Medical Records Retrieval for Lawyers: What You Get, What Arrives Broken, and What to Do the Day It Lands

The concession first, because it decides whether this page is useful to you. Hathr.AI does not retrieve records. We do not subpoena providers, chase releases, or call the release-of-information desk at a hospital for the fourth time. If what you need is retrieval, use a retrieval vendor — several are good, and this page names what to ask them.

What this page is actually about is the part nobody sells you: the day the production lands. Retrieval vendors are measured on completeness and turnaround. Nobody is measured on whether the 4,000 pages that arrive are in a state a human can work with. They usually are not.

Key takeaways

• Retrieval and review are different purchases. Buying one and expecting the other is the most common budgeting mistake in this workflow.
• Ask a retrieval vendor four questions: what is the completeness guarantee, is the output OCR'd or image-only, is it Bates-stamped on delivery, and what happens when the provider produces late.
• An image-only production costs you the entire review, twice — once to read it, once to re-read it when the supplemental arrives.
• Records in your possession are protected health information. Hathr.AI reads them inside AWS GovCloud under a FedRAMP High authorization boundary, with a signed BAA on every HIPAA-compliant plan.
• Hathr.AI's record-review output is measured at 97% accuracy in production, across more than 100,000 records a month. The 3% is where it asked for more information rather than guessing.

Retrieval, review, and the gap between them

Three separate jobs get collapsed into one line item on a case budget, and the collapse is where money leaks.

The middle row is the one that ruins schedules. A retrieval vendor's job ends when the file transfers. If that file is 3,000 image-only pages with no Bates numbers and three duplicate copies of the same discharge summary, the review clock starts with a day of unbillable janitorial work, and nobody quoted it.

Four questions to ask a retrieval vendor, in writing

1. What is the completeness standard, and what happens when a provider is late?

Ask what percentage of requests come back complete on the first pass, and what the process is when they do not. Late supplemental productions are normal. What matters is whether they arrive labelled as supplemental, or drop into the same folder with no marker — because the second case forces a full re-review to find what changed.

2. Is the delivered file OCR'd, or image-only?

This single answer changes your review cost more than any other. An image-only PDF cannot be searched, cannot be cited by text, and cannot be read by most tools. Ask for the OCR standard and whether handwriting is attempted at all — most pipelines skip it, and handwriting is where intake complaints and onset dates live.

3. Is it Bates-stamped on delivery?

If not, someone in your office stamps 4,000 pages before the chronology can cite anything. Ask whether stamping is included, at what point in the process, and whether supplementals continue the same sequence.

4. Where do the records sit while they are being handled?

Medical records are protected health information no matter whose server they are on. Ask where the vendor stores them, for how long after delivery, who on their staff can open them, and whether any part of the pipeline is offshore. Then ask the same question of every downstream tool you plan to use on the file.

What actually arrives

A typical personal-injury production, described honestly:

• Mixed provenance. Hospital records in one format, an imaging center's in another, a physical-therapy clinic's exported from a system that prints four visits to a page.
• Scans of scans. Faxed records photocopied and re-scanned, sometimes rotated, sometimes at an angle. The text is legible to a person and invisible to a text-only parser.
• Handwriting. Intake forms, nursing flow sheets, margin notes on printed orders. Disproportionately important, disproportionately skipped.
• Duplicates. The same discharge summary appearing in three custodians' productions, each with different page numbers.
• Billing mixed into clinical. EOBs and itemized statements interleaved with progress notes, which matters because the billing record often dates an encounter the clinical record does not.

None of this is anyone's fault. It is what a records production is. The mistake is planning the review as though a clean, searchable, deduplicated file is going to arrive.

What to do the day it lands

1. Inventory before reading. Every file, its custodian, its page range, and whether it is text or image. Twenty minutes, and it tells you where the week is going to go.
2. Decide the review method against the inventory, not against the page count. Four thousand clean text pages and four thousand image-only pages are different projects.
3. Get one pass of dates and page citations out first, before any summarizing. Sorting early surfaces the gaps and the missing custodians while there is still time to chase them.
4. Flag the illegible rather than guessing at it. An entry a reviewer knows to verify is safe. An entry silently filled in from context is a deposition problem.
5. Hold the whole production together. Onset conflicts and treatment gaps only exist across documents. A method that reads one file at a time cannot see them.

Steps 3 through 5 are what our medical chronology template is built around — the source citation, legibility confidence, conflict flag and gap columns exist precisely because productions arrive like this.

