You search Google for "pancreatic cancer symptoms" and the AI tells you to avoid fats. Sounds like reasonable advice, right? You could die if you follow it.
This is exactly what an investigation by The Guardian discovered in January 2026: Google's artificial intelligence summaries, called AI Overviews, are giving medical advice so dangerously incorrect that experts describe it as "alarming" and potentially fatal.
Google had to remove these summaries from several medical searches after doctors and health organizations raised the alarm. But the damage is done: 72% of adults search for health information on Google, and millions have received advice that directly contradicts medical evidence.
What went wrong? How much can you trust "Dr. Google"? And most importantly: what should you use instead?
What Are AI Overviews and Why Google Put Them in Health Searches
AI Overviews are artificial intelligence-generated summaries (using the Gemini model) that appear at the top of Google search results. Instead of showing you a list of links, Google tries to give you "the answer" directly.
Timeline of the Disaster
| Date | Event |
|---|---|
| May 2023 | Google presents AI Overviews at I/O |
| May 2024 | US launch for hundreds of millions |
| Late 2024 | Expansion to over one billion users |
| January 2026 | The Guardian exposes dangerous medical errors |
| January 2026 | Google removes AI Overviews from some health searches |
The original idea seemed good: instead of reading 10 articles about your symptoms, the AI gives you a summary. The problem is that AI cannot distinguish between correct and incorrect information when it comes to medicine.
Why Health Is Especially Dangerous for AI
Medicine isn't like searching for "best Italian restaurant nearby." A health error can:
- Delay a correct diagnosis
- Make you ignore serious symptoms
- Lead you to make decisions that worsen your condition
- In extreme cases, kill you
And according to an Ahrefs study, 44% of medical queries trigger an AI Overview, more than double any other category.
The Errors That Almost Killed Patients
The Guardian's investigation, supported by British medical organizations, documented chilling cases.
Case 1: Pancreatic Cancer - "Avoid Fats"
A search about pancreatic cancer generated an AI Overview advising to avoid high-fat foods.
The problem: This is exactly the opposite advice these patients need.
Anna Jewell, from Pancreatic Cancer UK, explained:
"The advice to avoid high-fat foods is completely incorrect. Following that advice could be really dangerous and compromise someone's chances of being well enough to receive treatment."
Why? Pancreatic cancer patients have difficulty absorbing nutrients. They need high-calorie, high-fat diets to:
- Maintain body weight
- Tolerate chemotherapy
- Be in condition for potentially life-saving surgery
Following Google's advice could literally prevent you from surviving a treatment that could cure you.
Case 2: Pap Test and Vaginal Cancer - 100% False Information
A search for "vaginal cancer symptoms and tests" returned a summary claiming that the Pap test detects vaginal cancer.
This is completely false.
The Pap test (cervical cytology) detects cervical cancer, not vaginal. They are different cancers in different organs with different tests.
Athena Lamnisos, from The Eve Appeal, stated:
"It's not a test to detect cancer, and it's certainly not a test to detect vaginal cancer – this is completely erroneous information. Getting incorrect information like this could potentially lead someone to not get their symptoms checked because they had a clear result on a recent cervical screening."
A woman with vaginal cancer symptoms might think: "My Pap test came back fine 6 months ago, it can't be cancer." And fail to see a doctor until it's too late.
Case 3: Liver Tests - Numbers Without Context
Searches for "normal liver test ranges" produced tables of numbers without considering:
- Patient's age
- Sex
- Ethnicity
- Specific laboratory methodology
Vanessa Hebditch, from British Liver Trust:
"A liver function test is a collection of different blood tests. Understanding the results and what to do next is complex and involves much more than comparing a set of numbers."
The danger: many liver diseases are asymptomatic until advanced stages. A person might see that their numbers "look normal" according to Google and not seek medical attention, when in reality they have developing cirrhosis.
Case 4: Mental Health - Advice That Pushes People Away from Help
For searches about psychosis and eating disorders, AI Overviews offered advice described by experts as "very dangerous."
Stephen Buckley, from Mind:
"Some AI summaries for conditions like psychosis and eating disorders offered very dangerous advice and were incorrect, harmful or could lead people to avoid seeking help."
In mental health, where stigma already makes it difficult for people to seek help, incorrect advice from an "authoritative" source like Google can be the difference between seeking treatment or not.
The Statistics That Should Scare You
The numbers behind this problem are alarming:
"Dr. Google" Usage
| Statistic | Meaning |
|---|---|
| 72% of adults search health on Google | Most people trust Google for health |
| 89% search symptoms before seeing a doctor | Google is the "first doctor" |
| 35% try to self-diagnose online | Avoid doctors completely |
| 50 million health searches in UK (2023) | Massive scale |
AI Errors in Health
| Statistic | Source |
|---|---|
| 44% of medical queries trigger AI Overview | Ahrefs |
| 70% of health AI Overviews rated as "risky" | Medical panel |
| 22% of AI responses have harmful recommendations | Stanford-Harvard |
| 1 in 8 summaries "very or extremely" risky | Nurse study |
The Problem of Blind Trust
| Statistic | Problem |
|---|---|
| 2/3 of users find AI results "trustworthy" | Trust incorrect information |
| MIT participants considered low-accuracy AI responses as "valid and reliable" | Can't detect errors |
| 25% of doctors say patient AI info contradicts medical advice | Conflict in consultations |
Google's Response: Insufficient
After The Guardian's investigation, Google removed AI Overviews for some specific medical searches:
What They Removed
- "what is the normal range for liver blood tests"
- "what is the normal range for liver function tests"
What They Did NOT Remove
Variations like:
- "lft reference range"
- "lft test reference range"
These searches still generate AI summaries. Simply changing how you write the question, you get potentially dangerous information.
