Say you run a small catering business. You ask an AI tool to write a page for your website about food safety rules in your state. Thirty seconds later you have something polished, professional, and confident.
And one of the “rules” it lists does not exist.
This is not a rare glitch. It is the single most important thing to understand about tools like ChatGPT, Claude, and Gemini: they sometimes make things up, and they do it in exactly the same confident tone they use when they are right.
The good news is that once you know why it happens and where it tends to happen, catching it becomes a quick habit instead of a constant worry.
Why AI makes things up
Here is the plain version. These tools do not look up facts in a big library of true things. They write by predicting which words are likely to come next, based on the enormous amount of text they learned from. That makes them very good at producing writing that sounds right.
Sounding right and being right are not the same thing.
When a tool does not actually know something — a specific price, a local regulation, the name of a study — it usually does not stop and say so. It fills the gap with something plausible. Technical people call this “hallucination,” which is just a fancy word for confidently filling a gap with something that sounds true.
It is not lying, exactly. Lying means knowing the truth and hiding it. This is more like a very eager new employee who would rather guess than admit they do not know. Useful to have around. Not someone whose guesses you print on a menu.
Where this bites a business
Made-up details cause trouble in a few predictable places:
- Invented facts. Ask for “five interesting facts about coffee” for your newsletter, and one of them may be fiction. It will read exactly like the four that are true.
- Wrong prices, dates, and numbers. If you did not give the tool a number, be suspicious of any number it gives you back.
- Laws and rules. Tax thresholds, food handling requirements, employment rules — AI can describe these fluently and still get them wrong, or describe last year’s version as if it were current.
- Fake-sounding references. Ask for sources and you may get studies, books, or articles with realistic titles and authors that simply do not exist.
- Details about your own business. If you never told it your opening hours and it states them anyway, it guessed.
Notice the pattern. The danger zone is specifics: names, numbers, dates, rules, and anything you did not supply yourself.
A simple trust-but-verify checklist
You do not need to fact-check every sentence. You need a filter. Before you use anything an AI wrote, run through four quick questions:
- Would it matter if this were wrong? A cheerful caption under a photo of scones: low stakes. A claim about allergens: high stakes. Spend your checking time where a mistake would cost money or trust.
- What are the checkable claims? Underline every name, number, date, price, rule, and quote. Those are the parts that can be wrong in a way that matters.
- Can I trace each one to a source I trust? That means an official website, your own records, or a supplier’s price list — not the AI itself. Asking it “are you sure?” is not verification. The tool may apologize and change its answer, or confidently repeat the mistake.
- Is this legal, medical, or tax territory? Then AI is a starting point for the questions you bring to a professional, never the final answer.
What you can relax about
Plenty of everyday AI work carries almost no risk of made-up facts, because there are no facts involved for it to invent.
Generally safe to trust:
- Rewording something you wrote, in a different tone or length
- Brainstorming ideas, names, and angles for you to judge
- Summarizing text that you pasted in yourself
- Turning your rough notes into a tidy draft
Always needs checking:
- Any fact, figure, or date you did not provide
- Anything touching laws, taxes, health, or safety
- Quotes, statistics, and references of any kind
One trick makes drafting much safer: tell the tool not to guess. Copy either of these and adjust the details:
Draft a frequently-asked-questions page for my mobile dog-grooming business using only the facts below. If a detail is missing, write [CHECK THIS] instead of guessing.
Rewrite this paragraph so it is warmer and shorter. Do not add any new facts, numbers, or claims — only use what is already here.
Both instructions draw the same clear line: your facts, its wording. That split plays to what AI genuinely does well and keeps its guessing habit out of your business.
Try this today
Find one thing an AI has written for you recently — or ask one right now for a short piece about your trade. Paste it into a note and underline every checkable claim: names, numbers, dates, rules.
Then spend ten minutes verifying each one against a source you trust. Count how many held up.
Whatever the score, you have just practiced the exact habit that lets you use these tools with confidence: enjoy the speed, keep the judgment.