasted pieces of confidential code and notes from an internal meeting into ChatGPT to get help. Within a few weeks the company banned generative AI tools on all corporate devices. Nobody had bad intentions: they just wanted to work faster.

In a small business the same thing happens, on a smaller scale: the customer list with phone numbers, the supplier contract, the payslip to “summarize,” the password typed into the message for convenience.

3 · Letting the chatbot make promises

It's the story every business with a chatbot should know. A customer asks the Air Canada chatbot how bereavement fares work. The chatbot tells him he can buy the ticket at full price and request a refund afterward. It wasn't true: the airline's policy said the opposite. The customer appeals, and in February 2024 a Canadian tribunal rules in his favor. The airline's defense — that the chatbot was “responsible for its own actions” — is rejected: whatever the chatbot on your website says, you're the one saying it.

Around the same time, in the United States, a car dealership's chatbot was talked into “selling” an SUV for one dollar. That one ended in laughter, but the mechanism is the same.

4 · Buying the app before you have the problem

Three active AI subscriptions, twenty euros each, and none of them actually used. It's perhaps the most common mistake and the least visible, because it does no damage: it just drains money every month. It happens when you start from the tool everyone's talking about (“I need that one too”) instead of from the work you want off your plate.

The right method runs the other way. Take a sheet of paper and write down the three tasks that steal the most hours from you every week: answering the same emails, preparing quotes, writing posts, copying data from one file to another. Only then do you look for the tool for the first one on the list.

5 · The one-line prompt (and the conclusion that “it doesn't work”)

“Write me a post about the sale.” Out comes a clean, correct and perfectly useless text: any store in Italy could have written it. From there, the most common conclusion: “AI is no use for my line of work.” In reality the AI did exactly what it was asked, which is almost nothing.

Think of a newly hired employee. You would never just tell them “write a post” and leave it at that: you'd tell them who you are, who you're talking to, what tone you use, what must never be said, and you'd show them a post you liked. With AI it works the same way.

6 · The last two: copy that all sounds the same, and AI used on the sly

Mistake 6 — losing your voice. When every post, email and description comes out of the same chatbot untouched, after a month they all look alike. And they look like your competitors', who use the same tool. A customer who picks a small business does so because there's a person on the other end: if AI erases that difference, you've saved time and lost the reason they chose you. The fix: keep a sheet with your typical phrases and the words you never use, and always add something only you could know — a customer, a detail from the workshop, an opinion.

Mistake 7 — the AI everyone in the company uses, but nobody knows about. The employee who translates emails with one app, the intern who writes posts with another, the owner who pastes quotes into a third. Each with their own free account, nobody knowing where the data ends up. You don't need a twenty-page policy: one page is enough, listing which tools are used, with which account, and what never gets pasted. Plus half an hour to talk it through out loud.

In short

AI doesn't do damage on its own: it does damage when we use it without rules. The seven mistakes to avoid are publishing without a second read, pasting confidential data into a chat, letting the chatbot promise discounts and refunds, buying subscriptions before knowing what to solve, writing one-line prompts, losing your own voice, and letting everyone in the company quietly use different tools. The fix, almost always, is the same: the AI prepares, a person checks and decides. An afternoon and one page of written rules are enough to stay on the right side.

Frequently asked questions