How Bursify uses AI.

AI drafts the letter. You sign it. We mail it. That is the whole loop. The model reads the medical bill or insurance denial you upload, then writes a document that cites the law you are protected under. The document does not leave Bursify until you put your name on it.

For each tool, the model reads your documents, finds the legal protection or program that applies to your situation, and drafts a document you can review and sign. The math, filing, and tracking run on deterministic code, not the model.

• Medical bills: reads an itemized bill, checks the charges against published Medicare rates, and flags lines that look overcharged or billed for care you did not receive.
• Insurance appeals: reads your denial, identifies the appeal right that applies (the No Surprises Act, ERISA, or your state insurance code), and drafts an appeal with the citation in the body.

We use Claude (made by Anthropic) for the drafting work, with higher-stakes documents routed to a stronger model and additional review. Claude runs inside our own enterprise cloud account, not over a public chatbot. Text extraction from your uploaded PDFs and images uses a managed OCR service.

Your data stays inside our environment. Anthropic does not retain your prompts or your documents for training. We do not send your bills, denials, or transaction data to consumer AI products (no ChatGPT, no Claude.ai web). The full vendor list lives in our Security and Privacy page.

No appeal, dispute, or demand letter leaves Bursify without passing all four:

1. Independent accuracy check. A second, separate review reads the draft and flags any claim it cannot trace back to your source documents (your bill, your EOB, your transactions). Flagged drafts are held, not sent.
2. Confidence bar. Every draft is scored for how well-supported it is. Anything that does not clear the bar for that product never reaches you as a ready-to-sign document, it is revised or routed to a human reviewer first.
3. Every citation is verified. If the draft cites a statute, that citation is checked against the official US Code or state code text. Made-up citations are caught here and the draft is rejected.
4. You sign the final document. Every appeal, dispute, and demand letter is signed by you in the app before it goes to fax, mail, or portal. No autonomous filing, no auto-pilot mode.

• It does not make automated decisions about you. The AI assists your choices and never decides your coverage, your care, your eligibility, or any matter with a legal or similarly significant effect. You review and approve every output, and you can route any case to a human reviewer.
• It does not move money. We have no payment rails and cannot initiate a transfer, pay a bill, or close an account on your behalf.
• It does not call your insurer or provider on your behalf.
• It does not file lawsuits. If a case needs litigation, we hand you the documentation and a referral.
• It does not promise outcomes. No guaranteed recovery, no promised result.
• It does not negotiate verbally on your behalf. That is the upcoming phone-call agent, which is opt-in only and records every call.

We benchmark accuracy on denial-code reading and medical-code lookups against known-correct answers. We also track real outcomes, paid, partially paid, denied, no response , in aggregate, so we can see which approaches work best for a given denial type and insurer and improve from there. This uses de-identified results, never your case to train anyone else's model.

When results lag over a meaningful sample, we improve how that product drafts and document the change. We do not silently change how your letters are written.

Every draft is editable before you sign. Rewrite a paragraph, change a number, swap a tone, delete a sentence. You can also opt out of AI-drafted content entirely from Settings and we will route your case to a manual queue handled by a human reviewer.

Questions about how the AI works on your case? Email support@bursify.app.