To understand lab results with AI, black out your name, date of birth, medical record number, Social Security number, and address. Paste only the result page you have. Ask what the words mean, what the report actually says, and what it does not say. Then ask for a short list of questions for your doctor or pharmacist. The chat does not diagnose, choose a treatment, or replace emergency care.
Next step: do Step 1. Copy the box, fill in the blanks, and paste it into the chat.
A prompt is the message you paste. A Director check is a short look before you use the result.
Who is this for?
This page is for someone staring at a lab report, a test result, or a portal screenshot, and wanting the words in plain English before they talk to a doctor, a nurse, or a pharmacist. You bring the page you already have. The chat reads only that page. You leave with questions, not a diagnosis.
This is not medical advice. This is not a diagnosis. This is not a treatment plan. A chat does not know your symptoms, your medicines, your history, or what the person who ordered the test was looking for. It must not tell you what disease you have, what drug to start or stop, or that you can skip the visit.
MedlinePlus says a lab test checks a sample to learn about your health, and that the results do not provide a complete picture on their own. Your provider uses the results along with an exam, your history, and any imaging. MedlinePlus also says to talk with your provider about what the results mean for you, even when you can already read them in a portal.
The Canadian Medical Association says people use a chat to turn jargon into plain language, and that this can help with an unfamiliar term or with questions for the next appointment. The same page says a chat can sound sure when it is wrong, that it misses the context a clinician has, and that you should never treat the reply as the final word. NPR reported the same kind of caution: physicians and patient advocates warn that chatbots can produce wrong answers, and that medical details may not stay private once they are pasted.
A confident paragraph is not a checked line. The habit of quoting the page you pasted is Catch facts the AI made up. When the only allowed source is that page, use Answer only from your sources. A number on the report is a number you paste, on Give it the numbers. The chat does not recall a "normal" from memory and swap it in.
A bill for the same visit is a different document, on Understand a medical bill with AI. Choosing a plan for next year is Compare health plans with AI. Medicare Open Enrollment is Compare Medicare plans with AI. Those pages do not read a lab report. This page does not read a bill.
What do you need before you start?
Gather one result before you open a chat. One page is enough. A whole chart is too much, and it is more of you than the chat needs.
- The result page you have: a printout, a PDF, or a screenshot of that result. Not the inbox full of old messages
- A pen, or a copy. You will black out identifiers before anything goes into a chat
- The name of the test as printed, if you can see it after you cover your name
- The reference range printed on that same page, if the lab included one. MedlinePlus says ranges differ by lab, and you use the range on your report. You do not swap in a range from a website
- A question you already have, in your words, even if it is "what does this word mean?"
- The phone number or portal message path your clinic already gave you, if you have it. Do not ask the chat to invent a clinic phone number
You do not need a paid plan. You need the page. A chat that "knows labs" and has not seen your result is guessing.
When should you not use AI for this?
Close the chat and get care from a person if any of these are true.
- You think you are having an emergency. In the United States, call 911. Chest pain, trouble breathing, a severe allergic reaction, heavy bleeding, and signs of a stroke are reasons to get emergency care. Do not wait to see what a chat says.
- A clinician told you to go to urgent care, the emergency department, or labor and delivery. Follow that instruction. This page does not overrule it.
- You want the chat to tell you which medicine to stop, which dose to change, or which supplement will "fix" the number. Ask the prescriber or the pharmacist.
- The only way to proceed is to paste the unredacted portal, the insurance card, or a Social Security number. Read the page yourself, or ask the clinic to explain it.
- You are about to turn the chat's paragraph in for school credit. Write the assignment yourself. Follow your teacher's rules.
How do you black out the identifiers?
The chat needs the test. It does not need to know who you are. NPR's report says to remove personal information before you paste. Do that on a copy.
Cover these, every time:
- Your name, and any other patient's name
- Date of birth
- Medical record number, account number, and member ID
- Social Security number
- Street address, phone number, and email address
- A barcode or a photo of the portal header that still shows those identifiers
Keep the test name, the result, the units, the reference range printed on that page, the flag the lab printed, the specimen type if it is on the page, and the date of the test. If a comment from the lab is on that same page, you may keep the comment after you are sure it does not include your name.
