Digital Exhaust or Digital Gold? The Value of AI-Generated Clinical Visit Transcripts | New England Journal of Medicine

Example of a Hypothetical AI-Generated Clinical Visit Transcript and the Resulting Draft Note.

AI Medical Scribes: Should Your Doctor's Notes Keep the Full Conversation?

AI Medical Scribes: Should Your Doctor's Notes Keep the Full Conversation?

BOTTOM LINE UP FRONT: AI scribes that automatically transcribe doctor visits are being widely adopted across healthcare, but most U.S. hospitals are deleting these valuable transcripts immediately after finalizing patient notes—primarily due to malpractice liability fears. This practice prevents researchers from evaluating AI accuracy, studying diagnostic errors, and advancing medical knowledge, while AI companies retain the data to train future products. Experts argue that with proper safeguards, these transcripts represent "digital gold" for improving healthcare quality and safety.

The Rise of AI Medical Scribes

Artificial intelligence has entered the exam room. AI scribes—software tools powered by large language models—now record, transcribe, and summarize millions of clinical visits annually. These systems create audio recordings and detailed transcripts of conversations between patients and doctors, then generate draft medical notes automatically.

The technology has been described as a transformative force in healthcare, with adoption accelerating rapidly among health systems. Unlike traditional medical devices, AI scribes don't require Food and Drug Administration oversight, allowing for widespread deployment without rigorous independent evaluation.

For physicians drowning in documentation requirements, AI scribes offer welcome relief. Studies have shown these tools reduce documentation burdens and potentially mitigate clinician burnout. One health system reported using AI scribes for over 2.5 million patient encounters in a single year.

The Transcript Deletion Problem

Here's where the story takes a troubling turn. Before producing a draft note, AI scribes create a complete transcript (and usually an audio recording) of the clinical conversation. In theory, clinicians can cross-reference this transcript when reviewing the AI-generated note. But after the note is finalized, most U.S. healthcare institutions are quietly deleting these transcripts—and many aren't receiving audio recordings from AI scribe systems at all.

The widespread destruction of AI scribe transcripts reflects healthcare systems viewing them as "digital exhaust"—potentially harmful byproducts that create risks with few perceived clinical benefits. The primary concern is medical malpractice liability, as transcripts could create a discoverable paper trail in lawsuits. By deleting transcripts once notes are finalized, institutions preserve them long enough for editing but destroy them before attorneys would typically file lawsuits.

Healthcare institutions haven't been publicly sharing their transcript-retention policies, but conversations among colleagues and discussions in national research and legal networks suggest that most—or perhaps all—U.S. institutions are following this deletion practice.

---

SIDEBAR: Taking Control of Your Medical Records

Apps for Recording Your Doctor Visits

With hospitals deleting AI-generated transcripts, creating your own recordings has become increasingly important. Here are patient-friendly options:

Voice Recording Apps:

• Otter.ai (iOS/Android): Free tier provides 600 minutes/month of transcription with speaker identification. Paid plans ($10-17/month) offer longer recordings and enhanced features. Creates searchable transcripts automatically.
• Rev Voice Recorder (iOS/Android): Free recording with optional paid transcription services ($1.50/minute). High-quality audio suitable for medical conversations.
• Apple Voice Memos (iOS) / Google Recorder (Android): Built-in, free options. Google Recorder includes automatic transcription on supported devices.
• Trint (Web/iOS/Android): AI-powered transcription with medical terminology recognition. Plans start at $80/month but offer high accuracy for complex medical discussions.

Medical-Specific Apps:

• Med Mnemonics Voice Recorder: Designed specifically for doctor visits with built-in medical terminology support.
• Abridge (iOS/Android): Created by physicians, specifically for medical visit recording and summarization. Free for patients, with focus on healthcare conversations.

Important Legal Note: Always ask permission before recording. In most states, only one party (you) needs to consent, but eleven states require all-party consent (California, Connecticut, Florida, Illinois, Maryland, Massachusetts, Michigan, Montana, New Hampshire, Pennsylvania, and Washington). A simple "I'd like to record our conversation for my records, is that okay?" usually suffices.

Creating an Automatic Case Diary: A Practical Procedure

For patients managing complex conditions like advanced prostate cancer, maintaining a comprehensive case diary can be invaluable. Here's a streamlined procedure:

Setup Phase:

1. Choose your recording app (recommendation: Otter.ai for automatic transcription)
2. Create a dedicated cloud storage folder (Google Drive, Dropbox, or iCloud)
3. Set up a simple naming convention: "YYYY-MM-DD_Provider-Name_Visit-Type" (e.g., "2025-12-03_Dr-Smith_Oncology-Followup")
4. Create a master spreadsheet or document to track all entries

Visit Recording Protocol:

1. Before the Visit:

   • Open your recording app and test it briefly
   • Note the visit purpose and any questions you want to ask
   • If you're feeling symptoms, record a brief pre-visit note about them

