HIPAA-Compliant AI for Skilled Nursing Facilities
Hathr.AI is a HIPAA-compliant AI platform that skilled nursing teams use on real resident data — MDS support, PDPM documentation review, survey and F-tag preparation, denial appeals, and everyday charting cleanup.
It runs Government approved versions of Anthropic's Claude models inside AWS GovCloud under a FedRAMP High authorization boundary, with a signed BAA on every account, zero data retention, and no seat minimums. $47 per user per month, 7-day free trial.
The record supports the billed clinical category: IV medication administration is documented on days 2 through 5 of the covered stay, with physician orders and MAR entries present.
Section GG entries and the nursing function score are consistent across the assessment window and align with the HIPPS code submitted.
Illustrative example. A clinician always reviews and signs.
What is Hathr.AI for skilled nursing facilities?
Hathr.AI is HIPAA-compliant AI software that skilled nursing facilities use on real resident data — checking whether a chart supports the PDPM classification that was billed, preparing for survey and Five-Claim Probe reviews, drafting Medicare denial appeals, and finding the Section GG or MDS detail missing from a stack of charts. It runs Anthropic's Claude inside AWS GovCloud under a FedRAMP High authorization boundary, with a signed BAA on every account and zero data retention.
Skilled nursing runs on documentation. Care is delivered by clinical staff, but it is the record — the specific entries that tie a resident's condition, services, and function score to the rate that was billed — that determines whether a claim survives a Five-Claim Probe, an Additional Documentation Request, or a CMS data validation review. In 2023, 79.1% of SNF improper payments were attributed to insufficient documentation alone. The care was delivered. The chart did not prove it.
Most AI vendors selling into skilled nursing quote roughly four hours saved per nurse per shift. That figure traces back to a case study at an academic acute-care medical center, not a skilled nursing facility, and it is about sixteen times larger than the best available evidence from long-term care. The strongest peer-reviewed evidence — a 2026 time-motion study in the Journal of Medical Internet Research covering 52 registered nurses across 14 long-term care facilities and 770 observed hours — found documentation time per morning shift fell by an adjusted mean of 15 minutes, about a 28% reduction from a 53.91-minute baseline.
That study also found where the time actually moves. Reductions were concentrated in active documentation (−13.28 minutes) and note-taking (−2.38 minutes), while reviewing entries (+0.64) and information retrieval (+1.28) went slightly up — a shift from producing text to checking and retrieving it. Hathr.AI is built for that second layer: the reviewing and retrieval work that grows when charting gets faster.
New to how Medicare pays for skilled nursing?
Start with our plain-English guide to PDPM documentation requirements.
Why other AI tools (even HIPAA Compliant AI tools) often fail a skilled nursing compliance review
Consumer ChatGPT, Claude, and even other "HIPAA Compliant" AI tools are often not HIPAA-compliant for skilled nursing use because they don't include a BAA by default, may retain your inputs, can train on your conversations, and aren't compliant with Medicare or Medicaid requirements to host CMS information. Entering PHI into them is a HIPAA violation or violation of CMS standards that require appropriate FedRAMP infrastructure, and following NIST data protection standards.
This is where Hathr.AI pulls ahead — FedRAMP High Infrastructure, NIST Standards, Certified Federal Contractor, Government approved LLMs. Things other teams won't, and frankly can't, offer.
There's a persistent myth that "the AI company is big, so it must be safe." Size isn't the standard — the Business Associate Agreement is. Under HIPAA, any vendor that touches PHI on your behalf must sign a BAA that makes them legally accountable for protecting that data. Without one, the tool is simply off-limits for anything involving a patient chart.
It's true that the landscape has shifted: both OpenAI and Anthropic now offer BAA-backed paths for eligible enterprise customers. That's real progress — but it changes where the differentiation lives, not whether it exists. For a skilled nursing facility, the questions a compliance officer actually asks are more specific:
Is there a BAA on this account — or only on a higher enterprise tier?
With Hathr, a BAA is included on every account. There is no seat minimum or enterprise gate to clear before you're covered.
Where does the data physically run?
