Reports › Why experts switch
Why four experts switched from InPractice AI, Dodon AI, ChatGPT, and Claude to Medrecords AI
In August 2026 four medical-legal professionals moved their record review to Medrecords AI: a nurse consulting team that left InPractice AI over duplicate-page billing, a legal nurse consultant who left Dodon AI over merged VA notes, a physician expert who left ChatGPT and Claude over HIPAA limits, and a forensic psychologist whose assistants went back to school.
Four switches to AI medical record review, one month
This report tells four switching stories from August 2026, with the complaints quoted near-verbatim from recorded onboarding and support calls. The customers are de-identified; the tools they left are not. One thing to hold onto while reading: everything said here about InPractice AI, Dodon AI, ChatGPT, and Claude is the customer's reported experience, not a claim from the vendor's published record. For what each vendor publishes about itself, see the InPractice AI review, the Dodon AI review, and the alternatives pages.
The four had little in common. A nurse consulting practice focused on medical billing analysis. A legal nurse consultant working personal-injury files. An internist who testifies in medical-malpractice cases. A forensic psychologist evaluating conservatorship and conditional-release cases. They left different setups for different reasons, and the reasons rhyme.
Why a nurse consulting team left InPractice AI
A consultant who runs operations for a busy legal nurse practice evaluated more than twenty medical chronology tools before choosing InPractice AI and prepaying roughly $4,000 in credits at what he described as seven cents per uploaded page. He had real praise for the product, its side-by-side chronology view in particular. The complaints below are his, quoted from the August 2026 calls.
Duplicate pages, billed at full rate
“We just did a case with over 2,000 pages, and the attorneys who gave it to us, there were tons and tons of duplicates. And there was no way to know that beforehand.”
The tool, in his account, could deduplicate “on a basic level, but not to the level of depth that we really need, because we're going to continue to get these case files that are just messy.” And because billing was per uploaded page, the practice paid full rate for copies of pages it had already paid for. His summary: “This doesn't deduplicate as well, which is kind of the whole reason we started looking into you.”
A deduplication black box you cannot audit
“What we're using right now, it's a black box with that functionality. I have no way to see how you handled the duplication.”
Include-only filters and mislabeled records
The practice builds billing-specific chronologies, which means excluding everything except billing records. His account: “It doesn't have the ability to exclude. It only has the ability to include,” and the include filters only work when every record is labeled correctly, which in his experience they were not: “this usually happens when the system doesn't accurately label different types of records.”
Exports missing the table of contents and the narrative
On the export side: “The exports that we have right now, it doesn't have the table of contents.” The case narrative had the same problem: “I haven't found a way to export it.” What the practice wanted in one file was the table of contents, the chronology, and the hyperlinked source documents together.
- ›Deduplication runs in the open: suspected duplicate pairs are shown side by side so a reviewer confirms each match, and page counts settle only after duplicates are removed.
- ›Self-service pricing bills 10 cents per deduplicated page; duplicate pages are removed and never charged.
- ›Content selection filters both include and exclude by record category, subcategory, provider, and facility, so a billing-only chronology is a filter, not a feature request.
- ›Hyperlinked exports carry the table of contents, the narrative, the chronology, and the source documents in one clickable file.
Why a legal nurse consultant left Dodon AI
A legal nurse consultant working personal-injury cases opened her demo call with the verdict on Dodon AI: “It just does not do what I need it to do.” And: “It's not user-friendly for me.” She had prepaid, in her words, “$1,200 for 120,000 credits.” Her complaints, quoted from the call, ran longer than anyone else's in this report.
Two or three VA notes merged into one chronology entry
“The VA notes are killing me. The VA has terrible records. So there's multiple notes on a single page.”
VA files often separate notes with nothing more than a horizontal line, and in her account the tool read each page as one unit: “What I'm finding is this platform is pulling two to three notes into one summary entry. So it'll pull two different imaging results for a shoulder and a leg into one entry under one provider.”
Admission dates stamped instead of note dates
“It's pulling the wrong dates. This is the admission date up here. It's not actually pulling my entered note date. So I have to manually go in and change all the dates.” A chronology with the wrong dates on every inpatient note is not a chronology; it is a worksheet.
