Can case managers use AI for case notes? A safe-use guide for nonprofits

Can case managers use AI for case notes? Yes, for drafting and polishing, as long as the tool is the right one and the text cannot point to a real person. A free chatbot or AI tool should never see anything that identifies a client. AI built into your case management system, or a business plan covered by a signed business associate agreement (BAA), is the safer route.

The question is coming up everywhere. A national survey of 1,179 social workers, run by the Moritz Center for Societal Impact at UT Austin between October 2025 and February 2026, found that nearly two-thirds use AI in their current role, most often for correspondence, reports, and documentation. About two-thirds also named ethical guidelines, stronger client protections, and more training as the profession’s most pressing AI need. Your staff are likely experimenting already. What they need from you is a clear set of rules.

The quick reference below gives you the short version. The sections after it explain the reasoning so you can adapt it to your programs.

Can case managers use AI to write case notes?

 

Yes, AI for case notes works well for drafting, formatting, and summarizing. The assessment, the judgment calls, and the final sign-off stay with the case manager. AI works best as a fast first-draft partner that you then check against what you saw and heard.

What AI does well

 

Case notes follow predictable structures, which plays to a language model’s strengths. Common uses include:

• Turning your bullet points into a DAP, SOAP, or BIRP structure.
• Tightening wording and removing jargon so the note reads clearly.
• Summarizing a long case history before a case conference.
• Drafting a follow-up plan for you to review and revise.

 

What stays with the case manager

 

Risk determinations, safety plans, mandated reporting decisions, diagnoses, and treatment recommendations all require a trained person who knows the client. A signature on a note means the author takes responsibility for every word, including any sentence the AI wrote. “The AI wrote that part” will not hold up with an auditor, a funder, or a licensing board.

What can go wrong

 

Language models sometimes invent details that sound plausible, such as a referral that never happened or a statement the client never made. They can also smooth over uncertainty, so a polished note may read as more confident than your observations support. Reading every draft against your own memory and records is the safeguard.

Is it safe to put client information into ChatGPT?

 

Not in a free or standard consumer account. These products come without a signed agreement that limits how the vendor uses, stores, or shares what you type, so pasting identifiable client details amounts to disclosing them to an outside company.

Social work’s own technology standards point the same direction. The NASW, ASWB, CSWE, and CSWA Standards for Technology in Social Work Practice say social workers should have the client’s informed consent before sharing confidential information outside the organization, and should limit what they send to what is required.

Retention adds another wrinkle. Deleting a conversation in a consumer app does not guarantee the vendor has erased it, because retention follows each vendor’s terms and any legal obligations that apply to them. Anyone using AI for case notes should read those terms before relying on a delete button.

Common ways client information slips in

 

Most exposure happens by accident rather than by intent. Watch for these routine habits:

• Pasting an email thread that includes names, phone numbers, or a signature block.
• Uploading a spreadsheet export or a screenshot of a client file for “quick help.”
• Feeding a recorded session or transcript into a transcription or summary tool, since a voice recording is itself a biometric identifier.
• Using a browser extension or AI add-on that reads whatever page is open, including your case management system.

 

A short reminder in your staff handbook, plus a pause before every paste, catches most of these.

What paid and enterprise plans change

 

A business associate agreement is a contract in which a vendor promises to protect health information it handles for you under HIPAA rules. OpenAI’s help center lists the products it offers with a BAA, including ChatGPT for Healthcare, ChatGPT for Enterprise with a Regulated Workspace, ChatGPT for Clinicians, and the API with Modified Retention. For the API, both the BAA and a Modified Retention setting are required, and only the listed features are covered. Standard individual and team plans do not appear on that list.

Other AI vendors offer BAAs under their own conditions. Whatever tool you consider, ask for the BAA in writing and confirm it covers the specific feature your staff will use. These lists change, so check the vendor’s current page before you decide.

Does HIPAA apply to your nonprofit?

 

Whether you use AI for case notes or any other task, the answer depends on what your organization does, and many human services nonprofits face other confidentiality rules even when HIPAA does not apply. The safest habit is to treat every client record as sensitive and let your compliance lead tell you which rules govern which program.

Covered entities and business associates

 

HIPAA applies to covered entities, which are health care providers that bill electronically, health plans, and clearinghouses. It also applies to their business associates, the vendors that handle protected health information (PHI) on their behalf. PHI is health information that identifies a person. A nonprofit clinic that bills Medicaid is likely covered, while a food pantry usually is not. Mixed organizations often have one covered program and several that are not.