Where the crossover is, in money

The three jobs are priced on three different units, which is why a case budget that looks fine in month one stops looking fine in month four.

• Retrieval is priced per request or per custodian, plus statutory provider copy fees. It scales with how many places your client was treated, not with how sick they were.
• Outsourced review is priced per page. It scales with the size of the production, and it re-prices every time you go back with a new question.
• A per-seat tool is flat. It does not scale with either.

The crossover is not really about page count. It is about question count. A production reviewed once, by a service, for a chronology, is a single invoice and a reasonable one. The same production asked forty follow-up questions — every mention of the prior shoulder, every provider who documented causation, every place the record contradicts the deposition — is forty more requests from a service and forty free queries in a platform.

So the honest way to model it is: how many times, realistically, will someone need to go back into this file? If the answer is once, buy the review. If the answer is "constantly, for eighteen months," the arithmetic inverts, and it inverts earlier than most firms expect. At $47 per user per month with no per-page fee, a 12,000-page production costs the same to interrogate as a 600-page one.

One caveat we would rather state than have you discover: a flat tool does not remove the human hours. It moves them. The reviewer still verifies the flagged entries and the lawyer still decides what the record means — what disappears is the transcription, not the judgment.

Where Hathr.AI fits, and where it does not

Where it does not: retrieval, subpoenas, provider follow-up, affidavits, or case management. If those are the bottleneck, this is the wrong tool and a retrieval vendor or a litigation platform is the right one.

Where it does: the moment the file is in your possession. Hathr.AI runs Anthropic's Claude models inside AWS GovCloud, under a FedRAMP High authorization boundary, with a signed BAA on every HIPAA-compliant plan, accepted electronically at signup. The OCR reads scanned and handwritten pages that text-only pipelines skip. Record sets of up to 100,000 pages are held in a single workflow, so a question that spans page 200 and page 3,400 is answerable in one pass rather than stitched from chunks.

Output accuracy is measured at 97% in production, across live records-review work running at more than 100,000 records per month. The remaining 3% is not wrong answers — those are the cases where Hathr stopped and asked for more information rather than filling a gap on its own. On a production this messy, that behaviour is the point.

Hathr surfaces what the record says. A human decides what it means. Nothing here substitutes for the reviewer's judgment or an expert's opinion.

At $47 per user per month with no seat minimums and no per-page fee, the cost of a 12,000-page production is the same as a 600-page one — which is the opposite of how retrieval and per-page review are priced, and worth modelling against your actual monthly volume.

Upload one production and get a cited chronology — free for 7 days →    The full buyer's guide →

Frequently asked questions

What is medical records retrieval for lawyers?

Medical records retrieval is the process of obtaining a client's complete medical file from every treating provider, using authorizations or subpoenas, and delivering it to the firm with any affidavits required for admissibility. It is a separate purchase from medical record review, which is the work of turning that file into a cited chronology and an issues memo.

How long does medical records retrieval take?

It varies by custodian and jurisdiction, and supplemental productions arriving after the first delivery are normal rather than exceptional. The number worth asking a vendor for is not average turnaround but first-pass completeness, and how late supplementals are labelled when they arrive.

Does Hathr.AI retrieve medical records?

No. Hathr.AI is not a retrieval service and does not contact providers. It reads records already in your possession — producing a cited chronology and answering questions against the full production — inside AWS GovCloud under a FedRAMP High authorization boundary, with a signed BAA on every HIPAA-compliant plan.

Is it safe to upload retrieved medical records to an AI tool?

Only if the vendor's data handling supports it. Medical records are protected health information regardless of who holds them. Ask whether the vendor signs a BAA and what surfaces it covers, where records are processed, and whether your data is used for model training. Hathr.AI answers those three the same way for a two-person firm as for a national practice.

What should I ask a medical records retrieval vendor?

Four things in writing: the first-pass completeness standard and the process for late supplementals; whether delivery is OCR'd or image-only and whether handwriting is attempted; whether Bates stamping is included and whether supplementals continue the sequence; and where the records are stored, for how long, and who can open them.

Why does an image-only production cost more to review?

An image-only PDF cannot be searched, cited by text, or read by most review tools, so the review begins with a conversion pass nobody quoted. It also breaks a second time when supplemental records arrive, because there is no reliable way to diff an image against an image.

Accuracy measured in Hathr.AI production environments during live records-review work at a volume exceeding 100,000 records per month. Last updated: August 2026. Informational, not legal advice.