Annotations lost with every regenerated chronology
“I can't add notes in it. So every time I pull a chronology, I have to pull it to a PDF. And when I find errors, all the notes that I've taken on my PDF, I have to manually pull them to the new PDF.”
Sections she added by hand had a second cost: “I have to enter entire new sections that are not hyperlinked, because I'm editing it and adding it.”
Hallucinations on big files, and “there's no fix”
“I've noticed there's hallucinations that are happening more often, especially with bigger documents. It's just having a hard time processing.” She raised all of it with the vendor and reported the answer plainly: “I have really good contact with the guys at Dodon, but they're telling me that there's no fix for this.”
The stakes are scale. She has a 9,000-page case in hand and a 69,000-page nursing-home case behind it: “The 69,000-page document is where I figured out that Dodon isn't going to work for me if it can't even process this 9,000-page document.”
- ›Medical record OCR splits at the note level, not the page level, so multiple notes on one page become separate dated entries; see also VA C-file review.
- ›Uncertain dates are flagged for confirmation instead of silently stamped with an admission date.
- ›In-app annotation: notes, highlights, and comments live in the workspace and survive reprocessing, and added records slot into the existing chronology instead of forcing a re-pull.
- ›Every chronology line carries a citation to its source page, so nothing depends on trusting an unlinked summary.
Why a physician expert witness left ChatGPT and Claude
An internist who reviews medical-malpractice cases at $650 an hour started, like many physicians, with a consumer chatbot. His words: “I started with Claude, but Claude is not protected. Claude was amazing. And then I'm like, oh, I need this HIPAA protected.” So he moved his reviews to a paid ChatGPT setup, and hit the next set of walls.
“If uploading 10,000 pages, it doesn't read everything clearly”
“I upload a document, and it's too much memory. They're like, well, if you upload too much, our AI can't... sometimes we'll miss things. We've got to be careful. If uploading 10,000 pages, it doesn't read everything clearly.”
For an expert whose files run to thousands of pages, a tool that admits it may skip content is disqualifying. His conclusion on the call: “I've got to cancel that other company.”
A 4,000-page case with zero nursing notes
Completeness cuts the other way too. “Right now I have a case where they sent me 4,000 pages and I said, hey, I reviewed it all. Not one page has any nursing notes. They're missing. And I'm like, I can't give you an opinion.” An expert opinion on an incomplete record is a liability; spotting the gap is part of the review.
The deposition questions every expert will get
“They'll ask me: oh, do you use AI? What program do you use? Is it HIPAA protected? Do you have a BAA contract with them? How do you know it's protected?”
A consumer chatbot signs no business associate agreement, and the wrong answers to those questions hand opposing counsel the cross-examination. What he needed was analysis he could keep and paperwork he could describe plainly under oath.
- ›A signed BAA and HIPAA-compliant, US-hosted processing he can name at deposition without flinching.
- ›Missing-records identification cross-checks what the file references against what it contains, so a 4,000-page case with no nursing notes gets flagged before the opinion, not after.
- ›Cited case answers: every answer lists the source records behind it, in clinical, evidence-based, or verbatim-extract mode.
- ›Cases beyond 50,000 pages are supported, with every page processed; there is no read-what-we-can ceiling.
From assistants back in school to AI: a forensic psychologist's switch
A forensic psychologist with 25 years in practice, “I've been doing this 25 years,” evaluates conservatorship and conditional-release cases on report templates the courts have seen from her for years. Her review pipeline was human: a few assistants who worked up each file before she wrote. Then the assistants went back to school, and the pipeline went with them. She considered hiring replacements, but that was a daunting task; finding the right people for work this specialized is very hard. So she went looking for an alternative and found Medrecords AI.
Her requirements were the opposite of generic. Her cases have no date of injury; the anchor is, in her words, “typically the date they were competent to stand trial or the date of the qualified offense.” Medication stability matters more than most of the medical file: “They need to be on the same mental health medication for a year.” And the clinical scope is specific: “any medical records connected to the brain is what I would be interested in.”
That is what her account became: her own templates rebuilt into the report builder, a forensic case type of her own, and drafts that follow her structure, which she edits and signs. The AI drafts; she decides.
- ›Custom report builder: her existing court-facing templates, rebuilt and reused, not replaced.