Hathr operates inside AWS GovCloud, a FedRAMP High authorized environment with data and processing only happening within the confines of the United States — the tier built for the most sensitive government and healthcare workloads, not a general commercial cloud region.
What happens to inputs after a session?
Hathr retains no customer data for training and does not use your prompts or documents to improve any model. Your patients' information leaves no residue.
For facilities operating under government contracts, VA relationships, active state survey scrutiny, or simply a risk-averse board, that infrastructure depth — GovCloud, FedRAMP High, zero retention, BAA-by-default — is the difference between "technically has a BAA somewhere" and "approved for use on real PHI today."
See the full breakdown: Is Claude HIPAA compliant? · Is ChatGPT HIPAA compliant? · The risks of using ChatGPT for clinical documentation
Why documentation is the skilled nursing facility's core financial risk
Skilled nursing leads every other Medicare care setting in documentation errors. The national SNF improper payment rate rose from 7.79% in 2021 to 17.2% in 2024 — roughly $5.9 billion — and in 2023, 79.1% of SNF improper payments were attributed to insufficient documentation alone. HIPAA-compliant AI that reviews the record before a claim goes out works on the error category that actually drives the money.
Documentation burden in skilled nursing is not a soft problem — it lands on the claim. In November 2025 the HHS Office of Inspector General reported that a single facility, Pinnacle Multicare Nursing and Rehabilitation Center, failed Medicare requirements on 99 of 100 sampled claims: $1.1 million in the sample, extrapolated to an estimated $31.2 million in overpayments. OIG cited records that did not support the assigned reimbursement rate code, services for residents who did not require skilled care, and claims that simply did not meet documentation requirements.
In January 2026 CMS launched the SNF Value-Based Purchasing Data Validation Process, with randomly selected facilities receiving notices through iQIES, so MDS accuracy is now actively cross-checked against the clinical record. The money is moving the other way on admissions as well: OIG found in June 2026 that Medicare Advantage organizations denied prior authorization for SNF-level care from nursing home residents 40% of the time, versus 11% for other enrollees — and overturned nearly all of those denials on appeal. Facilities that appeal well get paid; facilities without the staff hours to appeal do not. The trap is templating your way out of the time problem by reusing generic language to move faster, which is exactly what reviewers are trained to flag. Hathr.AI produces individualized, evidence-grounded review faster than a nurse working from a blank page, so you gain time and defensibility instead of choosing between them. It is not that notes take too long to type. It is that the chart must survive a reviewer who was not there, and proving that at scale is a reading-and-checking problem.
What skilled nursing teams use Hathr.AI for
SNF teams use Hathr.AI to review documentation against Medicare criteria before it is billed or surveyed — checking PDPM support, drafting and appealing denials, preparing for surveys and F-tag citations, and analyzing full resident records in a single pass. It is not an ambient scribe and does not replace your EHR; it is the review and analysis layer that sits on top of whatever charting system you already run.
MDS and PDPM documentation support
Hathr.AI reads the supporting record and checks whether the clinical documentation actually substantiates what was coded — Section GG scoring, primary diagnosis mapping, NTA comorbidity capture — before the claim is submitted.
This is the exact failure mode OIG cited at Pinnacle, where the record did not support the assigned rate code. Upload the chart alongside the relevant RAI Manual guidance and ask whether what the clinicians wrote supports this resident's coded classification. An AI suggestion that raises your case-mix index without documentation behind it is, from an auditor's perspective, indistinguishable from a human coding error. Hathr.AI is deliberately positioned to work the defensive direction: is this supported, rather than how do we code higher.
See the guides on PDPM and the MDS Section GG.
Survey readiness and F-tag preparation
Run current policies, care plans, and incident documentation against the F-tag requirements you are most frequently cited on. Hathr.AI processes a large volume of resident records in one pass and surfaces where documentation is thin before a surveyor finds it.
Survey readiness is not a binder — it is whether the record supports what the care plan claims. Hathr.AI helps Directors of Nursing and QAPI leads confirm the evidence a surveyor will look for is actually present, and surface the gaps while there is still time to document them.
Dig into the requirements: F-tags explained · survey readiness checklist · QAPI · the MDS coordinator role · the RAI Manual.
