Real news, real insights – for small businesses who want to understand what’s happening and why it matters.

By Vicky Sidler | Published 29 November 2025 at 12:00 GMT+2
There is a moment most business owners experience. A weird pain, a rash that looks suspiciously like a map of South America, or a late night panic about whether a headache is stress or something catastrophic. Instead of calling a doctor, we open the same browser we use for Netflix and ask an AI chatbot whether we are dying.
Recently, I wrote about why you cannot trust AI for legal advice. The short version was simple. AI guesses confidently, does not know what it does not know, and cannot access current information.
Medical advice has the same problems but the stakes are higher because a wrong answer can hurt you far more than the long arm of the law.
According to recent research and medical guidance databases, there is a growing concern that people trust AI responses more than advice from trained doctors. Confidence does not equal accuracy. In medicine, that difference matters.
AI cannot access paid medical databases or current research
Medical guidelines change frequently and AI cannot keep up
AI often invents medical facts and sounds confident doing it
Incorrect advice can delay treatment or cause harm
👉 Need help getting your message right? Download the 5 Minute Marketing Fix.
AI Medical Advice Is Usually Wrong and Sometimes Dangerous
Why Medical Information Changes Constantly:
When AI Gives Yesterday’s Medicine:
AI Hallucinations Are Not Harmless:
Real Consequences for Real People:
Why AI Cannot Replace a Healthcare Professional:
So What Should You Use AI For?
1. Why You Can’t Trust ChatGPT, Perplexity or Other AI For Legal Advice
2. Why AI Business Advice Fails Harder Than You Think
3. AI Therapy Is Dangerous Research Shows Why
4. AI Slop Is Breaking the Internet—Here’s What Small Brands Can Do
5. New MIT Study Links AI Use to Weaker Critical Thinking
1. Can I rely fully on AI tools for research or decision making?
2. Why does AI sometimes give confident but incorrect answers?
3. Should small business owners stop using AI entirely?
4. How do I check whether an AI response is accurate?
5. What types of tasks are safe to automate with AI?
6. Are some AI tools more reliable than others?
7. Is AI becoming more accurate over time?
8. How can I use AI without sounding generic or robotic?
Medicine evolves at an overwhelming pace. One estimate says more than seven thousand medical papers relevant to primary care are published every month. Doctors cannot read them all manually, so they use specialised software like UpToDate which continuously reviews and updates information.
UpToDate has more than seven thousand physician authors and monitors nearly five hundred medical journals. Some updates happen multiple times per day when new research comes out. Other trusted databases include PubMed, EMBASE, and the Cochrane Library. These are not free public websites. Doctors and hospitals pay for them because current medical knowledge saves lives.
AI systems do not have access to these paid sources. They work from older, publicly available information. If medical guidance changed after their training period and no one posted about it publicly, the AI will never know.
AI models have what researchers call a knowledge cutoff. That means their training ended on a certain date. Anything discovered, updated, recalled or approved after that point does not exist in their world.
Even tools that can browse the internet, like Perplexity or certain versions of ChatGPT, still face major limits. They only access what’s publicly available, which means no access to medical databases like UpToDate, PubMed, or Cochrane. They also work within token limits. That means they're not reviewing everything. They're sampling small chunks and trying to summarise them.
Just because a paper is online doesn’t mean AI read the whole thing—or understood it correctly. And it definitely doesn’t mean it picked the best source. It might rank a Reddit thread above a medical journal because the language is more common or the formatting is clearer.
In medicine, that’s a problem. Treatment recommendations shift. Drugs get recalled. Guidelines change fast.
And here's the critical point: AI is not a search engine. It usually doesn't fetch live results. It predicts the next likely sentence based on past patterns. That’s why it sometimes forgets what you told it ten minutes ago. It can’t hold the full conversation, let alone an entire medical database.
So while the AI may sound confident, it might be recommending treatments that were replaced last year. Or it might invent a study that never existed. Neither of those is helpful when your health is on the line.
If the problem stopped at outdated answers, it would already be concerning. Unfortunately AI also does something called hallucinating. That means when it does not know an answer, it creates one that sounds medically plausible.
And there is another layer to this problem. AI scrapes the public internet which means personal blogs, opinions, and half researched theories sometimes get repeated as if they are verified medical facts. The AI cannot tell the difference between a peer reviewed study and a persuasive stranger on a message board with a strong opinion and a username like DetoxWarrior87.
