Content Hub › Prompt library › The 22 core medical record review prompts

Prompt library

The 22 core medical record review prompts

The 22 core medical record review prompts are the shared spine every role runs before its own work: orientation, chronology, clinical status, integrity and risk, and causation. They ask what the record says rather than what a party wants it to say, so plaintiff and defense can run the identical set.

By Ahmed Jemaa, Co-Founder & CEO of Medrecords AI · Published 8 Sep 2026

Orientation

Incident summary

Summarize the incident as documented in the records: date, time, location, mechanism of injury, complaints recorded at first medical contact, and the date and setting of that first contact. Cite the source document for each element. Note any element the records do not establish.

Run when you first open the file, after S2 and S4. Feeds every lane's opening paragraph and the orientation of anyone joining the matter. Watch for mechanism of injury drawn from the patient's later account rather than the contemporaneous first record. These often differ, and the difference is frequently litigated. If the two conflict, S17 will surface it.

Provider roster

List every provider and facility appearing in these records. For each: name, specialty, facility, date of first contact, date of last contact, number of encounters, and volume of pages attributed. Sort by date of first contact.

Run first, before anything else. Feeds S4, the records-request list, and your sense of whether the production is complete. Watch for providers who appear once as a referral target and never again. Those are S4's raw material. Also watch for a provider with many encounters and few pages, which usually means their file is partially produced.

Body region inventory

List every body region with a documented complaint, finding, or treatment. For each: date first mentioned, date last mentioned, and the provider who documented it first.

Run early, right after S2. Feeds S9, S10, S18, and the structure of most lane packs. Watch for regions that appear only in the first week and vanish, and regions that appear months later with no intervening mention. Both patterns matter and neither is self-explaining.

Missing records audit

Identify records that should exist but are absent: referrals with no corresponding encounter record, imaging or reports referenced in a note but not produced, billing entries with no matching clinical record, and date ranges with billing activity but no notes. Present as a table with the reference and its source citation.

Run second, immediately after S2. This is the most under-run prompt in the library and the one that most changes what the rest is worth. Feeds the records-request list, and it scopes the reliability of every later output. Watch for a clean return. A record set with no missing references is unusual. If this comes back empty, re-run it as three separate passes using the three scaffolds in Part I Section 3, because the combined version tends to under-report.

Chronology

Master treatment timeline

Build a chronological timeline of every medical encounter. For each row: date, provider, encounter type, presenting complaint, objective findings, assessment, and any change to the treatment plan. Cite the source document per row. Give the date as it appears in the document alongside the normalized date.

Run third. This is the artifact everything downstream reads from. Feeds S6, S10, S14, S20, and every lane pack. Verify at Tier 1 even if the eventual use is Tier 2, because errors here propagate into everything and get harder to see with each step. Watch for date drift at the earliest and latest entries and around handwritten material. The dual-date column exists to make this visible.

Pre- and post-incident split

Divide the treatment history at <incident date>. Summarize the pre-incident period and the post-incident period separately, then state specifically what changed: new diagnoses, new body regions, changes in treatment intensity, changes in documented function. Cite each change.

Run after S5, before anyone forms a view on causation. Feeds S18, S21, and the central argument of every lane. Watch for an empty or thin pre-incident period. That usually means the prior records were never requested rather than that the person was previously healthy, and treating it as the latter is a common and expensive error. Check against S2 and S4 before relying on it.

Imaging inventory

List all diagnostic imaging: date, modality, body region, ordering provider, and the stated impression verbatim. Flag any imaging referenced in a note but absent from the records.

Run during the clinical detail stage. Feeds objective-findings analysis in every lane, and the defense and IME lanes especially. Watch for paraphrased impressions. The verbatim requirement is load-bearing here. Radiology language is precise and the difference between "no acute fracture" and "no abnormality" is an entire argument.

Coding inventory

List all ICD and CPT codes appearing in the records. For each: the code, a plain-language explanation, the encounter and date it came from, and the provider who assigned it. Group by body region.

Run during clinical detail. Feeds billing reconciliation, the adjuster lane, and any communication to a lay audience. Watch for codes carried forward from a prior visit without reassessment, which is common in high-volume practices and does not mean the condition was re-evaluated that day.

Clinical status

Injury summary

For each documented injury: body region, mechanism attributed by the provider, severity as documented, treatment received, and duration of symptoms. Distinguish provider findings from patient-reported symptoms.

Run after S5 and S6. Feeds damages work, exposure evaluation, and every report-writing lane. Watch for speaker collapse. This prompt is where it happens most often, because injury narratives are where patient account and clinical finding are most tangled.

S10

Symptom trajectory

For each body region, track the documented symptom course over time and characterize it as improving, plateaued, worsening, or resolved. Support each characterization with dated citations. Note where providers disagree on trajectory.

Run after S5. Feeds MMI analysis, permanency, life care planning, and both sides of the damages argument. Watch for trajectory inferred from treatment frequency rather than documented symptoms. Fewer visits is not the same as less pain, and the record has to say which.

S11

Medication history

List all medications prescribed or documented: name, prescriber, start date, stop date if documented, indication, dose changes, and any documented reason for change or discontinuation.

