Best AI for QME work: picks by task (2026)

By Ahmed Jemaa, Co-Founder and CEO of Medrecords AI. Updated 2026-11-12.

Limits: We read each product's own pages and the DWC sources on 2026-11-12 and did not upload records to any product. Prices, plans and rules change, and nothing here is legal advice; ask your own counsel about Labor Code 4628.

Which AI should a California QME use?

Use 2 kinds of AI and keep them apart. The first reads the record file: it needs a BAA, handles thousands of pages and ties every line to a page so you can check it before you write. The second answers medical questions with no patient in them, such as what the literature says about cumulative trauma and lumbar disc disease. No AI should take the history, write your discussion or reach your conclusions. Labor Code 4628(a) reserves those for the physician who signs, and a report that breaks the rule can be struck and go unpaid.

Picks by QME task, as of 2026-11-12

Each row names the tool to try first for 1 task, the rule that limits the task, and the catch. Facts come from each maker's own pages and from dir.ca.gov, read on 2026-11-12. The tool directory has a source link on every row, and the task guide covers each task in depth.

What the QME rules say about AI

No Title 8 regulation mentions artificial intelligence. We searched dir.ca.gov on 2026-11-12 and found none. The rules that bind AI use are the ones that already govern who prepares the report:

• 8 CCR 41(c)(7): the evaluator writes all portions that discuss medical issues, the research behind them and the conclusions. 41(c)(2) requires a list and summary of every record reviewed.
• Labor Code 4628: only the signing physician may take the history, review and summarize prior records, or draft the conclusions. If someone else excerpts the records, the physician reviews the excerpts and the entire outline. The report names everyone who did nonclerical work. Breaking the rule makes the report inadmissible and unpaid, and a knowing violation carries a civil penalty of up to $1,000 each.
• The DWC California Medical-Legal Quality Assurance Checklist (last updated 2026-10-10) suggests that a physician who used AI name the application and certify that every statement and conclusion was verified. It says this is the authors' suggestion, not an official DWC determination, and it neither allows nor bans AI.
• Gurrola Martinez v. H & H Wallboard (WCAB panel decision, 2025-08-25): a workers' comp judge found that a QME's reports should be struck because a third party summarized the records. The board sent the case back so the QME could be heard, and said the 4628 finding was very likely correct.
• Lopez Mendoza (WCAB panel decision, 2025-12-01): a defense petition cited cases an AI chatbot invented. The board gave notice of sanctions up to $2,500 and said the person who signs is responsible for checking the work.

Panel decisions do not bind other panels, and neither case ruled on AI in a QME report. Together they show where the risk sits: undisclosed help with the record summary, and unchecked AI output under your signature.

General assistants vs QME record tools

The table shows 4 of our 5 tests; the 5th, price against the ML fee, has its own section below. They come from what a QME report must survive: a privacy review, a 4628 challenge, a 3,000-page file from 2 parties, and a deposition where opposing counsel asks where a fact came from.

What AI costs against the ML-201 fee

Synthetic example, no real patient. Take a 1,500-page panel QME file for an orthopedic ML-201 evaluation, the only file you run that month. 8 CCR 9795 sets the ML-201 fee at $2,015 and says it already covers record review, report preparation including typing and transcription, and overhead. Here is what each published price would charge for the file, read from each maker's pricing page on 2026-11-12:

Monthly plans only look cheap if you use the allowance. 8 CCR 9795(b) folds report preparation and overhead into the fee, so we read the subscription as overhead you absorb, not a charge you pass on; we found no DWC statement on AI billing. Pages past the first 200 are billed separately under the same regulation, and how to count them is a topic of its own.

Where AI fails on QME files

• Records with no declaration. Under 8 CCR 9793(n), documents sent without the provider's page-count declaration do not count as received. An AI tool reads whatever you upload, so separate those files first.
• Mental health records. 8 CCR 35(c) routes some mental health records away from the worker. They still reach you, and they need the same BAA cover as the rest of the file.
• Condensed depositions. A transcript with 4 pages on 1 sheet reads as 1 page to software. 9793(n) counts each condensed page separately, so check the tool's page count against the declaration.
• Late records. 8 CCR 35(i) has you serve the report and note records not received within 10 days after the exam. Make sure the AI chronology does not blend a later upload into the first report's summary.
• Video and audio. Most tools we list read PDFs. Docura is the one that says it organizes audio and video. Sub rosa review time is your time under ML-205.

How to choose in 5 steps

1. Split the tasks. Most QMEs need a record tool and a literature assistant. A dictation tool is the third, if you dictate.
2. Get the BAA before the first upload, and read which features it covers. Then read the chatbots and PHI page for what stays out of consumer plans.
3. Run a closed case you served and know well through 2 tools. Score missed priors, wrong page cites and invented facts with the closed-case scoring sheet.
4. Write your disclosure line now. Name the tool, say what it did and say you reviewed the records yourself. The DWC checklist suggests both the name and a verification statement.
5. Price the file at your real volume, using the table above with your own page counts.

Questions QMEs ask about AI