Denial appeals and ADR response
Given the 40% Medicare Advantage denial rate on SNF admissions — and the near-total overturn rate on appeal — the constraint on appeal revenue is usually staff hours, not merit.
Hathr.AI drafts appeal letters grounded in the actual chart, maps the argument to the denial reason, and assembles the supporting documentation reference for each of the five appeal levels.
Billing accuracy before submission
Check claims against consolidated billing rules, verify the right CPT codes for Part A versus Part B, and confirm ABN issuance was appropriate — before the claim reaches the MAC rather than after a denial comes back.
Most SNF billing errors surface after the money is already at risk. Moving the check earlier turns a denial-and-appeal cycle into a correction, and it is the cheapest place in the revenue cycle to spend attention.
See the guide to SNF billing and consolidated billing.
PBJ, staffing, and restorative program documentation
Reconcile staffing documentation against PBJ submission requirements, and check that restorative nursing programs are documented to the O0500 standard that PDPM requires.
Staffing and restorative documentation are quiet sources of both citation risk and lost PDPM value. Hathr.AI reads the submission requirements alongside your own records and reports where the two do not line up.
PBJ reporting requirements and restorative nursing programs.
Everyday chart review and QA
Summarize a long resident record, extract a specific finding across an entire stay, or standardize a note against your own facility's template. Hathr.AI reads single documents exceeding 500,000 words in one pass and runs handwriting OCR on scanned and hand-annotated charts.
A 400-page record split into twelve uploads makes a finding that spans page 30 and page 380 invisible. Reading the whole record in one pass is what makes a cross-cutting question answerable — and handwriting OCR is something most HIPAA-compliant tools skip entirely.
What Hathr.AI is not
Hathr.AI is not an ambient scribe and not an EHR. Hathr.AI's Enterprise tool allows you to integrate direclty into PointClickCare or MatrixCare, while our webapp allows you to copy and paste or export information wherever you need to. Hathr.AI doesn't replace your MDS coordinator, but it helps them move at a speed they couldn't before - Hathr.AI reads what your charting system already produced and checks it against Medicare criteria before the claim goes out or the surveyor arrives — which is why it works the same way across a multi-site operator running different systems in different buildings.
How Hathr.AI works
Sign up, get your BAA (included on every account), and start working in a secure chat interface. Upload or paste a resident record, ask in plain English (or 50 other separate languages including Spanish, Chinese, and languages from across the world), and Hathr.AI responds using Claude models inside AWS GovCloud. Nothing is retained for training.
Hathr.AI is deliberately simple to adopt — our webapp is set up for teams to get fast access and start using secure, HIPAA Compliant AI today. No lengthy IT integration and no EHR project to get value on day one. The research is also clear that AI time savings are real, modest, and unevenly distributed, so where you point it matters: start with the reviewing layer rather than bedside charting, ground it in your own criteria, assign it first to the roles where documentation risk concentrates, and measure clean-claim and appeal-overturn rates rather than minutes saved.
- 1
Create an account and activate your BAA.
Coverage is included on every account, so you're compliant from the first session — no enterprise procurement cycle required.
- 2
Bring your context.
Paste text, upload documents, or reference your own policies and coverage determinations. Hathr.AI does retrieval across up to 100 files at once, so answers are grounded in your material — your policies, your MAC's local coverage determinations, the current RAI Manual, your own denial history — rather than a model's general training data.
- 3
Ask in plain language and review the output.
Check whether a chart supports the billed PDPM classification, draft a denial appeal, prepare for a Five-Claim Probe, or find the Section GG detail missing across 40 charts. Licensed staff always review and sign — Hathr.AI supports clinical judgment, it never replaces it.
Everything runs inside the FedRAMP High authorized GovCloud environment, and none of your inputs are used to train any model.
