Why AI Makes Things Up: What Medical and Dental Offices Should Know
By Stacey Tallitsch | July 21, 2026
You asked an AI tool a simple question. It gave you a clear, confident answer. Later you found out the answer was wrong.
If that has happened to you, you are not imagining things. AI tools make things up. They do it often, and they do it with a straight face. The people who build these tools even have a name for it. They call it a "hallucination."
That is a strange word for a piece of software. So let me explain what it actually means, why it happens, and what it means for your business. This matters more for a medical or dental office than for most, because in your world a wrong answer can hurt someone.
What "hallucination" actually means
An AI hallucination is when the tool states something false as if it were true. It is not lying. Lying means knowing the truth and hiding it. The tool does not know things the way you do. It is guessing, and sometimes the guess is wrong.
Think of it like a very well-read intern. This intern has read almost everything but remembers none of it exactly. When you ask a question, the intern gives you the most likely-sounding answer based on everything it has seen. Most of the time that answer is right. Sometimes it is confidently, completely wrong.
The hard part is the confidence. A hallucination does not come with a warning label. The wrong answer looks exactly like the right ones. Same tone. Same certainty. That is what makes it a problem.
Why AI makes things up
To understand why, you need one plain fact about how these tools work. An AI language model (the kind of tool behind ChatGPT and similar products) does not look up facts in a filing cabinet. It predicts words. It learned patterns from huge amounts of text, and it uses those patterns to guess what should come next.
Because it is predicting rather than remembering, it will sometimes fill a gap with a plausible guess. If it does not have a solid pattern for your exact question, it does not stop and say "I do not know." It produces the most likely-sounding answer anyway. That guess is the hallucination.
Researchers at the company behind ChatGPT published a paper in 2025 that explains one reason this keeps happening. The way these tools get trained and tested tends to reward confident guessing over admitting uncertainty. A tool that guesses scores better on a test than a tool that says "I am not sure." So the tools learn to guess. That habit follows them into your office.
If you want the plain version of how these tools learn in the first place, I covered what "training data" means and whether AI learns from what you type in an earlier post.
What this looks like in a dental or medical office
Picture a patient calling your dental office after hours. You have set up an AI assistant to answer common questions. The patient asks whether they can take a certain painkiller along with a medication they already take. The AI gives a clear answer. It sounds like it came from a knowledgeable staff member. It is also the kind of question where a wrong answer could send someone to the emergency room.
This is the heart of the issue. The tool is good at sounding informed. It is not a doctor. It does not check its answer against a drug database unless you specifically built it to do that. Left on its own, it predicts an answer. For a health question, a predicted answer is not good enough.
The same risk shows up in quieter ways. An AI tool might state that you accept an insurance plan you dropped last year. It might invent an appointment policy nobody ever gave it. Ask it to back up a claim with a source, and it may hand you a study that does not exist. None of this is the tool trying to deceive you. It is the tool guessing, the same as always.
Why does this land harder on a practice than on, say, a hardware store? Two reasons. First, your patients trust what your office tells them, and they act on it. A wrong answer does not just annoy them. It can change what they put in their body. Second, you carry a level of legal responsibility that most small businesses do not. A confident, wrong answer with your name on it is a real exposure, not just a bad customer moment. That is why the guardrails below are worth the small amount of effort they take.
Five ways to protect your office
You do not have to avoid these tools. You have to use them in the right spots and put a guardrail around the risky ones. Here is how.
Use AI for drafts, not final answers, on anything touching health or money. Let it write a first version of a patient email or a recall reminder. Then have a person read it before it goes out. The tool speeds up the work. The human catches the mistake.
Keep it away from clinical advice. An AI assistant answering "what are your hours" or "where do I park" is fine. An AI assistant answering "should I take this pill" is not. Draw that line clearly and tell your staff where it is.
Feed it your real information. A tool guesses far less when it answers from your own documents instead of its general memory. This setup is called "retrieval," which just means the tool pulls from files you gave it, like your actual policy sheet or price list, before it answers. Ask whoever set up your tool whether it works this way.
Ask for sources and check them. If the tool gives you a fact that matters, ask where it came from. Then confirm the source is real. This one habit catches most of the damaging mistakes.
Pick a tool that admits when it is unsure. Newer tools are better at saying "I do not have that information" instead of making something up. The differences between the main options matter here. I compared ChatGPT, Claude, and Gemini for a small business in another post, and how each one handles uncertainty is part of the difference.
So should you trust AI at all?
Yes, in the right places. The mistake is treating one of these tools like a search engine that always returns facts. It is not that. It is a fast, fluent assistant that is usually right and occasionally very wrong, and it will never tell you which is which.
That is a fine helper for drafting a newsletter, summarizing a long document, or roughing out a reply. It is a poor choice for a question where being wrong has a real cost. The skill is knowing which task is which. If you are still weighing whether your kind of business needs one of these tools at all, I wrote an honest take on whether a salon or restaurant actually needs AI that applies just as well to a practice.
Hallucination is not a flaw someone will patch away next month. It comes from how these tools work at the core. The good news is that you do not need it gone. You need to know it is there, keep the tool out of the seats where a wrong answer hurts, and put a person between the tool and anything that matters. Do that, and AI becomes a useful pair of hands instead of a quiet risk.
-- Stacey | The Standalone
About the Author
Stacey Tallitsch runs The Standalone, an AI Implementation Diagnostic practice for small business owners. He has 30 years of experience in technology and has written 21 books on systems thinking and decision-making. More than 30,000 students have learned from his online courses.
- Stacey Tallitsch, The Standalone