Run during clinical detail. Feeds the life care planner lane heavily, and pain-management arguments generally. Watch for medications appearing in a history list versus medications actually prescribed in that encounter. Models merge these.

S12

Maximum medical improvement

Determine whether any provider documented that maximum medical improvement was reached. Quote the language verbatim with citation. If no provider states it explicitly, report that and list any observations bearing on the question.

Run once the timeline is built. Feeds valuation, permanency, life care planning, and settlement posture in every lane. Watch for an inferred MMI. Models will reason from "condition stable" to "at MMI," and those are different statements with different consequences. The absence clause in this prompt is the whole point of it.

S13

Functional findings

Extract every documented work restriction, activity limitation, functional capacity finding, and impairment rating. For each: date, provider, the restriction verbatim, and its stated duration. Note where restrictions were lifted, modified, or expired.

Run during clinical detail. Feeds wage loss, earning capacity, life care planning, and exposure evaluation. Watch for restrictions that expired and were never renewed being reported as ongoing. The duration column is there to catch this.

Integrity and risk

Run every prompt in this section regardless of lane. These findings exist whether or not you look for them, and the party who finds them second is the party who gets surprised.

S14

Treatment gaps

Identify every gap of more than <30> days between medical encounters. For each: start date, end date, duration, the encounter on either side, and any documented reason for the gap. Distinguish gaps with a documented reason from unexplained gaps.

Run in the integrity stage. Feeds the defense lane directly and the plaintiff lane defensively. Watch for gaps that are artifacts of an incomplete production rather than real interruptions in care. Cross-check every gap against S4 before treating it as a finding. This is the single most common false positive in the library. Refine by varying the threshold. Run at 30 days, then at 60 and 90, because the shape of the answer changes and the clusters tell you different things.

S15

Missed appointments

Identify all cancelled, missed, rescheduled, or no-show appointments. For each: date, provider, and any documented reason. Report the total count and the count with a documented reason.

Run alongside S14. Watch for rescheduled appointments counted as missed. A rescheduled visit that occurred is not a gap in care, and conflating the two overstates the finding in a way that is easy to rebut.

S16

Cross-provider contradictions

Identify contradictions between providers on the same finding, symptom, history element, or functional assessment. For each: the two conflicting statements quoted verbatim, both source citations, both dates, and the nature of the conflict.

Run in the integrity stage. Feeds the LNC, defense, and expert lanes most heavily. Watch for apparent contradictions that are actually change over time. Two providers describing different findings four months apart may both be right. The date columns are there so you can tell.

S17

History contradictions

Identify contradictions between the patient's reported history and what the records document, and between the patient's reports to different providers. Quote both sides with citations and dates.

Run alongside S16. Watch for intake forms. Much of what looks like a contradiction originates in a checkbox on a form the patient filled out in a waiting room, and that context matters when the contradiction gets argued.

S18

Pre-existing conditions

Identify all pre-existing conditions affecting body regions involved in this incident. For each: the condition, pre-incident treatment history with dates, documented status immediately before the incident, and how it is documented after.

Run after S6. Feeds apportionment, causation, and the central dispute in most contested files. Watch for the thin-prior-record problem from S6. A pre-existing condition that appears only in a post-incident history recital is documented very differently from one with its own treatment file, and the two should not be reported the same way.

S19

Intervening events

Identify any injury, accident, fall, or new medical event documented after <incident date> that could bear on the same body regions. For each: date, description, source citation, and the treatment that followed.

Run in the integrity stage. Watch for events mentioned once in passing, often in a note about something else entirely. These are exactly what this prompt is for and exactly what a reader skims past.

S20

Legal timeline correlation

Correlate the dates of medical encounters against documented legal events including filing, deposition dates, and IME dates. Identify any clustering or gap in treatment that aligns with a legal event, and report the alignment as an observation with dates rather than a conclusion.

Run last in the integrity stage, once you have both timelines. Watch for the model editorializing. The neutral-phrasing clause is load-bearing. Correlation here is genuinely ambiguous: treatment may cluster before a deposition because the deposition prompted a check-up, or for no reason at all. Report the pattern and let a human argue it.

Causation

S21

Provider causation statements

For each diagnosis, extract every provider statement bearing on causation. Quote verbatim with citation, date, and provider. Separate explicit causation opinions from statements that merely record the patient's attribution.

Run after the integrity stage. Feeds every lane's central question. Watch for the separation clause failing. "Patient states injury began after the collision" is not a causation opinion, and it is the single most over-read line in personal injury records.

S22

Unrelated care separation

Identify all conditions treated during the post-incident period that are unrelated to the incident. Separate them from incident-related care and state the basis for the separation, with citations.

Run last in the spine. Feeds billing analysis, exposure evaluation, and damages scoping. Watch for over-separation and under-separation both. Comorbidities that genuinely affect recovery are not unrelated, and the basis-for-separation requirement is what makes the call reviewable. ---

Run these on a file you already have

Send one real record set. You get back a cited chronology, and you review it, you revise it, you sign it.

Test a file · Book a demo

Last verified: 2026-09-08 · ← All 71 prompts