Hathr.AI vs. consumer AI vs. generic HIPAA chatbots
Consumer ChatGPT and free Claude are fast but carry no BAA by default and shouldn't touch PHI. Generic "HIPAA chatbots" may include a BAA but often run on standard commercial clouds with thin capability. Hathr.AI pairs frontier Claude models with GovCloud, FedRAMP High, zero retention, and a BAA on every account — and adds the skilled nursing workflows that matter before a claim goes out or a surveyor arrives.
| Consumer ChatGPT / free Claude | Generic HIPAA chatbot | Hathr.AI | |
|---|---|---|---|
| BAA | Not by default | Sometimes | On every account |
| Hosting | Commercial cloud | Commercial cloud | AWS GovCloud (FedRAMP High) |
| Data retention | May retain / train on inputs | Varies | Zero retention; no training on your data |
| Model quality | Frontier | Often weak & lots of offered models | Frontier Models + Optimization |
| Grounding on your docs (RAG) | Limited | Varies | Yes — answers from your material |
| Skilled nursing workflows | None | None | PDPM, Section GG, F-tags, appeals, PBJ |
| Safe for PHI | ×No | !Depends | ✓Yes |
The takeaway: capability alone isn't enough (consumer tools have it but can't touch PHI), and compliance alone isn't enough (generic chatbots have a BAA but weak models and no skilled nursing fluency). Hathr.AI is built to satisfy both the MDS coordinator who wants a genuinely capable assistant and the compliance officer who has to sign off on it.
Security and compliance built for vendor review
Hathr.AI runs on AWS GovCloud in a FedRAMP High authorized environment, includes a BAA on every account, retains no customer data for training, and is a certified Federal Contractor at the Federal, State, and Local Level Service-Disabled Veteran-Owned Small Business (SDVOSB). Hathr.AI supports thousands of customers from solo practitioners, to large, multi-state networks, the Centers for Medicare and Medicaid (CMS), and the Dept of Health and Human Services. It's designed to clear a skilled nursing facility's security and compliance review, not just a marketing claim.
Skilled nursing procurement increasingly runs through a real security review before any long-term vendor contract is signed — and rightly so. Hathr.AI is built to answer those questions directly.
If your compliance officer or IT reviewer has a security questionnaire, that's exactly the conversation Hathr is built to have.
Talk to our team about a security reviewBusiness Associate Agreement
Included on every account, establishing Hathr's legal accountability for PHI under HIPAA.
AWS GovCloud, FedRAMP High authorized environment
The infrastructure tier for the most sensitive regulated workloads, isolated from general commercial cloud regions.
Real, Government Approved Standards, not commercial Check boxes
Hathr.AI follows real, government standards - NIST800-53, NIST800-171, FedRAMP High Infrastructure that run of the mill commercial teams can't even get access to.
Hathr.AI's security is validated by the Department of Defense and other federal organizations, not random commercial companies looking to do an audit for cash.Zero data retention
Your prompts and documents are not stored for training and are not used to improve any model.
Government Approved Frontier models under the hood
Unlike teams who use Commercial Servers for your data, and regular Large Language Models, with Hathr.AI you're not trading capability for compliance; you get both.
SDVOSB
Hathr.AI's employees maintain US Government Security Clearances, and are National Security Professionals who have worked across the world, and supporting US Government Security needs - it's what's pulled us to serve Healthcare, and other regulated Industries. Our business is a Service-Disabled Veteran-Owned Small Business, an advantage for facilities and operators with local, state, or federal government contracting, set-aside, or supplier-diversity considerations.
Who it's for
Hathr.AI fits the people who own skilled nursing documentation and its consequences — from a single building to a multi-state operator:
MDS coordinators
Check that Section GG, diagnosis mapping, and NTA capture are supported by the record before the assessment is submitted.
Administrators and multi-site operators
Reduce documentation risk, protect PDPM revenue, and adopt AI without creating a compliance liability.
Compliance officers and QAPI leads
A vendor with a BAA, GovCloud hosting, and zero retention you can actually approve.
Directors of Nursing and clinical leads
Run policies and care plans against the F-tags you are most often cited on, and close the gaps before a surveyor does.
Business office, billing, and denial specialists
Check claims before submission and draft appeals grounded in the actual chart. Cleaner documentation upstream means fewer denials downstream.
Whether you run one building or a multi-state portfolio, Hathr.AI scales from a single MDS coordinator's daily review to facility-wide documentation quality — with the same BAA-on-every-account coverage throughout.
