How we judge AI for IME work
Short version: we judge every AI product on 4 tests (BAA, page citations, large files, format control) using only the maker's own published pages, read on a stated date. We did not upload records to any product. We make Medrecords AI, and we say so wherever it appears.
Limits: We read each product's help pages, terms and pricing on 2026-11-12 and did not test how any product handles PHI. Plans, prices and laws change, and nothing here is legal advice for your state.
The 4 tests
- BAA available: does the maker publish that it signs a BAA, for which plans, and who can get it (an individual or only an organization)?
- Page citations to your file: do the maker's pages say that output links each statement to a page of the uploaded record? Citations to journals do not count for this test.
- 1,000+ page files: what file size, page or token limits does the maker publish, and does it say what happens past them?
- Report-format control: can you set the output layout with templates or examples, beyond typing instructions into a prompt?
We picked these 4 because an IME report has to survive a privacy review, a challenge about where each fact came from, a large referral packet, and a requester that expects a set format. Speed and interface polish are left out on purpose.
Where the facts come from
- Products: each maker's help center, pricing page, security page or product page, read on 2026-11-12 and linked in the row. "Not stated" means those pages did not say.
- Laws and rules: statute or regulation text where we could read it (Texas SB 1188, California Labor Code 4628 and HSC 1339.75, New York WCL 137, 12 NYCRR 300.2, 45 CFR 164.508, HHS guidance, the NAIC bulletin). For Texas HB 149, California AB 489 and Colorado SB 26-189 we used a dated secondary summary that links each statute, and we mark those rows.
- Search results: the Google top 10 for "best ai for ime" and 3 close searches, pulled 2026-11-12, to see what the field already covers.
- Synthetic examples: the 1,240-page file on this site is made up, labeled "synthetic, no real patient", and used only to illustrate.
We never publish a statistic without a source, and we do not publish reviews, ratings or testimonials.
What we did not do
We did not upload records, test output quality or measure error rates in any product, ours included. A table cell that says a tool cites pages means the maker says so. Run your own closed file through 2 tools before you trust either, and count missed events and wrong pages.
Nothing on this site is legal advice. The rules table is a starting list for a conversation with your counsel or your IME company.
How we handle owning Medrecords AI
This site is published by Medrecords AI, which makes 1 of the products it covers. That is a conflict of interest, and we do not claim to be neutral about it. These are the rules we hold ourselves to:
- Medrecords AI is marked "(ours)" wherever it appears and gets the same columns as every other product.
- It is never called "best overall". Where we pick it, we name the use case, and we name the cases where another product is the better choice, including cheaper options at low volume.
- Links to medrecords.ai appear only on the Medrecords AI page and in boxes marked as ours.
- Links to every other company carry a nofollow attribute.
- Vendors can send corrections to [email protected], and confirmed corrections go live within 2 business days, whether or not they help us.
How often we recheck
Every product row and every rule is rechecked at least every 45 days, and each page shows the date it was last updated. AI plans and health AI laws both changed several times in 2025 and 2026, so a row older than 45 days should be read with that in mind.