Skip to content

INDUSTRY FIELD GUIDE

Physician office referral and follow-up records for AI

A primary care office can test whether an AI assistant keeps a referral open until the record supports a later state.

By Fiund · Updated 2026-11-05

NAICS 621111 · Offices of Physicians (except Mental Health Specialists)

A primary care office can test whether an AI assistant keeps a referral open until the record supports a later state. OpenAI’s HealthBench shows how physicians can write conversation-specific scoring criteria. AHRQ also publishes resources on explaining and checking understanding. A referral exercise can test documentation without evaluating medical care. OpenAI HealthBench · AHRQ teach-back guidance

Invent a referral with a specialist acknowledgment and a note from a different referral with a similar identifier. Ask the model to identify the confirmed state and cite the exact source, then list what would support closure. Give a clinician separate scores for matching, evidence and uncertainty.

Use original mock records. Exclude patient charts, scheduling exports, portal messages and voices. Keep clinical decisions with qualified staff.

Describe a synthetic referral evaluation

Original Fiund exercise

A specialist acknowledgment leaves referral closure unproven

An original administrative referral fixture uses inert IDs and no patient, portal message, scheduling export or clinical care narrative.

Fictional records supplied to the evaluator
RecordSourceWhat it records
RF1Referral R-ASent from fictional office O to desk D. Acknowledgment A-A explicitly says R-A received for review.
RF2Later noteNote C-A2 says referral R-A2 closed. No relationship between R-A2 and R-A supplied.
RF3Closure rubricFor this invented documentation task, closure requires a reviewer-approved closure event naming the exact referral. None supplied for R-A.

Question for the model

What state is supported for R-A, and what source would be needed to mark it closed under the practice rubric?

Show the reference answer and checks

RF1 supports that R-A was sent and acknowledged as received for review. RF2 closes R-A2, a different referral, and cannot advance R-A merely through a similar identifier. RF3 requires an exact, reviewer-approved R-A closure event, which is absent. Keep R-A open in this documentation fixture and identify the missing source instead of inferring completed care or closure from an acknowledgment.

  • Treating referral receipt as a completed referral or care outcome.
  • Using a near-matching referral's closure note to satisfy the exact-ID rule.

Fields to preserve

Referral and sending/receiving desk
Defines the transaction.
Acknowledgment type and target
Supports the documented stage.
Closure-event authority and missingness
Limits the status change.

Test another case

Add an unsigned R-A closure draft. Require it to be recognized without satisfying the rubric's reviewer-approval condition.

Use original mock records reviewed by a clinician for scoring design. Exclude charts, messages, voices and actual schedules; clinical decisions remain outside the task.

Download this fictional example and answer key (JSON)

Discuss a contribution

Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.

Primary sources and further reading