How our medical bill checker works

Think your bill has a mistake? Many people start there. MedBillCheckup never decides that a charge is a mistake: it finds the lines worth asking the billing office about, and only the provider or your insurer can confirm one. The common kinds are in medical bill mistakes to check.

1. Read the bill, on your device

Open a PDF, or take a photo of each page. Your own browser does the reading. A PDF from a provider or insurer portal usually carries text, which is read directly. A photo or scan has no text to read, so your browser runs text recognition on it, on your device. A scanned PDF whose only text is a hidden layer its scanner added is read the same way, from its picture. The file is never uploaded, and it stays in your browser’s storage.

You choose which country the bill is from, because billing rules differ. The United States has the deepest support today. If a photo is hard to read, we say so straight away rather than giving a confident-looking result from an unreadable page.

What comes out is a record of the provider, the dates, each line and each amount, with how confidently each value was read. Anything we are unsure of is left empty or marked for you to check. Your name, address, account and member numbers and date of birth are kept apart on your device, and removed from what is sent wherever your browser recognizes them.

2. Check what we read

Hold your paper bill next to the list and change any number that differs. Your corrections are marked as yours and trusted from then on. Everything up to this point is free and needs no account.

3. See what to ask

For free, you find out how many things on the bill are worth asking about, of what kind, and which comes first and why. If nothing is, you stop there and pay nothing. The full list, with every line, the reason for each, the questions to ask and a letter for the billing office, is $6.99 for this bill, or included with Pro. Pro is $9.99 a month or $59.99 a year.

Each time the checks run, your browser sends the figures you just checked: the charges with their descriptions, codes, dates and amounts, the bill’s totals and dates, who billed each charge when it is printed as an organization, your corrections, the country, and any answers you gave to the optional money questions (amounts and yes or no only). Not the file, not a picture of the page, and not the provider’s name at the top of the bill, your insurer or the claim and statement numbers: the checks only learn that the bill shows them. They are checked in memory, the answer comes back, and nothing is kept. If you buy the full list for one bill, the payment also carries a fingerprint of those figures (a code that cannot be turned back into them), so your receipt opens that bill only. Apart from these, nothing about your bill leaves your device.

The checks are fixed rules, not an AI model: multiplication, addition, date comparison and text similarity. Arithmetic should be exact and give the same answer every time. They look for:

• Charges and billing codes that appear more than once
• Quantities on services usually given once
• Line totals against quantity times unit price
• The lines against the stated subtotal and total
• Insurance paid, the adjustment and what you owe against the total charges
• Dates that conflict with each other or with the statement date
• Information missing from the statement that you can ask for
• A similar charge from two different providers for the same service
• Adjustments or credits that look larger than the charges they apply to
• Routine supplies often included in a room or procedure charge, billed on their own line
• Everyday items, such as common pain relievers, priced far above what a shop charges
• Drug and supply quantities above the unit limits Medicare publishes (its Medically Unlikely Edits)
• Money you may get back or may not owe, from a few optional answers, such as the amount your insurer’s EOB says you owe for this visit
• With Pro: your plan’s yearly maximum, typed once and counted across the bills on this device

Each of the 19 checks, with what it compares and why a flag can still be normal.

Each item on the full list says what we found, why it’s worth asking, what to check, the lines involved and how sure we are. Each item comes with the question to ask. From the list your browser writes a letter for the billing office and a PDF report, on your device. You can edit the letter, copy it, or download the report.

With Pro you can also type a question about a US bill on its full list. A small model on your device matches it to the closest question we prepared and shows the answer we wrote, filled with your bill; when none is close enough, it offers the closest topics instead. It does not write answers or judge the bill.

Call with your call guide, and if they say no, follow the next official step. The full list gives a call guide for each thing worth asking about, and shows the usual next official step for situations like yours, with its deadline rule and source. All of these steps are in our free guide What to do if the billing office says no.

See a sample full list, made by the same checks from an example bill.

Paying, and coming back

Payment happens on Stripe’s page; we never see your card. When you come back, your browser keeps a signed receipt for the bill you paid for, or a pass for your subscription, so you’re never asked to pay twice for the same bill.

There is no account for a single bill. Subscribers can sign in on another device: enter the email used at checkout and we email a link. That is the only kind of account there is.

What we won’t do

We won’t tell you a charge is fraudulent, illegal or definitely wrong. We won’t tell you what you owe or what you’re entitled to. We won’t explain a billing code beyond Medicare’s public code list, except to say in our own words what kind of service a procedure code is (try the free code lookup). And we won’t invent anything that isn’t on your bill. We never call, write to or negotiate with anyone for you, and we never mail anything for you.

How sure we are about each item is worked out by fixed rules from how strong the signal is, how clearly we read the values it depends on, and the quality of the document.