Google's Defense
A Google spokesperson stated:
"We invest significantly in the quality of AI Overviews, particularly for topics like health, and the vast majority provide accurate information."
Google also claimed that:
- Their internal team of doctors reviewed the examples
- In many cases, the information "wasn't inaccurate"
- The summaries link to "known and reputable sources"
The problem with this defense: Even if it links to Johns Hopkins, the AI can misinterpret that information or present it out of context.
Medical Organizations' Criticism
Vanessa Hebditch, from British Liver Trust:
"Our biggest concern is that this is a limited approach to one search result. Google could simply disable AI Overviews for that query, but that doesn't solve the bigger problem of using AI Overviews in healthcare."
In other words: Google is playing "whack-a-mole" with specific searches instead of recognizing that AI shouldn't give medical advice.
Why AI Fails in Medicine (And Always Will)
There are fundamental reasons why artificial intelligence cannot replace medical judgment:
1. AI Cannot Examine You
A doctor can:
- Feel a lump
- Listen to your heart
- See your skin color
- Notice your anxiety level
AI only has text. It cannot perform a physical examination.
2. It Doesn't Know Your History
Your doctor knows:
- What medications you take
- Your allergies
- Previous illnesses
- Family history
- Previous test results
AI answers an isolated question without context.
3. "Normal" Ranges Are Not Universal
What's normal for:
- A 25-year-old woman
- A 65-year-old man
- A person of Asian descent
- Someone taking certain medications
...is completely different. AI gives generic ranges that don't apply to you.
4. AI "Hallucinates" With Total Confidence
Language models can invent information that sounds completely credible but is 100% false. And they present it with the same confidence as real information.
5. Medicine Evolves, AI Falls Behind
Medical recommendations change. What was standard 5 years ago may be obsolete today. AI may be trained on outdated information.
What to Use Instead of Google AI for Health
If you need to search for medical information online, use verified sources:
Top-Tier Medical Institutions
| Source | Specialty | URL |
|---|---|---|
| Mayo Clinic | General, high quality | mayoclinic.org |
| Cleveland Clinic | Medical research | my.clevelandclinic.org |
| Johns Hopkins | Global research | hopkinsmedicine.org |
| M.D. Anderson | Cancer | mdanderson.org |
Government Resources (US)
| Source | Description |
|---|---|
| MedlinePlus | NIH, verified information |
| CDC | Diseases and prevention |
| NIH | Medical research |
| DailyMed | FDA drug information |
For UK Audience
| Source | Description |
|---|---|
| NHS | National Health Service |
| NHS Inform | Scotland health information |
When to ALWAYS See a Doctor
Don't search Google if:
- You have intense or sudden pain
- Symptoms that are worsening
- Unexplained bleeding
- Difficulty breathing
- Confusion or disorientation
- Persistent high fever
- Any symptom that seriously worries you
The golden rule: If you're searching symptoms because something worries you, that worry is already reason enough to call a doctor.
The Bigger Problem: Trust in AI
This case reveals something deeper about our relationship with artificial intelligence.
The Authority Bias
When Google gives you an answer, you assume it's correct because:
- Google is a leading technology company
- The answer "comes from advanced AI"
- It's at the top of the page
- It looks professional and well-written
But AI doesn't know what's true. It only knows what "sounds like" medical information.
The Danger of Automation in Health
Stephanie Parker, from Marie Curie:
"People turn to the internet in moments of worry and crisis. If the information they receive is inaccurate or out of context, it can seriously harm their health."
When you search symptoms, you're vulnerable. You're scared. And you trust that Google will give you correct information.
That trust can kill you.
What Google Should Do (But Probably Won't)
Medical organizations have called for concrete actions:
1. Remove AI Overviews from ALL Medical Searches
Not just from specific queries that generate complaints, but from any health-related search.
2. Show Clear Warnings
If they insist on keeping AI Overviews in health, they should include:
- "This information may be incorrect"
- "Consult a medical professional"
- "Ranges vary based on your individual situation"
3. Transparency About Errors
Publish statistics about:
- How many corrections they've made
- What error rate they accept
- How they verify medical information
4. Collaboration with Medical Organizations
Work with hospitals, medical associations, and patient organizations to review content before displaying it.
Conclusion: Don't Trust Your Life to AI
The Guardian's investigation has exposed what many doctors feared: Google's AI is giving dangerous medical advice to millions of people.
A pancreatic cancer patient could die of malnutrition following the "avoid fats" advice. A woman could ignore vaginal cancer symptoms thinking a Pap test protects her. People with liver disease could think they're healthy looking at numbers without context.
And Google, instead of completely removing this dangerous functionality, plays whack-a-mole with specific holes while the fundamental problem persists.
What You Should Do Now
- Don't trust AI Overviews for health - Ever.
- Use verified medical sources - Mayo Clinic, MedlinePlus, your national health system.
- Consult professionals - If something worries you, call the doctor.
- Distrust answers that are "too clear" - Medicine is complex, simple answers are usually wrong.
- Share this information - Your family members also use Google for health.
Artificial intelligence can be useful for many things. But when it comes to your life, you need a human being with medical training, not a language model that "sounds like" it knows medicine.
Your health is too important to leave in the hands of an AI that's wrong 44% of the time.
Have you ever received incorrect health advice from Google? Next time you search symptoms, remember: AI is not your doctor.