If you are comparing systems with HIPAA in mind, our guide to HIPAA compliant case management software walks through what to ask vendors.

Rules that apply even when HIPAA does not

 

Several other frameworks can govern the same case note:

• 42 CFR Part 2 protects substance use disorder treatment records. HHS issued a final rule in February 2024 that aligned Part 2 more closely with HIPAA, and the rule still sets its own consent requirements.
• Federal VAWA, FVPSA, and VOCA confidentiality provisions apply to many domestic violence and victim service programs.
• State privacy laws and social work licensing board rules apply in many settings.
• Confidentiality clauses in your funder and government contracts add their own limits.

 

How to find out which rules apply to you

 

Ask your compliance lead or attorney, and bring a list of your funding sources and the types of data each program collects. That list usually settles the question quickly. This guide is general information and not legal advice.

What counts as de-identified information in a case note?

 

HHS recognizes two methods. Safe Harbor removes 18 types of identifiers, and Expert Determination relies on a qualified expert who documents that the risk of identifying anyone is very small. Safe Harbor is a HIPAA standard, but it works as a practical benchmark for any nonprofit deciding what to strip out before using AI for case notes.

In case-note language, the HHS list of identifiers covers these categories:

• Names of the client and of relatives, employers, and household members.
• Anything smaller than a state, including street, city, county, and ZIP code (the first three ZIP digits are allowed only when that area has more than 20,000 people).
• Every date except the year, including service, admission, and birth dates, with ages over 89 grouped as “90 or older.”
• Phone, fax, and email, plus Social Security, medical record, health plan, account, and license numbers.
• Vehicle and device identifiers, web addresses, and IP addresses.
• Biometrics such as voice prints, full-face photos, and any other unique code or characteristic.

 

Why the story itself can identify someone

 

Safe Harbor also requires that you have no actual knowledge that what remains could identify the person. HHS notes that clinical narratives are information rich and can supply enough context to identify a patient. One of its own examples is an occupation listed as “former president of the State University.” A rare event, a small town, or an unusual family situation works the same way, so generalize those details as you write.

The table below shows how to rewrite common details. All examples are invented for illustration.

Removing the name alone rarely does the job. Check dates, places, employers, and schools too, and remember that even properly de-identified data keeps a small chance of being linked back, as HHS acknowledges.

Which AI tools are safe for case notes?

 

When it comes to AI for case notes, the safest option is AI built into your case management system. Next comes an enterprise AI plan with a signed BAA, then a paid plan without a BAA, and finally a free chatbot. The matrix shows what each type of tool can receive.

*Special records include substance use treatment records, domestic violence and victim service records, and records of minors in care.
Personal details means identifiable information that is not health related. Generic wording means text with no client facts.
lass=”yoast-text-mark” />>Note 1: Only if your policy and the vendor’s terms allow it.
>Note 2: Confirm the BAA or agreement covers the specific AI feature, and that it is configured.
>Note 3: Confirm client consent, access roles, and program rules first.

Read “Caution” as a stop-and-check signal. It means the answer depends on your policy, the vendor’s terms, or the client’s consent, so confirm those before pasting anything. “Okay” still means a person reviews and signs the note. This matrix is general information and not legal advice, so check your funder contracts and state rules.

How do you use AI for case notes step by step?

 

Write your own de-identified bullets, ask for a draft in your note format, check it against what you observed, add identifiers back inside your secure system, and sign. The steps below walk through each part of using AI for case notes safely.

1. Confirm your organization’s policy and approved tools

 

Check whether your organization has approved a tool for this purpose. If no policy exists yet, pause and ask your supervisor or compliance lead before using any AI with work product.

2. Write short de-identified bullets from your own observations

 

Jot down what happened in your own words, using the rewrites from the table above. Short bullets keep identifying detail out and give the AI less room to guess.

3. Ask for a draft in your note format

 

Name the format and set boundaries. Our guide to DAP, SOAP, and BIRP note formats shows what each section should contain. A prompt might look like this:

Using the DAP format, turn these bullets into a progress note in plain, neutral language. Do not add any facts that are not in the bullets. Bullets: [your de-identified bullets]

4. Review the draft against what you observed

 

Read every line. Look for invented details, shifts in tone, and any sentence that sounds like a clinical conclusion you did not make. Delete or correct anything you cannot support.