So instead of saying something responsible like “results vary and research is ongoing,” AI will confidently reply as if it is quoting a medical textbook.
Examples include:
Fake medical studies
Incorrect treatment options
Wrong diagnostic criteria
Fabricated drug interactions
Someone’s opinion stated as clinical truth
One study showed AI chatbots diagnosed medical conditions correctly less than half the time. Another found that eighty eight percent of chatbot medical responses contained false information.
For someone already anxious about health, the confident tone makes the response feel trustworthy. That is the danger. It feels right even when it is wrong.
Medical mistakes do not wait months to reveal themselves. Sometimes the impact is instant. One documented case involved a patient who experienced concerning symptoms after a cardiac procedure. Instead of seeking medical help, they asked an AI tool whether it was serious. The AI suggested the symptoms were normal. Doctors later confirmed it was a stroke which required urgent attention.
Another study found people trusted AI health explanations more than ones written by licensed physicians even when the physician labelled the AI’s answer as low accuracy.
Trusting a system because it sounds smart is not the same as receiving professional care.
Doctors do more than interpret facts. They examine patients, observe subtle body language, consider history, culture, lifestyle, and context. They apply judgment formed over years of training and experience.
AI cannot touch a patient. It cannot notice a change in skin tone. It cannot hear a tremor in someone's voice or pick up the unspoken fear behind symptoms. It cannot access the latest medical studies behind paywalls.
Most importantly, doctors are accountable for their decisions. AI is not.
At this point you might be wondering whether AI has any useful place in healthcare or whether it belongs somewhere between fortune telling and horoscope apps. The truth sits somewhere in the middle. AI can be helpful if you treat it like an enthusiastic assistant with no medical responsibility and no authority.
AI can help with:
Explaining medical terminology in plain English
Summarising general context
Helping you prepare questions before seeing a doctor
Those uses are similar to reading a medical dictionary or watching a short educational video. They are informational, not diagnostic.
What AI should never do is tell you what condition you have, whether a medication is safe for you, or whether a symptom is serious enough to get checked. Diagnosis requires human judgment, access to current evidence, and sometimes physical examination.
It should never be used to diagnose, prescribe, or decide whether symptoms require care.
Your health and your life deserve more than a confident guess.
If clarity matters in marketing and business, it matters ten times more in medicine. Guessing may work for dinner recipes. It does not work for medical decisions.
If you want clarity in how you communicate with customers and write messages that help people trust you, my free 5 Minute Marketing Fix can help you start small with one sentence that explains what you do and why it matters.
If AI’s medical advice is made up, its legal guesses shouldn’t surprise you. Read how it confidently invents laws that don’t exist.
Same confident tone. Same lack of judgment. Find out how AI business advice falls apart under real-world pressure.
When health issues are emotional, AI still can’t help. This one covers the real psychological risks of turning to bots for support.
The web is being overrun with confident nonsense. This article explains the consequences and how small businesses can still stand out with real expertise.
If you’re relying heavily on AI to think for you, this research shows what might be happening behind the scenes to your own decision-making ability.
No. AI tools can generate helpful starting points, but they often present guesses or opinions as facts. Always verify the information with credible human sources.
AI responds based on patterns in data, not understanding. If the data is incomplete, biased, or unclear, the output can sound certain while being completely wrong.
Not necessarily. AI can save time and support early brainstorming. It works best when paired with human judgement, expertise, and proper verification.
Look for real citations from trusted organisations, recent sources, or subject-matter experts. If the topic affects money, safety, legal decisions, or strategy, verify manually.
Templates, drafts, outlines, summaries, and simple content ideas are often safe. Anything requiring nuance, compliance, or strategic thinking should be checked by a human.
Different tools are trained on different datasets and have different priorities. Reliability varies, so treat all outputs as a starting point, not the final answer.
Accuracy is improving, but so are the risks. As models grow, hallucinations become harder to spot because the responses sound more believable.
Use AI for structure or rough drafts, then rewrite in your own tone. Real experience and examples make content sound human.
Assuming the tool thinks or cares. AI doesn’t know whether it is right or wrong. It generates text, not truth.
Legal advice, financial decisions, healthcare, compliance, and anything where a mistake could cause harm or cost money. Always involve a qualified human in these cases.
👉 Need help getting your messaging right?Download the 5 Minute Marketing Fix.

Created with clarity (and coffee)