2. At the Appointment:

   • Politely inform your provider: "I'd like to record our conversation for my personal records to help me remember everything we discuss. Is that okay?"
   • Start recording when the conversation begins
   • Don't let recording replace active engagement—take brief notes of action items or key concerns
   • Record any verbal test results, medication changes, or follow-up instructions

3. After the Visit:

   • Let the recording run until you're completely finished and have left the exam room
   • Save the recording immediately with your naming convention
   • If using auto-transcription, review the transcript within 24 hours while the conversation is fresh
   • Correct any medical terminology errors in the transcript (AI often mishears drug names)

Building Your Case Diary:

1. Weekly Review: Set aside 30 minutes weekly to review all recordings and transcripts

2. Extract Key Information:

   • Test results and values
   • Medication changes (dose, frequency, new prescriptions)
   • Symptom discussions
   • Treatment plans and timelines
   • Questions you asked and answers received
   • Follow-up appointments scheduled

3. Organize by Category:

   • Create sections in your diary: Labs, Imaging, Medications, Symptoms, Treatment Plans
   • Link each diary entry to the corresponding recording/transcript
   • Tag entries by topic for easy searching (e.g., "PSA," "pain management," "clinical trial discussion")

4. Maintain a Timeline:

   • Create a simple chronological list showing: Date → Event → Outcome → Recording Link
   • This becomes invaluable when new providers ask "When did you start treatment X?"

Automation Tips:

• Use IFTTT or Zapier to automatically move recordings from your app to your storage folder
• Set up automated reminders to review recordings within 48 hours
• Use voice-to-text to add quick symptom notes between appointments
• Consider using a password manager to keep all your medical app logins secure

Long-term Value: Your case diary becomes:

• A second opinion resource (new doctors can review actual conversations)
• A research tool (you may spot patterns in symptoms or treatment responses)
• A legal protection (you have documentation of what was actually discussed)
• A legacy document (family members can understand your medical journey)
• A clinical trial qualification resource (detailed history helps determine eligibility)

Privacy and Security:

• Encrypt your storage folder if possible
• Never share recordings on social media or unsecured platforms
• Consider which family members should have access and document this
• Back up your recordings in multiple locations
• Include instructions for accessing your diary in your advance directive

Cost Considerations: The basic setup can be completely free using built-in recording apps and free cloud storage tiers. For enhanced features (longer transcriptions, better medical terminology recognition), budget $10-20/month.

For Advanced Users: Consider using Claude or ChatGPT to help analyze your transcripts by uploading them and asking questions like: "What were my PSA trends across these three visits?" or "What side effects did I report and when?" This can help you spot patterns your medical team might miss.

---

What We're Losing: The Research Perspective

The same features that make transcripts worrisome to hospital lawyers make them invaluable to researchers and quality-improvement experts. Properly validated transcripts provide "ground truth" to assess the accuracy and quality of AI-generated summaries—fundamental performance characteristics that haven't been widely reported.

Consider what we don't know: Despite evidence that AI scribes reduce documentation burdens, there's a lack of peer-reviewed studies from real-world clinical visits evaluating how frequently AI scribe summaries include hallucinations or other errors, and whether clinicians catch those errors.

This isn't just theoretical. A recent report described an AI scribe hallucinating that a patient had diabetes and suspected heart disease—information that made it into the patient's chart and influenced care. In the UK, a health service AI tool generated false diagnoses for a patient, leading to him being wrongly invited to a diabetes screening appointment.

The hypothetical example in the research paper illustrates the problem clearly. A patient describes back pain that's "like fire across the side," hurts to touch, wasn't relieved by ibuprofen, and is worse with contact but doesn't worsen with movement. The AI-generated note flattens this to "musculoskeletal back pain" with a plan for physical therapy. The nuanced details that might suggest zoster sine herpete (shingles without a rash) or other non-musculoskeletal causes are lost in translation.

The Research Gold Mine

Beyond quality control, transcripts open entirely new research frontiers. By capturing the exact words used by patients and clinicians during visits, transcripts enable studying clinical communication, diagnosis, and disease at unprecedented scale.

Researchers could finally answer questions that were previously impossible to investigate:

• Does the likelihood of diagnostic error increase when clinicians interrupt patients more frequently?
• Do patients with rare cancers mention subtle, unappreciated symptoms years before diagnosis?
• What is the true breadth of patient-described symptoms, outside standardized medical nomenclature and oversimplified catchphrases?

For cancer patients specifically, this could be transformative. Imagine discovering patterns in how patients describe early symptoms that might lead to earlier detection. Or identifying communication factors that improve treatment adherence and outcomes.

The Irony: Who Gets to Keep the Data?

While healthcare systems destroy transcripts, AI companies that retain deidentified transcripts and audio recordings are amassing clinical dialogue to build future products. This creates a troubling asymmetry: the institutions providing patient care lack access to data that could improve quality and safety, while technology companies accumulate valuable training data for commercial purposes.

A Path Forward

Experts propose several solutions to shift the risk-benefit calculation:

1. Clinical and Research Advocacy: Decisions about transcript retention typically fall to risk-management committees led by institutional lawyers who see potential harms but not benefits. When these lawyers learn that AI scribes haven't undergone rigorous FDA evaluation, they may better appreciate the need to retain transcripts for studying tool performance.

