How to Use Free AI to Read Medical Bills and Insurance Letters Without Oversharing

By Austin Little

A medical bill is a document written by one institution for another institution, and you happen to be the person who has to pay it. Free AI can help you read it. The trick is to give the chatbot the structure of the problem without handing it your identity, and to remember that it explains, it doesn't decide.

AI disclosure. This page was drafted with AI assistance and edited for Apiary. We don't invent quotes, stats, people, or events.

This is not medical, legal, or insurance advice. Nothing here tells you what you owe, whether a claim should have been paid, or what treatment you need. It's a method for understanding paperwork well enough to ask the right people the right questions. When money or care is on the line, the final word belongs to your insurer, your provider's billing office, the official government help lines named below, or a professional you trust.

Why this is worth doing at all

Most people who open a hospital bill or an "Explanation of Benefits" letter feel two things at once: confusion and a low hum of dread. The paperwork uses abbreviations, codes, and phrases like "allowed amount," "patient responsibility," and "not a bill" on something that looks exactly like a bill. That confusion isn't a personal failing. These documents are built for billing departments and claims systems, not for the person at the kitchen table.

A free AI chat tool is surprisingly good at one narrow job here: turning jargon into plain language. Ask it what "coinsurance" means, why a statement shows three different dollar figures for the same visit, or what a denial reason phrase usually refers to, and you'll typically get a clear, patient explanation. That's real value. It can turn a forty-minute phone call into a ten-minute phone call, because you show up knowing which line you're asking about.

But there are two ways this goes wrong. The first is oversharing: pasting a full bill, with your name, date of birth, member ID, account number, and diagnosis, into a service you don't control. The second is over-trusting: treating the chatbot's confident answer as if it were a ruling. Both are avoidable. The rest of this guide is about avoiding them.

What free AI is good at, and what it isn't

Good jobs for a chatbot

Defining terms. Deductible, copay, coinsurance, out-of-network, allowed amount, adjustment, write-off, balance billing, prior authorization. These are general concepts, and a chatbot can explain them without knowing anything about you.
Explaining the layout of a document. "My statement has columns labeled Charges, Adjustments, Insurance Paid, and Patient Balance. How do those usually relate?" That's a structure question, and it doesn't need your personal details.
Translating a denial reason into plain English. Insurance letters often include a short reason phrase. You can paste just that phrase.
Drafting questions. "Help me write five polite, specific questions to ask a hospital billing office about a charge I don't recognize." This is one of the most useful things AI does.
Organizing a timeline. If you've had several letters and calls, AI can help you turn messy notes into a clean list of dates and what happened, which is gold if you end up appealing.
Checking arithmetic. If you type in the numbers (not the document), it can help you see whether charges minus adjustments minus insurance payments equals the balance shown. Still double-check with a calculator. Chatbots do slip on math.

Bad jobs for a chatbot

Deciding what you owe. Only your insurer's processing of the claim and your plan documents determine that. AI doesn't have your plan contract, and even if you pasted it in, it isn't the party that decides.
Telling you a charge is "illegal" or "fraudulent." It might say so confidently. It may be wrong. Billing errors are common enough that it's worth asking about anything odd, but ask the billing office or the official help desk, not the chatbot, whether something breaks a rule.
Interpreting a diagnosis or treatment. If a bill makes you realize you don't understand what was done to you medically, that's a conversation for your clinician. A billing code describing a procedure is not medical guidance.
Quoting the law. AI tools can mix up federal rules, state rules, and plan-specific rules, and they can be out of date. Any patient-rights claim should be checked against official pages. We've listed the ones we verified today below.
Knowing whether your provider is in-network. That's specific to your plan and changes. Call the number on your insurance card or use the insurer's own directory.

A good mental model: the AI is a well-read friend who has never seen your plan and isn't allowed to make phone calls. Useful for understanding. Useless as a final authority.

The redaction habit: strip it before you paste it

Here's the core rule of this article. Never paste or upload a medical bill or insurance letter as-is. Before any text goes into a chatbot, take out anything that identifies you, your account, or your health condition, unless the specific question genuinely needs it (most don't).

What to remove every time

Make this a checklist. Tape it next to wherever you deal with mail.