Learn more: the skilled nursing documentation library
The Hathr team maintains an in-depth, continually updated library of PDPM, MDS, survey, and billing guides for skilled nursing. Start with the pillar and branch into the topics you need:
Frequently asked questions
What AI documentation platforms are HIPAA-compliant for skilled nursing use?
A platform is only HIPAA-compliant for SNF use if the vendor will sign a business associate agreement covering your resident data. Hathr.AI signs a BAA on every account within 24 hours, on every plan, with no seat minimum — and runs in AWS GovCloud under a FedRAMP High authorization boundary with zero data retention. Several vendors offer BAA-backed paths; the differences are in hosting environment, retention terms, and whether the BAA requires an enterprise contract.
Can I use ChatGPT or Claude to write MDS or resident documentation?
Only under a signed business associate agreement covering that specific product and account. Both OpenAI and Anthropic now offer BAA-backed enterprise paths, but the consumer versions of ChatGPT and Claude.ai are not covered by default. Using an uncovered tool with resident PHI is a disclosure to a business associate without an agreement in place.
What HIPAA-compliant tools help with repetitive medical documentation in a SNF?
Look for three properties: a BAA that covers your actual plan, the ability to process a full resident record without splitting it (Hathr.AI handles single documents over 500,000 words in one pass), and retrieval over your own policy and coverage documents so answers reflect your MAC's rules rather than general training data.
Will AI-assisted documentation increase our audit risk?
It depends entirely on direction of use. An AI that suggests higher-acuity coding without documentation to support it recreates exactly the failure OIG cited at Pinnacle Multicare, where 99 of 100 sampled claims did not meet requirements. Used defensively — checking whether the record supports what was already coded — AI reduces the exposure that produced a 17.2% national SNF improper payment rate, of which 79.1% was insufficient documentation.
Is Hathr.AI's BAA included on every plan, or only enterprise?
Every plan. A single MDS coordinator on one seat gets the same BAA as a multi-state operator, signed within 24 hours. There is no seat minimum and no enterprise gate on compliance terms.
Where is our resident data hosted, and is it used to train AI models?
Data is processed in AWS GovCloud (US) inside a FedRAMP High authorization boundary. Hathr.AI has zero data retention and does not use customer data to train models — a contractual term, not a settings toggle. Prompts and outputs are fully audit-logged for your compliance file.
How much does Hathr.AI cost?
$47 per user per month with unlimited use, no setup fees, and no per-document or per-page charges, with a 7-day free trial. An enterprise tier adds SSO, custom retention, and dedicated support. See current pricing on our pricing page.
How much time does AI actually save on nursing documentation?
The best peer-reviewed evidence from long-term care — a 2026 JMIR time-motion study of 52 nurses across 14 facilities and 770 observed hours — found documentation time fell by an adjusted mean of 15 minutes per morning shift, about a 28% reduction from a 53.91-minute baseline. Savings vary widely by individual, and roughly 17% of nurses in a related study reported saving no time at all. Claims of four hours saved per shift come from acute-care settings, not skilled nursing.
Is Hathr.AI SNF software or an EHR, and does it integrate with PointClickCare or MatrixCare?
Hathr.AI is a HIPAA-compliant AI review layer, not an EHR. Skilled nursing facility software generally divides into the clinical system of record — nursing home charting software and MDS software such as PointClickCare or MatrixCare — and the tools that work around it. Hathr.AI is the second kind: it is not an embedded EHR module and does not write into PointClickCare or MatrixCare. You bring documents and records into Hathr.AI for review, analysis, drafting, and appeal preparation. That independence is deliberate: the review layer is not limited to what one EHR vendor's roadmap supports, and it works the same way across a multi-site operator running different systems in different buildings.
Does Hathr.AI replace our MDS coordinator?
No. It reviews, drafts, checks, and summarizes. The MDS coordinator remains the coder of record and the licensed staff who sign assessments retain full authority. The value is in giving them a second read of every chart before it is billed or surveyed — not in removing them from the process.
Give your skilled nursing team frontier AI they can actually use on PHI & CMS Data
Start checking PDPM support, preparing for survey, and appealing denials today — with a BAA on every account.