AI chatbots can explain lab results in plain language: what a test measures, what the units mean and why a value might be flagged. What they cannot reliably do is tell you what your result means for you, because that depends on your history, medicines and trends they usually cannot see. Use AI to understand the report, then ask your clinician what it means.
The notification arrives on a Friday evening. Your blood test results are in the patient portal. You open them before anyone at the clinic has looked. Three numbers are marked in red, one with the letter H beside it. The practice is closed until Monday.
A few years ago you would have searched each test name and read five websites that disagreed. Now you can paste the whole report into a chatbot and get a calm, organised explanation in seconds. Millions of people do exactly this.
Here is the argument of this piece. AI is genuinely good at translating a lab report. It is much less good at interpreting one, because interpretation needs context about you, and the chatbot is working with a page of numbers and whatever you remember to tell it. The translation is where AI helps. The interpretation is where your clinician still earns their place.
Why are patients seeing lab results before their doctor?
In the United States, the change is largely regulatory. Rules under the 21st Century Cures Act, which took effect in 2021, restrict “information blocking,” and many health systems responded by releasing test results to patient portals immediately rather than holding them until a clinician had reviewed them. Other countries have moved in similar directions, if unevenly.
Patients, on the whole, like it. A 2023 survey published in JAMA Network Open by Steitz and colleagues, covering 8,139 portal users at four US health systems, found that 96% preferred immediate online access to their results, even if that meant seeing them before their clinician. Only 8% said reviewing results increased their worry. But among people whose results were not normal, that figure rose to 17%, compared with 5% for people with normal results.
That 17% is the group AI is now serving. They have an abnormal-looking result, no clinician yet, and a question at 9pm on a Friday.
Can AI accurately explain lab results?
For individual tests, often yes. For individual patients, less so.
Researchers led by Zhe He at Florida State University tested this in a study published in the Journal of Medical Internet Research in 2024. They took real questions that lay people had posted online about their lab results and compared answers from several language models with answers from other users of a public Q&A site. The strongest model, GPT-4, gave explanations the researchers described as comprehensive, thoughtful, sympathetic and fairly accurate for individual tests. Peer answers were often short, focused on one number and sometimes opinionated or wrong.
But the researchers also listed the model’s weaknesses, and they are the ones that matter. Its answers were not individualised. It was not clear what sources it drew on. And it did not ask clarifying questions. There were also cases where its responses were simply inaccurate.
That combination describes the risk precisely. An explanation of what ALT measures is the same for everyone. Whether your ALT of 58 matters depends on whether you started a new medicine last month, ran a half-marathon yesterday, or have had a similar value for ten years.
What context does AI miss when reading lab results?
A clinician reading your results is usually looking at more than the page in front of you. They are thinking about several things a chatbot rarely has.
Your trend. A haemoglobin slightly below range that has been stable for five years is a different story from one that has fallen steadily over six months. Portals often show one result at a time, and people paste one report.
Your medicines and supplements. Many common drugs shift lab values. Some supplements, such as high-dose biotin, can interfere with certain lab tests. If you do not mention them, the chatbot cannot account for them.
Why the test was ordered. A routine check and a test ordered to follow up a specific concern are read differently. The same borderline value can be reassuring in one case and a reason for further tests in the other.
The lab’s own reference range. Reference ranges vary between laboratories, methods and sometimes by age, sex and pregnancy. A chatbot applying a textbook range to your lab’s number can mislabel a normal result, or miss an abnormal one.
How you are actually feeling. A mildly raised inflammatory marker in someone who feels well means something different from the same value in someone with a fever and weight loss.
None of this is exotic. It is the ordinary work of interpretation, and it is mostly invisible on the report you are holding.
A worked example: why “flagged” does not mean “wrong”
This example is illustrative, and the arithmetic is the point.
Most reference ranges are built to include the middle 95% of results from a healthy reference population. By definition, that means about 1 in 20 healthy people will fall outside the range on any single test, just by normal variation.
Now imagine a routine panel of 20 separate measurements. If the results behaved independently, the chance that a perfectly healthy person has at least one value outside its range is 1 minus 0.95 to the power of 20. That comes to roughly 64%. In real life, lab values are correlated, so the exact figure differs, but the lesson holds: a healthy person getting one or two flagged values on a big panel is common.
Suppose Ana, 45, pastes a 20-item panel into a chatbot. Two items are slightly high. The chatbot, doing its job, explains each: what the test measures and what conditions can cause high values. The list for one of them includes liver disease. Ana spends the weekend worried.