Additionally, as more patients create their own visit transcripts using phone recordings and AI-enabled devices, healthcare systems that destroy institutional transcripts will lose alternative records that could resolve disputes.

2. Policy-Sharing Mechanisms: Anonymous surveys and cross-institutional dialogue could help institutions understand how many peers are retaining transcripts and share best practices. This would cover obtaining patient consent for research use and securely storing and granting access to transcripts.

3. Phased Retention Approaches: Healthcare systems could permanently deidentify transcripts so they cannot be linked to individual patients and clinicians, eliminating many legal risks while preserving conversational dialogue.

Transcripts used strictly for quality improvement—such as assessments of AI scribe accuracy or initiatives to enhance clinical communication—may be shielded from legal discovery.

4. New Legal Protections: For transcripts carrying the highest risk but offering the most value—those available for broad research use and linkable to individual patients and clinicians—new statutory protections may be needed, including legal safe harbors or malpractice "blackout" periods that shield transcripts from discovery.

What This Means for Patients

For patients navigating serious illnesses like advanced prostate cancer, this issue matters in several ways:

Immediate care quality: Without transcript retention, we cannot verify that AI systems accurately capture the nuances of your symptoms and concerns. That subtle detail you mentioned—the one that might be clinically significant—could be lost in AI summarization without anyone knowing.

Future research: Transcripts could unlock breakthrough discoveries about disease patterns, early warning signs, and optimal communication strategies. Destroying them forecloses these possibilities.

Power dynamics: The current situation gives AI companies valuable data while denying healthcare systems, researchers, and ultimately patients the ability to verify accuracy and conduct independent research.

Your own records: Consider recording your own medical visits (with permission). As AI scribes become standard, your personal recording may become the only complete record of what was actually discussed.

The Bottom Line

AI scribes are the rare innovation embraced by both healthcare executives and frontline clinicians, with use rapidly accelerating. As long as transcripts and raw data are destroyed, however, the medical field loses critical chances to assess AI scribe accuracy and forgoes powerful research opportunities made possible by capturing patient-clinician dialogue.

The question facing healthcare is whether these transcripts represent "digital exhaust" to be discarded or "digital gold" to be carefully preserved. For patients dealing with serious illnesses, the answer seems clear: data that could improve diagnosis, reveal disease patterns, and enhance care quality is far too valuable to throw away.

What You Can Do

• Ask your healthcare provider about their AI scribe policy and transcript retention practices
• Consider recording your own medical visits (always ask permission first)
• Advocate for research access to deidentified transcript data
• Support policy changes that balance malpractice concerns with research benefits
• Stay informed as this technology evolves
• Start building your own case diary using the procedures outlined in this article

---

Sources

1. Goodman KE, Morgan DJ. Digital Exhaust or Digital Gold? The Value of AI-Generated Clinical Visit Transcripts. New England Journal of Medicine Perspectives. Published online December 3, 2025. DOI: 10.1056/NEJMp2514616. Available at: https://www.nejm.org

2. Schiff GD. AI-driven clinical documentation—driving out the chitchat? New England Journal of Medicine. 2025;392:1877-1879. DOI: 10.1056/NEJMp2502345

3. Goodman KE, Yi PH, Morgan DJ. AI-generated clinical summaries require more than accuracy. JAMA. 2024;331:637-638. DOI: 10.1001/jama.2023.27898

4. Tierney AA, Gayre G, Hoberman B, et al. Ambient artificial intelligence scribes: learnings after 1 year and over 2.5 million uses. NEJM Catalyst Innovations in Care Delivery. 2025;6(5). DOI: 10.1056/CAT.25.0040. Available at: https://catalyst.nejm.org/doi/full/10.1056/CAT.25.0040

5. Olson KD, Meeker D, Troup M, et al. Use of ambient AI scribes to reduce administrative burden and professional burnout. JAMA Network Open. 2025;8(10):e2534976. DOI: 10.1001/jamanetworkopen.2024.34976

6. Nolan B. UK health service AI tool generated a set of false diagnoses for one patient that led to him being wrongly invited to a diabetes screening appointment. Fortune. July 20, 2025. Available at: https://fortune.com/2025/07/20/uk-health-service-ai-tool-false-diagnoses-patient-screening-nhs-anima-health-annie/

7. Otter.ai. Medical Visit Recording and Transcription Features. Available at: https://otter.ai

8. Abridge AI. Patient Recording Solution for Medical Visits. Available at: https://www.abridge.com

9. Digital Media Law Project. Recording Laws by State. Berkman Klein Center for Internet & Society at Harvard University. Available at: http://www.dmlp.org/legal-guide/recording-phone-calls-and-conversations

---

This article is intended for educational purposes and does not constitute medical or legal advice. Always consult with your healthcare provider regarding your specific medical situation and verify your state's recording laws before recording medical visits.

Digital Exhaust or Digital Gold? The Value of AI-Generated Clinical Visit Transcripts | New England Journal of Medicine