Do this on a personal account you control. A workplace Copilot chat can be visible to your employer. Your lab result is not a work file. If the only chat you have is a work or school account, do not paste the result there.
Keep your data private with AI is the wider list. A training switch does not make a lab report safe to paste with your name still on it.
How do you paste one result page and ask what it says?
Paste the redacted page, or upload the redacted PDF if the tool you are using accepts a file. Ask for a translation of what is written. A translation that adds a diagnosis is not a translation.
Grok's own docs say the apps can take a PDF and other files. ChatGPT, Claude, Gemini, and Copilot can take a paste. The rule is the same in all of them. The page you pasted is the source. The model's memory is not.
You are helping me read one lab or test result. This is not medical advice. Do not diagnose me. Do not name a disease I have. Do not tell me to start, stop, or change a medicine, a supplement, or a treatment. Do not tell me I am fine. Do not tell me to skip a clinician.
I removed my name, date of birth, medical record number, Social Security number, and address. Use only the page below. If a sentence is not on that page, write NOT STATED.
For each result on the page:
- Quote the test name, the result, the units, and the reference range if the page shows one
- Say that line in plain English
- Say whether the page itself flags the line, such as high, low, or abnormal. If the page does not flag it, write NOT STATED
- Define the term in everyday words. Say if you are unsure
Then stop. Do not add a diagnosis. Do not suggest a treatment.
The page:
[paste the redacted result, or say you uploaded the redacted file]
Read the reply next to the page, not on its own. If a number in the reply is not on your paper, delete that sentence. If the chat names a disease, ignore that sentence and send the prompt again. Catch facts the AI made up.
MedlinePlus says you cannot compare your result to a reference range from somewhere else, because labs use different methods and different ranges. If the chat quotes a range that is not on your page, that range is not yours. Mark it and drop it.
How do you mark what the report does not say?
The useful part of a careful read is the gap. A lab page often does not say why the test was ordered, what your symptoms were, what medicines you take, or what the last result was. The chat should say those things are not on the page. It should not fill them in.
Stay with the same page. Do not diagnose. Do not recommend a treatment.
List what this report does not say. Use these labels only when they are true:
- NOT ON THE PAGE: symptoms, medicines, history, or a prior result I did not paste
- ON THE PAGE: a sentence you can quote
Answer only these:
1. What words would a careful reader still not understand?
2. Which lines are flagged by the lab, quoted from the page?
3. Which lines have no flag on the page?
4. What would you need from me, or from my clinician, before anyone could interpret this for my life?
Do not guess my age, my sex, my diagnosis, or my medicines.
If I did not paste it, it is NOT STATED.
Then stop.
The Canadian Medical Association uses a plain example: a chat may be able to define a term such as blood urea nitrogen, and it still should not put that number in the context of your whole health or tell you what to do next. It does not know your age, your history, or your medicines unless you typed them, and you should not type a life story to "help it." The questions in the next step are how you take the gap to a person who can see you.
How do you write questions for your doctor or pharmacist?
Leave with a short list you can ask. Five questions is enough. A question is not a request for the chat to answer as if it were the clinician.
Draft questions I can ask my doctor, nurse, or pharmacist about this one result.
Questions only. Do not answer them. Do not diagnose. Do not choose a treatment.
Include:
- One question that asks what this result means for me, in their words
- One question about any line the lab flagged, using the test name from the page
- One question about medicines or supplements I already take, without you naming a change
- One question about whether I need a repeat test, a different test, or nothing further
- One question I should ask if I do not understand the answer the first time
Use the test names from my page. If a detail was NOT STATED, the question can ask the clinician. It must not pretend the page said it.
Keep the list to 5 questions. Then stop.
Take the list to the appointment, the portal message, or the pharmacy counter. You ask. They answer. If the chat's earlier paragraph and the clinician disagree, the clinician who can see you is the one who decides. Tell them you used a chat to prepare questions if you want to. You do not have to hide it. You also do not have to treat the chat as a second opinion.
A pharmacist is the right person for a question about a medicine you already take. The chat does not replace that counter.
What mistakes should you avoid?
These are the misses that waste the result. Skip them before you send anything.