Your full name and the names of family members on the plan.
Date of birth.
Street address, email, phone number.
Member ID, group number, subscriber ID, anything printed on your insurance card.
Account numbers, claim numbers, invoice numbers, patient ID numbers. These are keys to your records. Write them down for your phone call; don't give them to a chatbot.
Social Security number if it appears anywhere. It should rarely appear on modern paperwork, but if it does, it never goes into AI. Ever.
Provider names and the facility name, if you don't need them for the question. "A hospital" is usually enough.
Exact dates of service, unless timing is the actual question. "Earlier this year" or "Visit 1, Visit 2" works fine.
Diagnosis descriptions and diagnosis codes, unless your question is specifically about what a code category means. Even then, consider asking about the code format in general rather than pasting your specific code.
Barcodes, QR codes, and payment portal links. They can carry account information, and some scam letters use them too.

What you can usually leave in

Generic line-item descriptions ("office visit," "lab services," "facility fee"), or replace them with placeholders.
Dollar amounts, if you need help with arithmetic. Dollar amounts alone rarely identify anyone, but you can round them or swap them for "Amount A, Amount B" if you'd rather.
Column headings and the document's general layout.
The exact wording of a denial reason or a phrase you don't understand.

Retype, don't upload

The safest technique is the most boring one: retype the parts you need into a fresh note, then paste that note. Don't photograph the letter and upload the picture. Photos and PDFs carry everything on the page, including the parts you'd never think to type, like the tiny account number in the corner or the barcode on the return slip. Retyping forces you to choose what goes in.

If you must work from a scanned document, cover sensitive parts with paper before you photograph it, or crop aggressively.

A redaction example

Say your statement contains this (all invented for illustration):

Patient: [your name] · DOB [date] · Acct [number] Svc date [date] · Provider: [hospital name] Emergency dept visit, level [x] — Charges $[A] · Adjustment $[B] · Ins paid $[C] · Pt resp $[D] Remark: "Out-of-network provider. Member responsibility applies."

What you'd paste:

I got a hospital statement for an emergency room visit. One line shows Charges, Adjustment, Insurance Paid, and Patient Responsibility. A remark says: "Out-of-network provider. Member responsibility applies." In plain English, what does that remark usually mean, and what questions should I ask my insurer and the hospital billing office about it?

Notice what the AI can still do with that: explain the remark, explain the columns, and help you prepare questions. It has nothing that could identify you.

Choosing a free tool with your privacy in mind

You don't need to pay for anything to do this well. A few options, in rough order of privacy:

Option 1: A local model on your own computer

If you or someone in your family is comfortable installing software, a local model keeps text on your machine. Ollama's official quickstart says you can download it for macOS, Windows, or Linux and run models locally without an API key.

Local doesn't mean you can skip redaction. Computers get shared, backed up, and synced. But it does mean you aren't sending a bill to someone else's server.

A local model will usually be less polished than the big cloud chatbots. For defining terms and drafting questions, that's often fine.

Option 2: A free tier of a mainstream chatbot

Several major chat assistants offer free plans. For example, Anthropic's pricing page lists a Free plan at $0 for chatting on web, desktop, and mobile, with usage limits. It also lists an "Incognito chats" feature for free users, and shows "Model training: Opt-out" on consumer plans, which means you have to turn training off yourself if you don't want your chats used.

Before using any cloud chatbot for anything health-related:

Find the data or privacy settings and decide whether you want your conversations used to train models. Turn it off if you can and you'd rather it be off.
Look for a temporary, incognito, or no-history mode and use it for this kind of task, if your tool offers one.
Remember that consumer chatbots are generally not your health care provider. HIPAA rules apply to health plans and providers, and a general chatbot is usually not one of those. Anthropic's own pricing table, for example, shows its "HIPAA-ready offering" only on enterprise plans, not on Free.

Option 3: No AI at all, just the official help lines

Sometimes the right tool is a phone. For questions about the No Surprises Act, CMS runs a help desk; its consumer page lists 1-800-985-3059 and says it offers help in many languages. Your insurer's member services number is on your card. The hospital billing office number is on the statement. AI helps you prepare for those calls. It doesn't replace them.

