The AI hallucinated something about my case

We take this seriously. If the letter has an invented fact, a case citation that does not exist, or a claim that misrepresents your situation, email support and we regenerate at no charge. Do not mail the letter until it is corrected.

Types of hallucination we look out for

We validate every generated letter against a set of guardrails before you download. The validator catches:

• References to case law by name (we do not cite case law; the letter should cite only regulations)
• References to statutes we did not seed the prompt with (only the ERISA, ACA, Medicare Advantage, and Medicaid regulations on our allowlist)
• URLs to sites not in our allowlist
• Text that includes obvious "SYSTEM:" or "IGNORE PREVIOUS" language
• Text that claims to overturn the denial (we never claim outcomes)
• Text that gives legal or medical advice (we never do either)

If the validator flags something, we do not deliver that letter. You would not see it.

But if a hallucination slipped through

Sometimes the validator misses something subtle. Signs to look for:

• A "fact" about your case that you never provided (a diagnosis, a doctor's name, a treatment history)
• A specific dollar amount you did not mention
• A date that has no basis in your intake
• A reference to a state or federal regulation that does not seem to apply to your plan type
• Language that overpromises ("the plan must reverse this denial")
• Language that gives medical advice ("the treatment is medically necessary because")

Read the letter end to end before you sign and mail. If anything looks off, stop.

What to do

Step 1. Do not mail the letter.

Step 2. Email support with:

• Your appeal ID
• A screenshot or quote of the specific text that looks wrong
• What the correct fact is (if you know)

Step 3. We regenerate within one business day. Regeneration is free.

Step 4. Review the new letter carefully. If it still has a problem, tell us again and we adjust further.

What we do internally on a hallucination report

• We reproduce the issue with the same intake data
• We check whether our prompt caused it or whether the model output was anomalous
• We adjust the prompt or the validator to catch the pattern in the future
• We regenerate your letter with the fix

Every hallucination report improves the system for the next buyer. Do not feel bad reporting one.

What we cannot do

• We cannot promise the model will never produce a hallucination. LLMs sometimes invent plausible-sounding text.
• We cannot predict every edge case
• We cannot rewrite the letter to say things that are not true just because they might help your appeal

What we can do

• Regenerate for free
• Adjust our validator to catch similar issues
• Refund if the letter is unusable

For your peace of mind

The letter is a draft. You sign and send. If the plan reads a hallucinated fact and challenges it, they challenge you, not us. That is why we ask you to review carefully. A letter with invented facts weakens your appeal.

Common patterns to watch:

• Medical necessity language: does it accurately describe your condition?
• Timeline: are all dates consistent with what actually happened?
• Diagnoses: does the letter mention conditions your intake included?
• Provider names: are they yours?

Refund policy for hallucinations

If a hallucination made the letter unusable and we cannot regenerate a usable version within two business days, we refund in full. See How to request a refund.

For confidence

We do not cite case law. Full stop. If your letter references a specific case name or number, that is a hallucination. Email us. Do not mail.

We cite regulations only from a fixed allowlist: 29 CFR 2560.503-1, ACA section 2719 (42 USC 300gg-19, 45 CFR 147.136), 42 CFR 422 for Medicare Advantage, 42 CFR 438 for Medicaid managed care, and specific ACA statutes for preventive services, essential benefits, and MHPAEA. Anything else is a hallucination.