A treatment date can look definitive until the same encounter appears three different ways: the date of service in a billing ledger, the date a clinician signed the note, and the date an imaging report was finalized. If an AI-assisted demand workflow treats those fields as interchangeable, it can create a chronology that is polished, internally consistent—and wrong.
This is not a reason to abandon automation. It is a reason to design the workflow so conflicts stay visible until a lawyer or trained reviewer resolves them. The goal is not to make every source agree. It is to identify which date answers which question, preserve the underlying evidence, and prevent an uncertain fact from becoming a confident sentence in the demand.
Why treatment dates conflict in ordinary PI files
Most date discrepancies are not evidence of bad faith or poor care. They are the predictable result of records created for different purposes. A treating provider documents care. A billing system records charges. A radiology vendor finalizes an interpretation. A records custodian exports data. Each system may attach its own timestamp to the same episode.
A single encounter can therefore produce several legitimate dates:
- Date of service: when the patient was examined, treated, or imaged.
- Entry or signature date: when the clinician completed or signed the note.
- Report date: when a diagnostic interpretation was issued or finalized.
- Billing-post date: when a charge entered the provider’s billing system.
- File or export date: when the document was generated, scanned, or produced.
Those fields are related, but they are not substitutes. For a treatment-gap analysis, the date of service usually matters more than the date a note was signed. For an argument about when a diagnosis was documented, the final report date may matter. For damages support, a billing-post date may explain why an item appears in one ledger but not another without proving when care occurred.
PI files add another complication: records often arrive in batches and formats. A 200-page PDF may contain office notes, ledgers, imaging reports, referral forms, and duplicate exports. Optical character recognition can misread faint stamps, reverse month and day values, or associate a header date with the wrong page. A human can make the same error when moving quickly through an unindexed production.
What an AI demand workflow should do with conflicting dates
A responsible system should not “pick the most likely date” in silence. It should distinguish extraction from resolution. Extraction identifies candidate dates and their source locations. Resolution determines what the dates mean and which one should control a particular statement. The second task requires context and, often, attorney judgment.
Keep the date tied to its source and field
Every extracted date should retain enough provenance for a reviewer to get back to the supporting material. At minimum, that means the source document, page or section, provider, record type, and nearby label. “March treatment” is not useful provenance. “Physical therapy daily note, date-of-service field, source page 47” gives the reviewer something testable.
Source labels also matter. A date next to “signed electronically” should not be promoted to a date of service merely because it is the clearest timestamp on the page. A charge line should remain a billing record unless another source confirms the clinical encounter. Preserving those distinctions reduces the risk that clean output masks a category error.
Classify the conflict instead of flattening it
Not every mismatch deserves the same response. A useful workflow can classify discrepancies into a few operational groups:
- Different fields, no true conflict: the service occurred on one day and the note was signed later.
- Duplicate records: two exports describe the same encounter but carry different file-generation metadata.
- Probable extraction error: the surrounding note supports one date while OCR produced a different value.
- Substantive source conflict: two records assign different service dates to what appears to be the same event.
- Missing support: a billing line exists, but the corresponding clinical note is absent.
This classification prevents teams from wasting review time on harmless administrative differences while ensuring material conflicts receive attention. It also stops an AI-generated narrative from treating “unresolved” as “resolved by whichever date appeared first.”
Apply context-specific rules, not one global hierarchy
There is no universal rule that the medical note always controls, or that the bill always controls. The controlling source depends on the proposition. A clinical note is usually stronger evidence of what happened during an examination. A radiology report is the direct source for when an interpretation was finalized. A ledger supports the amount charged and may corroborate an encounter, but it may not explain the care.
The workflow should therefore ask: what fact will this date support in the demand? If the sentence concerns treatment continuity, compare service dates. If it concerns notice of an imaging finding, compare order, performance, and report dates. If it concerns specials, reconcile the clinical chronology against the itemized charges without collapsing the two into one dataset.
Escalate uncertainty before prose generation
The best checkpoint comes before the system writes the treatment narrative. Once a date is embedded in fluent prose, reviewers are more likely to read past it. A pre-draft conflict table should identify the competing dates, sources, likely explanation, and required action. The action might be “accept as separate timestamps,” “verify against original PDF,” “request the missing note,” or “attorney decision required.”
This is where a structured medical chronology workflow earns its place. Chronology is not merely date sorting. It is the controlled process of connecting events to reliable sources while preserving exceptions for review.
A practical review protocol before drafting the demand
Firms do not need to re-read every page from scratch every time a discrepancy appears. They need a repeatable protocol that routes the right issue to the right reviewer.
- Normalize without overwriting. Convert dates into a consistent display format, but retain the original extracted value and field label.
- Group records by encounter. Compare notes, imaging, referrals, and billing lines that appear to describe the same episode of care.
- Separate harmless timestamp differences. Mark later signature or report dates as distinct fields rather than conflicts when the service date is clear.
- Check the source image. When OCR and structured output disagree, review the original page rather than another generated summary.
- Reconcile gaps and duplicates. Confirm whether an apparent treatment gap is real, caused by missing records, or created by using a billing-post date.
- Route material conflicts. Escalate discrepancies that affect causation, treatment continuity, diagnosis timing, or claimed specials.
- Lock approved chronology facts. After review, preserve the accepted date, the reason for the decision, and the supporting source so later revisions do not reintroduce the conflict.
- Run a final narrative-to-source check. Before sending, verify every material date in the demand against the approved chronology and underlying record.
The review should be proportional. A one-day difference between service and signature may require only classification. A conflict that appears to create a six-week treatment gap can change how an adjuster reads causation and mitigation, so it deserves direct review. The workflow should surface that difference rather than treating every warning as equally urgent.
For a broader discussion of where automation assists—and where attorney review remains essential—see Medical Chronology AI for PI Firms.
How Legal Power AI fits
Legal Power AI helps plaintiff PI teams organize records, build treatment chronologies, and draft demands from structured case materials while keeping attorney review at the center of the workflow. The platform processes data with BAAs from OpenAI and Anthropic, but technology and vendor controls do not replace professional responsibility: counsel must verify dates, source support, and the accuracy of every AI-assisted document before it leaves the firm.
The record should remain more authoritative than the draft
A demand letter is downstream from the file. When the file contains conflicting dates, good automation makes the disagreement easier to see; bad automation hides it behind confident prose. Plaintiff firms should require source-linked extraction, field-aware date handling, conflict classification, and a documented attorney-review checkpoint before narrative generation.
That discipline does more than prevent a typo. It protects the credibility of the treatment story, keeps gap analysis anchored to actual service dates, and gives the drafting team a defensible path from source record to final assertion. AI can reduce the mechanical burden of finding and comparing dates. The legal judgment about what those dates prove remains with the attorney.
Build a source-linked chronology before the demand is drafted.