A step-by-step routine for one confusing letter

Here's a routine you can repeat for any bill or insurance letter. It takes about fifteen minutes the first time and less after that.

Step 1: Identify what kind of document it is

Before AI, just look at the top of the page. Is it:

A bill or statement from a provider (doctor's office, hospital, lab, ambulance)? It usually asks for payment and has a due date.
An Explanation of Benefits (EOB) from your insurer?
A denial or adverse determination letter? It says the insurer won't pay for something, or won't pay as much, and usually explains how to appeal.
A good faith estimate? It's a pre-care estimate, most relevant if you're uninsured or not using insurance.
A collections notice? It comes from a third party collecting a debt.

If you can't tell, you can ask AI: "What's the difference between a provider statement, an Explanation of Benefits, and a denial letter?" No personal details needed.

Step 2: Write down the identifiers for yourself, offline

On paper or in a private note, write the claim number, account number, dates of service, and provider name. You'll need these for phone calls. They don't go into the chatbot.

Step 3: Retype the confusing parts with placeholders

Use "Provider A," "Visit 1," "Amount A." Keep exact wording only for the phrases you don't understand.

Step 4: Ask for an explanation, not a verdict

Phrase your questions so the AI explains rather than rules. Compare:

❌ "Do I have to pay this?"
✅ "What does 'patient responsibility' usually mean on a hospital statement, and what could cause it to be higher than I expected?"
❌ "Is this bill illegal?"
✅ "What are common reasons a bill for emergency care might show out-of-network charges, and what official resources explain consumer protections for that situation?"

The second version of each gets you something useful without inviting the chatbot to pretend it's a judge.

Step 5: Ask for questions to take to a human

This is the most valuable prompt in the whole routine:

Based on this explanation, write me 5 short, polite questions to ask my insurance company's member services line, and 5 to ask the provider's billing office. Keep each question to one sentence.

You'll get a script. Edit it so it sounds like you.

Step 6: Check any rights or rules against official pages

If the AI mentions a law, a deadline, or a right, don't act on it until you've checked an official source. We list the ones we verified below. If the AI cites a specific rule we haven't listed, look it up on CMS.gov, HHS.gov, HealthCare.gov, or your state's insurance department website. If you can't find it there, treat it as unconfirmed.

Step 7: Make the call and take notes

When you call, write down the date, time, name of the person you spoke with, and what they said. Ask for a reference number for the call. Afterward, you can paste your (redacted) notes into AI and ask it to clean them up into a tidy timeline.

Step 8: Keep the paper trail

Put the letter, your notes, and any reply into one folder, paper or digital. If you end up appealing or disputing, that folder is your case.

Patient rights we checked today (and how to read them)

We fetched these official pages on October 3, 2026. Laws and pages change, so check them yourself before relying on any detail. We're quoting or closely paraphrasing them, not interpreting how they apply to you.

The No Surprises Act (CMS)

CMS's consumer page describes the No Surprises Act as a federal law that went into effect on January 1, 2022. It says the law applies to most types of health insurance and protects people from unexpected out-of-network bills from:

Emergency room visits
Non-emergency care related to a visit to an in-network hospital, hospital outpatient department, or ambulatory surgical center
Air ambulance services

The same page says you can submit a complaint if you believe your facility, provider, or insurer isn't following these rules, and lists the help desk number 1-800-985-3059.

Whether a particular bill falls under these protections depends on details only the insurer, provider, or CMS help desk can sort out. AI can help you understand the categories. It can't tell you which one your bill falls into.

Good faith estimates if you're uninsured or self-paying (CMS)

CMS says that, usually, if you don't have or aren't using health insurance, providers must give you a good faith estimate of expected charges if you request one or schedule care at least 3 business days in advance. CMS's consumer page also says you may be able to dispute your bill if it's at least $400 more than the estimate. Its good faith estimate guide adds that you'll need the estimate to dispute a bill, so keep it.

That guide also lists timing rules. If you schedule 3 to 9 business days ahead, you should get the estimate within 1 business day. If you schedule 10 or more business days ahead, within 3 business days. If you schedule 0 to 2 business days ahead, you aren't entitled to one. Check the page for the full details and any updates.