Common mistakes
- Pasting the full portal, the insurance card, or a photo that still shows your name.
- Asking "what disease is this?" and keeping the answer.
- Letting the chat replace the reference range on your page with a range from its memory.
- Treating "you are fine" or "this is urgent" as a triage decision. Emergency symptoms go to emergency care. Everything else goes to your clinician.
- Sending the chat's paragraph to a family group as if it were the doctor's note.
- Using a work account because it was already open.
What should you keep private?
A redacted result can still feel sensitive. Use a personal account. Turn off model training if the product has that switch, and still assume a pasted page is not a secret you can get back. Delete the chat when you are done if the product lets you.
Privacy: what not to paste
Do not paste a family member's result unless you are their caregiver and you are allowed to handle it. Black out their identifiers too. A child's result is not a school project.
Grok's consumer FAQ says not to share personal or sensitive information in questions, and it tells you where the training control sits. The same warning belongs on ChatGPT, Claude, Gemini, and Copilot. The privacy guide walks the switches those products actually show.
What should you check before you use it?
Answer each line before you act on the reply.
Director check: look before you use it
- Are your name, date of birth, medical record number, Social Security number, and address still out of the chat? If one slipped in, delete that chat and do not paste it again.
- Can you point to the page for every number in the reply? If the chat added a range or a result, drop that sentence.
- Did it diagnose you or tell you to change a medicine? Ignore that part. Ask your clinician or pharmacist.
- Did it say NOT STATED where the page is silent? If it filled the silence with a story, delete the story.
- Do you have five questions or fewer, in words you will actually ask? If the list is a speech, cut it.
- Are you treating this as medical advice? It is not. You still talk to a clinician. Emergency symptoms are emergency care.
What is one extra tip?
When you want every sentence tied to a quote, use Answer only from your sources. The same move, said for a lease, a bill, or a policy, is Ask for plain English. A lease is Read a lease with AI. That page is not legal advice. When a second page disagrees with the first, the comparison habit is Compare two documents with AI. You still do not ask either page to diagnose you.
One extra tip
The charge for the test, if a bill arrives later, is Understand a medical bill with AI. If the insurer denies that claim, the appeal is Appeal a denied insurance claim with AI. You file it. The chat does not invent the deadline. The plan that might cover the next test is Compare health plans with AI or, for Medicare, Compare Medicare plans with AI. Those guides are not insurance advice. This guide is not medical advice.
The same prompts work in a free chat the guide names, including Grok when you want a file upload, and ChatGPT, Claude, Gemini, or Copilot. More health pages are on Health plans.
Which sources back this page?
Open these. A summary on this site is not a substitute.
What else do people ask?
Open a question. The short answer under it is the one to use.
Can AI read my lab results?
It can translate words on a page you paste, after you remove your name and ID numbers. It can list questions for your doctor or pharmacist. It cannot diagnose you, choose a treatment, or replace the clinician who ordered the test.
What should I remove before I paste a lab report?
Your name, date of birth, medical record number, Social Security number, address, phone number, and email. Also black out a member ID and a barcode that shows those identifiers. Keep the test name, the result, the units, the reference range printed on that page, and the flag the lab printed, such as high or low.
What if a number is outside the range on the page?
Ask the chat what the report says, and ask what the report does not say. A value outside a reference range is a reason to talk to your clinician. MedlinePlus says a result outside the range may or may not be a health problem, and a result inside the range is not a guarantee. The chat must not decide which one you are.
Can I use this instead of calling the doctor?
No. The Canadian Medical Association says a chat can help with a term, plain language, or questions for an appointment, and that the doctor is the one who diagnoses and decides treatment. If you were told to get emergency care, or you think you are having an emergency, use emergency services. Do not wait for a chat.
Is a lab report the same as a medical bill?
No. A lab page is about a test. A bill is about charges. Read a bill on Understand a medical bill with AI, after you black out names and account numbers. This page does not explain what you owe.
Can a student use this?
Only to learn how a lab page is worded. Use a sample that is already public, or a result that is yours and that your teacher said you may use. You explain it in your own words. You do not turn in a diagnosis the chat wrote. Follow your teacher's rules.
By Joe Rangel · Last updated · More Life guides