- ›Forensic psychological evaluation workflows with case types anchored on qualifying offense and competency dates instead of a date of injury.
- ›Narrative drafts she edits and signs. AI drafts, humans decide, always.
What experts want from AI medical record review software
- ›Verification over trust. Every one of the four asked to see the chronology next to the source page before anything else.
- ›Billing that respects the file. Counting pages after deduplication, with duplicates never charged, came up in every pricing conversation. Paying full rate for copies is the fastest way to lose a reviewer's trust.
- ›Coverage before opinion. A chronology of an incomplete file is a liability. Flagging what the record references but does not contain figured in three of the four stories.
- ›Fit over features. Custom templates, custom filters, custom case types: each switcher asked the tool to match how they already work, not the other way around.
- ›Paper that survives a deposition. A signed BAA and HIPAA-compliant processing were entry conditions, not bonuses.
How these stories were collected
All four accounts come from recorded onboarding, demo, and support calls between these professionals and the Medrecords AI team in August 2026. Nobody was compensated and nobody was interviewed for marketing purposes; these were working sessions, condensed, with the complaints quoted near-verbatim. Roles, specialties, file sizes, and the problems described are reported as stated. Names, employers, locations, and all patient information are removed or generalized. Where a customer described a competitor, that account is attributed to the customer, not presented as the vendor's published record; each vendor's own published claims are covered separately in the vendor reviews.
Switching to Medrecords AI: free to test, nothing to lose
If you are looking for an InPractice AI alternative, a Dodon AI alternative, or a HIPAA-compliant replacement for ChatGPT or Claude in your record review, the terms these four switched on are the standing offer:
- ›Test up to 500 pages free. Upload a real case at test a file and check the output against the source pages before paying anything.
- ›Pay only for deduplicated pages. 10 cents per page on self-service, and duplicate pages are removed and never charged.
- ›Cases beyond 50,000 pages are supported. Big files are the point, not the exception.
- ›Reach a person within a couple of hours, including outside business hours. Every customer talks to the team that builds the product.
- ›Migration is seamless. You upload the same case files you already have; nothing is lost in the move, and your templates come with you.
The team goes out of its way to keep customers satisfied, and that is why more people are migrating off competitors each month. Testing costs nothing and risks nothing: run 500 pages, compare, decide.
Test a file free (up to 500 pages) or book a demo and bring your messiest case.
Questions about these switching stories
Are these real Medrecords AI customers?
Yes. All four accounts come from recorded onboarding, demo, and support calls held in August 2026, and the complaints are quoted near-verbatim from those recordings. Roles, specialties, and case sizes are reported as stated. Names, employers, locations, and all patient information are removed or generalized.
Are the claims about InPractice AI, Dodon AI, ChatGPT, and Claude verified?
They are the customers' own accounts of their experience, quoted from recorded calls in August 2026, and they are attributed that way throughout. They are not assertions drawn from the vendors' published materials. For what each vendor publishes about its own product, see the vendor reviews at /reports/ and the comparison tables at /alternatives/, which cite only published materials.
Does Medrecords AI charge for duplicate pages?
No. Self-service pricing starts at 10 cents per deduplicated page, and duplicate pages are removed and never charged. Suspected duplicates are shown side by side in the workspace so a reviewer can confirm each match instead of trusting a black box.
Is Medrecords AI free to test?
Yes. You can test up to 500 pages of a real case free before paying anything, the same way the four professionals in this report started. Upload a file, check the chronology against the source pages, and only buy credits if the output holds up.
How large a case can Medrecords AI handle?
Cases beyond 50,000 pages are supported. The largest file discussed in these stories is a 69,000-page nursing-home case, staged behind a 9,000-page test case by the consultant who left Dodon AI.
What does de-identified mean in this report?
The professional roles, specialties, file sizes, and quoted complaints are real and unchanged. Everything that could identify a person or firm is removed: names, employers, cities, patient details, and any case fact specific enough to point to one matter.
Ahmed Jemaa is Co-Founder & CEO of Medrecords AI. He runs the onboarding, demo, and support calls these stories come from, and turns what customers say on them into product changes, usually within days. Reach him at aj@medrecords.ai or on LinkedIn.