Your right to see your billing records (HHS)

HHS's HIPAA guidance for individuals says the Privacy Rule gives you, with few exceptions, the right to inspect, review, and get a copy of your medical records and billing records held by covered health plans and providers. It also says:

A provider can't deny you a copy because you haven't paid for services.
A provider may charge reasonable costs for copying and mailing, but can't charge a fee for searching for or retrieving your records.
If you think something in your medical or billing record is wrong, you can request an amendment. If the provider or plan disagrees, you can submit a statement of disagreement that must be added to your record.

The HHS page also notes that parts of its guidance are limited by a 2020 court decision, so read the page itself before relying on fee details.

Why this matters for AI: an itemized bill or your billing record is often easier for both you and a chatbot to understand than a summary statement. Ask the billing office for an itemized version, then redact it before you paste anything.

Appealing an insurance decision (HealthCare.gov)

HealthCare.gov says that if your insurer refuses to pay a claim or ends your coverage, you have the right to appeal, and that insurers must tell you why they denied the claim and how you can dispute it. It describes two paths:

Internal appeal: asking the insurer for a full and fair review of its own decision, which must be sped up if the case is urgent.
External review: taking the appeal to an independent third party, so the insurer no longer gets the final say.

Appeal deadlines and procedures depend on your plan type and situation. Your denial letter should explain them.

Medicare, Medicaid, and state rules

If you have Medicare, Medicaid, or a state-regulated plan, other rules and help lines may apply. We didn't verify those pages today, so we aren't summarizing them.

Prompts you can copy

These are written so they work without personal information. Swap in your own redacted text where you see brackets.

Terms

Explain these insurance terms in plain English, one or two sentences each, with a simple example using made-up numbers: deductible, copay, coinsurance, out-of-pocket maximum, allowed amount.

Statement layout

My medical statement has these column headings: [list headings]. Explain what each column usually means and how they typically add up to the balance due. Don't assume anything about my specific plan.

Denial phrase

An insurance letter gave this reason for not paying: "[exact phrase]." What does this phrase typically mean? What documents or information would usually help address it? Please don't tell me whether the denial is correct. I'll ask my insurer.

Question script

I'm calling [my insurer / a hospital billing office] about a charge I don't understand. Here's what I know: [redacted summary]. Write 6 short, polite questions I can ask, and a one-line opening I can say when the call starts.

Itemized bill request

Write a short, polite letter asking a provider's billing office for an itemized bill for services on [Visit 1]. Leave blanks for my name, account number, and address so I can fill them in by hand.

Timeline cleanup

Here are my messy notes from three phone calls about a bill: [redacted notes]. Turn them into a clean, dated timeline with who I spoke to (use "Rep 1, Rep 2"), what they said, and what they promised to do next.

Arithmetic check

Here are four numbers from one line of a bill: Charges [A], Adjustments [B], Insurance paid [C], Patient balance [D]. Does A minus B minus C equal D? Show the math step by step.

Sanity check on the AI itself

In your last answer, which statements are general information and which ones depend on my specific plan or state? List anything I should verify with my insurer or an official government website.

That last prompt is worth using every time. It pushes the chatbot to separate what it knows in general from what it's guessing.

When AI says something alarming

Sometimes a chatbot will say things like "this charge appears to be illegal" or "you don't have to pay this." Here's how to handle that:

Don't stop paying or ignore a bill because a chatbot said so. Unpaid bills can go to collections.
Ask the AI where that claim comes from. If it can't point to an official source, treat the claim as a lead, not a fact.
Look it up yourself on CMS.gov, HHS.gov, HealthCare.gov, or your state insurance regulator.
Call the relevant help line. For surprise-billing concerns, CMS lists 1-800-985-3059.
If a lot of money is involved or you're in collections, consider a patient advocate, a legal aid office, or an attorney.

Watch for scams that look like medical bills

Fake medical bills and fake insurance calls are a known scam pattern. AI can help you spot the red flags, but the habit that protects you is simple: never pay or give information through a link or phone number in a message you weren't expecting. Instead:

Call the number on your insurance card, or the number on the provider's official website.
Log in to your insurer's or provider's portal by typing the address yourself, not by clicking a link.
Be suspicious of urgency ("pay today or be sent to collections"), unusual payment methods (gift cards, wire transfers, crypto), and requests for your full Social Security number.

