Executive Summary for Claims Executives
AI demand generation engines rely on large language models that are fundamentally probabilistic rather than deterministic. When processing 500+ page medical packets, these systems frequently “hallucinate” non-existent disc herniations, attribute pre-existing degenerative findings as acute trauma, and double-count medical billing. Defense examiners must implement line-by-line Bates verification rather than negotiating from synthetic summaries.
The Rise of High-Volume AI Demand Factories
Over the past three years, the plaintiff personal injury bar has undergone rapid industrialization. Platforms like EvenUp have raised over $150 million to automate the drafting of demand letters and medical chronologies. By promising plaintiff attorneys that a 40-page demand package can be generated in 48 hours for a few hundred dollars, these tools have exponentially increased the volume of represented bodily injury claims submitted to auto and casualty insurers.
However, behind the marketing claims of “proprietary legal AI models,” serious operational and evidentiary vulnerabilities have emerged.
The Three Primary Hallucination Modes in AI Demands
Our audit of demand packages generated by automated legal-tech platforms identifies three consistent failure patterns that inflate demand valuations:
1. Diagnostic Hallucination & Severity Inversion
Large language models summarize clinical notes by predicting likely token sequences. When an MRI report states “mild disc bulge at L4-L5 without nerve root impingement; changes consistent with chronic degenerative disc disease,” automated systems frequently synthesize this as: “Client suffered severe traumatic lumbar disc displacement requiring intensive spinal intervention.”
The AI transforms a radiological finding of non-traumatic degenerative aging into an acute catastrophic injury, dramatically skewing the plaintiff’s initial settlement anchor.
2. The Phantom Billing and Double-Counting Multiplier
Medical billing packets in bodily injury claims typically contain overlapping documentation: the medical provider’s master ledger, individual CMS-1500 billing forms, itemized invoices, and collection agency notices. Automated data extraction tools without strict document de-duplication frequently tally the same $14,000 surgical facility fee two or three times.
| Extracted Item | Asserted in AI Demand Letter | Primary Source Document Audit | Discrepancy / Inaccuracy |
|---|---|---|---|
| Ambulatory Surgery Center | $28,500.00 | $14,250.00 (Single procedure, billed on facility CMS-1500 and collection statement) | 100% duplicate double-count |
| Physical Therapy Visits | 36 documented visits ($9,200) | 18 attended sessions ($4,600); 18 cancelled or no-show billing codes | Inclusion of non-rendered care |
| Radiculopathy Diagnosis | Traumatic acute cervical radiculopathy (M54.12) | Documented in physical therapy intake 18 months prior to accident | Pre-existing condition omitted |
3. Treatment Gap Erasure
In personal injury litigation, treatment gaps (e.g., 60 days between emergency room discharge and initial chiropractic consultation) are material to proximate causation. Generative AI summaries tend to smooth over timeline gaps, presenting intermittent, fragmented care as continuous, uninterrupted therapy.
Legal Liability: Who Owns the Errors?
When an insurance defense examiner catches an inflated or hallucinated demand, who bears legal responsibility? EvenUp’s own terms of service provide the answer:
“EvenUp is not a law firm. Our services do not constitute legal advice... The Customer is solely responsible for reviewing, verifying, and confirming the accuracy of all work product prior to use or transmission.”
While plaintiff attorneys bear formal ethical liability under Rule 11 and state professional responsibility rules for submitting false or unverified claims, insurance carriers bear the financial loss if an untrained adjuster negotiates against an unverified synthetic anchor.
Carrier Defense Playbook: The 5-Point Verification Audit
Insurance claims leadership should mandate the following five verification controls whenever evaluating suspected AI-generated demand packages:
- Reject Summary Demands Without Page-Linked Bates Citations: Require plaintiff counsel to tie every claimed injury and billing line directly to a specific page number in the medical record submission.
- Automate Cross-Provider Duplicate Detection: Reconcile all CMS-1500, UB-04, and provider ledger statements into a deduplicated master ledger before evaluating medical specials.
- Cross-Check Prior Medical History: Specifically inspect the earliest post-accident primary care and emergency records for physician notes documenting pre-existing arthritis, prior chiropractic care, or degenerative spinal history.
- Audit Attendance vs. Billed Codes: Verify that billed physical therapy or chiropractic treatments reflect actual attended treatment notes rather than planned prescription sessions.
- Deposition Readiness: Ensure the defense adjuster’s file contains an evidence-linked chronology that defense counsel can hand directly to the judge or jury to impeach the plaintiff’s demand.
Conclusion: Countering AI Demands with Verifiable Proof
The solution to AI-generated plaintiff demand inflation is not to deploy ungrounded “counter-AI” that arbitrarily slashes claims. Doing so exposes the insurer to bad-faith litigation under state Unfair Claims Settlement Practices Acts. The only legally defensible counterweight is complete, Bates-stamped document provenance: software that anchors every single dollar and clinical fact to the underlying source record.