Grok Bot requested itemized bill, found 47 errors, called billing and disputed; bill dropped from ~$38k to $2,000
unicodeX (Twitter) · 2026-11-16
Read originalHOLY F*CK! MY HOSPITAL BILL WAS $38,000. GROK BOT FOUND THE ERRORS IN IT AND SAVED ME $36,000. I PAID $2,000 grok bot can read medical bills now. i gave it one job: find every charge i shouldn’t be paying first it requested the itemized bill. hospitals never send it unless you ask 47 errors. a room for 3 days when i stayed 2. the same MRI billed twice. $11 per tylenol, 14 times then it called billing: billing: “the balance is $38,412, we can set up a payment plan” bot: “recorded line. the itemized bill shows a duplicate MRI under code 70553 and a room charge for a day after discharge” bot: “the anesthesiologist was out-of-network. under the No Surprises Act my client owes in-network rates only” billing: “...let me send this to review” it sent a written dispute for all 47 and copied my insurance 3 weeks later, new bill: $2,000 i was about to put $38k on a payment plan for the next 6 years. now i just open my mail ask yours: “get the itemized version of this hospital bill. find duplicates, upcoding and anything the No Surprises Act covers. call billing, dispute every error and send a written notice after”
Business model
Author’s hospital bill was about $38,000. Grok Bot requested the itemized bill, found 47 errors (duplicate MRI, extra room day, high-price Tylenol, etc.), called billing citing specific codes and the No Surprises Act, and sent a written dispute copied to insurance. Three weeks later the new bill was $2,000. Reader can reuse the request-itemized → find-errors → call+written-dispute → human-confirm pattern.
Workflow
- Bot requests full itemized bill from hospital
- Bot identifies duplicates, extra days, anomalous charges (author reported 47)
- Bot calls billing with citations, sends written dispute copied to insurance
- Any legal statement or payment requires author confirmation (AgentFlow reconstruction)
Adaptation plan
Results and numbers are as reported by the original post author (bill ~$38,000, 47 errors, revised bill $2,000). $38,412 appears only in the quoted billing dialogue; the attached video uses $38,412 billed and $36,412 saved, differing from the post text. Workflow steps, freePrompt, and trial I/O are AgentFlow reconstructions and were not live-tested here.
Prompt
AgentFlow draft · Untested.
You are my medical bill review assistant (Grok Bot). Task: follow the author’s own instruction — “get the itemized version of this hospital bill. find duplicates, upcoding and anything the No Surprises Act covers. call billing, dispute every error and send a written notice after”.
Variables: Original bill or summary: {{bill}}; Actual stay dates/services: {{stay_dates}}; Insurance info: {{insurance}}; Confirmation rule: any phone statement, written dispute, or payment requires my explicit “confirm” reply before execution.
Steps: 1) Request the full itemized bill. 2) Compare against my dates and services; flag duplicates, extra days, anomalous prices with codes. 3) Draft phone talking points and written dispute citing the errors. 4) If information is missing, list the gaps and ask me to supply them—do not invent medical facts. 5) Treat written itemized data as authoritative on conflicts.
Output format: Error list (code, amount, reason), suggested call script, written dispute draft.
Confirmation boundaries: Do not place calls or send legal statements on your own; all external actions wait for my confirmation.
Untested: AgentFlow reconstruction; not live-tested here.Running costs and pitfalls
Results and numbers are as reported by the original post author (bill ~$38,000, 47 errors, revised bill $2,000). $38,412 appears only in the quoted billing dialogue; the attached video uses $38,412 billed and $36,412 saved, differing from the post text. Workflow steps, freePrompt, and trial I/O are AgentFlow reconstructions and were not live-tested here. No independent cost benchmark is available. Check subscriptions, permissions and external actions before running a workflow.