You can paste a redacted version of a suspicious message into AI and ask, "What signs in this message are common in scams?" That's a fine use. Just don't let a chatbot's "this looks legitimate" stand in for calling a number you know is real.

Helping a parent or relative

Plenty of people reading this are doing it for someone else: an aging parent, a spouse in recovery, a relative who doesn't read English comfortably. A few dignity-first habits:

Ask before you read their mail. Health paperwork is personal. Offer help; don't take over.
Do the redaction together. Showing someone how to cover their name and member ID before using AI is a skill they keep.
Let them ask the questions. You can draft the script with AI; they can make the call, with you on speaker if they want.
Be careful about authorization. Insurers and providers may not discuss someone's account with you unless that person gives permission. HHS says only you or your personal representative has the right to access your records.
Use plain language, not baby talk. "This column is what the hospital asked for; this one is what the insurance actually allowed" respects the person. Talking down doesn't.

A short worked example (invented)

Here's a fictional walkthrough, so you can see the routine end to end. Every detail is invented.

Maria gets two envelopes in one week: an EOB from her insurer and a statement from a hospital, both for a visit to the emergency room after a fall. The EOB shows several lines, one of which has a remark she doesn't understand. The hospital statement shows a balance that looks bigger than she expected.

She writes the claim number, account number, and service date on a sticky note. Those stay off the computer.
She retypes the remark and the column headings into a note, replacing names with "Hospital" and amounts with "A, B, C, D."
She asks a free chatbot to explain the remark and the columns. It explains that the remark appears to be about network status and that the columns show the billed amount, a discount, what insurance paid, and what's left.
She asks: "What official resources explain protections for emergency care billing?" It mentions the No Surprises Act. She goes to the CMS consumer page herself and reads that the law covers emergency room visits for most types of health insurance.
She asks AI for a question script. She edits it, then calls her insurer, with the claim number from her sticky note, and asks how the claim was processed and whether the emergency-care protections apply.
She writes down the rep's name, the date, and a call reference number. Then she pastes her redacted notes into AI to clean up into a timeline.

At no point did the chatbot learn her name, her member ID, her diagnosis, or which hospital she went to. It still did its job.

The short version

AI explains; it doesn't decide. Your insurer, provider, and official help lines decide.
Redact before you paste. Name, date of birth, address, member ID, account and claim numbers, SSN, diagnosis, and barcodes all come out.
Retype instead of uploading. Photos and PDFs carry more than you think.
Prefer local or privacy-respecting free tools, and check each tool's data settings yourself.
Check every rights claim against an official page. We verified CMS, HHS, and HealthCare.gov pages today; anything else should be confirmed by you.
Use AI to prepare for the human conversation. Questions, scripts, timelines, and arithmetic checks are where it earns its keep.

You're allowed to understand your own paperwork. A free chatbot, used carefully, can make that a lot less exhausting.

FAQ

Is it safe to upload a photo of my medical bill to a chatbot?

We don't recommend it. A photo carries everything on the page, including account numbers and barcodes you might not notice. Retype only the parts you need, with names and numbers replaced by placeholders.

Can AI tell me if I'm being overcharged?

It can help you understand the line items and check arithmetic you type in. It can't tell you what you owe under your plan. Ask for an itemized bill, then put your questions to the billing office and your insurer.

Does HIPAA protect what I type into a chatbot?

Don't assume it does. HIPAA rules apply to health plans and providers, and a consumer chatbot is generally not one of those. Treat anything you type into a cloud chatbot as possibly stored, and redact accordingly.

What is a free option that doesn't send my text to the cloud?

A local model app, such as Ollama, runs models on your own computer. Its quickstart says local models don't need an API key. Check its docs for current hardware guidance before downloading.

Who can I call about a surprise medical bill?

CMS's No Surprises help desk lists 1-800-985-3059. Your insurer's member services number is on your insurance card. Your provider's billing office number is on your statement.