On Monday her doctor looks at her previous results, sees that the same value has been marginally high for years, notes that she recently started a medicine known to nudge it, and suggests repeating it in three months. Nothing the chatbot said was wrong. It simply answered “what can cause this?” when the real question was “is this a change, for me?”
This is also why a flagged result without a plan can do more harm than good, a theme we explored in a test without a treatment is just bad news.
How to use AI to understand your lab results safely
If you choose to use a chatbot, these habits make it more useful and less alarming. They do not turn it into a substitute for your clinician.
Ask it to explain, not to diagnose. Good questions include “What does this test measure?”, “What do these units mean?” and “What questions should I ask my doctor about this result?” These play to its strengths.
Give it the lab’s reference range, not just the number. The range printed on your report is the one that applies to your sample.
Mention context you know is relevant. Recent illness, medicines, supplements, pregnancy, intense exercise or fasting status can all matter. Say so.
Ask what would make a result urgent. Some values, such as a very high potassium, can need prompt attention. Labs usually contact the clinician directly about critical results, but if a chatbot or your own reading suggests something may be urgent, contact your clinic, an out-of-hours service or emergency care rather than waiting.
Protect your data. Consumer chatbots are generally not bound by the health privacy rules that apply to your hospital or clinic. Remove your name, date of birth and record numbers before pasting a report. We have written about how health apps often sit outside HIPAA, and the same caution applies here.
Bring the conversation to your appointment. Tell your clinician what you read and what worried you. It usually leads to a better conversation, not a worse one.
The strongest counterargument: isn’t any explanation better than none?
There is a serious case that AI explanations are a net gain, and it deserves a hearing.
Lab reports are written for clinicians. Abbreviations, units and flags are close to meaningless for most patients. Waiting days for someone to explain them is its own kind of harm, especially for anxious people. The He study found model answers were often better than what people would otherwise find from peers online. For patients with low health literacy, a plain-language explanation may be the first time a result has made sense. We have argued that hospitals should measure their own health literacy, not blame patients for theirs, and AI translation could be part of that.
All of that is true. But the argument supports AI as a translator, not as an interpreter. The danger is the slide from “this test measures liver enzymes” to “you may have liver disease,” which is the step the chatbot is least equipped to take. The best use keeps the two apart.
What this means for patients, clinicians and health systems
For patients. Use AI to make the report readable. Do not use it to decide whether you are ill, whether to change a medicine or whether to skip follow-up. If anything worries you, contact your clinician, and for severe symptoms seek urgent care regardless of what your results say.
For clinicians. Assume patients will read results before you do and may run them through AI. A short plain-language note attached to results, even one line such as “stable and expected,” can prevent a weekend of worry that no chatbot can resolve.
For health systems. If you release results immediately, consider building explanations into the portal itself, using approved content and your own lab’s reference ranges, with clear routes to ask a person. That keeps the translation in a place where the context actually lives.
Frequently asked questions
Can I trust AI to explain my blood test results?
You can use it to understand what each test measures, what the units mean and why a value might be flagged. You should not rely on it to judge whether a result is significant for you, because that depends on your history, trends, medicines and symptoms. A 2024 JMIR study found model answers were often accurate but not individualised. Discuss results with your clinician.
Why are some of my results flagged if I feel fine?
Most reference ranges include the middle 95% of healthy people, so about one in twenty healthy results falls outside range on any single test. On a large panel, one or two flagged values in a healthy person is common. Your clinician can tell you whether a flagged value is new, changing or expected for you.
Is it safe to upload my lab report to a chatbot?
Consumer chatbots usually are not covered by the same health privacy laws as hospitals and clinics. If you use one, remove your name, date of birth, address and record numbers first, and check the service’s privacy settings. Consider whether you need to share the whole report or only the specific values you want explained.
Should I wait for my doctor before reading my results?
That is a personal choice. Most patients in a large 2023 survey preferred immediate access, though people with abnormal results were more likely to feel worried. If you read results early and something concerns you, contact your clinic rather than waiting in uncertainty, and seek urgent care if you have severe symptoms.
What questions should I ask my doctor about my lab results?
Useful questions include: Is this result new or has it been like this before? Could any of my medicines or supplements explain it? Does it need repeating, and when? Does it change my treatment? What symptoms should prompt me to contact you sooner? These focus the conversation on what the result means for you specifically.
AI has made lab reports readable. It has not made them personal. The number is on the page; its meaning is in your history, and that still lives with the people who know it.
Have you used AI to decode a test result? Did it calm you down or make things worse?
