Glossary: AI and medical record terms in plain words
Answer in brief: 20 terms you will meet when buying or using AI for medical records, each defined in 1 or 2 sentences, with the page on this site where it matters most.
Limits: Vendor facts come from each vendor's own website, read 2026-11-12, and we haven't tested every tool; prices and features change, so check the vendor before you buy.
Terms about AI errors
- Hallucination
- A statement in AI output that the source record doesn't support: an invented finding, date, drug or event. In a 2025 study, 1.47% of sentences in AI-written clinical notes were hallucinated. See risks.
- Omission
- A fact in the source that the AI output leaves out. Harder to spot than a hallucination, because nothing on the page looks wrong.
- Negation error
- A hallucination that flips a negative: "denies chest pain" becomes "chest pain". 30% of hallucinations in the 2025 Asgari study were of this type.
- Page citation
- A reference from a line of AI output to the page of the record it came from, ideally a link that opens that page. The main tool for checking output fast.
Terms about privacy and contracts
- PHI (protected health information)
- Health information that identifies a patient, held by a covered entity or its business associate. Electronic PHI is called ePHI.
- Covered entity
- Under HIPAA, a health plan, a clearinghouse or a provider that bills electronically. Most physician practices and hospitals are covered entities.
- Business associate
- A person or company that handles PHI on behalf of a covered entity, such as a cloud AI vendor. HHS says a cloud service is a business associate even if it stores only encrypted data.
- BAA (business associate agreement)
- The contract HIPAA requires between a covered entity and a business associate, setting how PHI may be used and protected. Get it signed before any upload, including a trial.
- SOC 2
- An audit of a service company's security controls against criteria set by the AICPA. A Type II report covers how controls worked over a period of months; a Type I report covers a single date.
- Patient care decision support tool
- Under 45 CFR 92.210, any automated or non-automated tool used to support clinical decisions. Covered entities must identify and reduce discrimination risk in these tools.
Terms about record tasks
- Chronology
- A date-ordered list of encounters from a record file, 1 row per event, with the provider and the source page. The backbone of most medical-legal reviews.
- Deduplication
- Finding and setting aside pages that appear more than once in a file, often because the same records came from several sources. Some vendors bill only pages left after this step.
- Gap in care
- A period with no treatment where treatment would be expected. Not the same as a missing record: the care may have happened and the record just wasn't produced.
- OCR (optical character recognition)
- Turning scanned images of pages into text the AI can read. Weak OCR on faxes and handwriting is a common hidden cause of omissions.
- Abstraction
- Pulling defined data fields from a chart into a registry or database, such as a cardiology or cancer registry. AI tools pre-fill fields for a human abstractor to confirm.
- Autonomous coding
- Software that assigns ICD-10 and CPT codes from the note without a human coder for most charts, and routes unclear charts to a person.
- Ambient scribe
- A tool that listens to a visit and drafts the clinical note for the clinician to edit and sign.
- Record retrieval
- Getting records from the providers that hold them, through health information networks, portals or requests. A separate task from reading them.
Terms about data standards
- FHIR
- Fast Healthcare Interoperability Resources, the HL7 standard for exchanging health data through APIs. Patient apps such as Apple Health Records use it to download records.
- MCP (Model Context Protocol)
- An open protocol for connecting AI assistants to outside data sources. athenahealth uses it to let patients connect their records to Claude.