5. Add identifiers back inside your case management system

 

Names, dates, and locations go in after the draft moves into your secure record, never in the chatbot. This keeps the identifying details in a system built to protect them.

6. Sign the note and follow your disclosure rules

 

Your signature confirms the note is accurate. Where your policy requires it, record that AI assisted the draft.

A hypothetical walk-through

 

Imagine a case manager who uses AI for case notes right after a home visit, with three minutes before the next appointment. She jots down: “Client reports improved sleep. Attended two of three scheduled appointments. Discussed transportation options. Plan: confirm bus pass by Friday.” None of those bullets contains a name, date, or address. She asks the AI for a DAP note, reads the result, deletes a sentence that claims the client “seemed hopeful” because she never observed that, then saves the final text in her case management system and adds the client details there.

The whole exchange takes a few minutes, the identifying details never leave the secure system, and the note still reflects what she saw. That is the pattern to repeat.

Tasks AI for case notes should never handle

 

• Risk assessments and safety plans.
• Mandated reporting decisions.
• Diagnoses and treatment recommendations.
• Any content you have not read and verified yourself.

 

Should you use AI built into your case management software instead?

 

Usually yes, because client data stays inside a system already covered by your security controls and agreements. Built-in AI for case notes avoids copying sensitive text into a second tool, and your existing access rules keep working.

What changes with built-in AI

 

Role-based access applies to the AI the same way it applies to the rest of the record. Activity can be logged, and one vendor agreement covers the whole workflow. Staff also stop juggling browser tabs, which removes a common source of accidental pasting.

Questions to ask any vendor

 

• Is our client data used to train AI models?
• Where is data processed and stored?
• Will you sign a BAA or equivalent agreement?
• Which AI features does that agreement cover?
• How long are prompts and outputs kept?
• Can we turn AI off for specific programs or roles?
• Is AI use logged and auditable?
• What happens to our data if we leave?

 

As one example, LiveImpact’s AI runs inside its own AWS infrastructure, client data is not used to train external models, and the platform is HIPAA compliant with a BAA available on request. Its AI features today center on natural-language reporting and conversational data analysis, while writing and signing each note stays with the case manager. You can see how that fits program work in our nonprofit case management software with built-in AI.

What should a policy on AI for case notes include?

 

A short written policy on AI for case notes that covers approved tools, what data can go where, client disclosure, review and signature rules, supervision, and incident response. One page is enough to start.

Disclosure deserves special attention. NASW’s technology standards (Standard 3.01) call for explaining to clients whether and how technology is used to gather, manage, and store their information, which makes intake paperwork the natural home for a plain-language AI disclosure.

• An approved and prohibited tools list, with BAA status recorded for each tool.
• Data classes that spell out what is generic, identifiable, PHI, or specially protected, and which tools may receive each.
• Client disclosure wording for intake paperwork.
• A review and signature requirement, plus supervisor audits of a sample of AI-assisted notes.
• Training for every staff member who uses AI, since two-thirds of surveyed social workers asked for more of it.
• A response plan that tells staff what to do if sensitive information lands in the wrong tool.

 

Sample disclosure wording for intake paperwork

 

Adapt this sentence with your compliance lead: “Our staff may use secure software tools, including AI features inside our case management system, to help draft notes. A staff member reviews and signs every note, and AI does not make decisions about your services.” Clear, short language works better than a paragraph of legal terms.

Revisit the policy every six months. Vendor terms, agreements, and rules change often enough that an annual review can leave gaps.

Frequently asked questions

 

Can case managers use ChatGPT for case notes?

 

Yes, but only with text that cannot identify a client. Free ChatGPT is a consumer product with no business associate agreement, so names, birth dates, addresses, and case details should stay out. Use a de-identified summary, a business plan with a signed BAA, or AI built into your case management system.

Is it a HIPAA violation to put client information into ChatGPT?

 

It can be when your organization is a HIPAA covered entity or business associate and the information is protected health information. Sending it to a vendor without a BAA is generally an impermissible disclosure. Organizations outside HIPAA still face funder, state, and contract confidentiality rules.

What counts as de-identified information?

 

HHS recognizes two methods. Safe Harbor removes 18 identifier types, including names, street addresses, dates other than year, phone numbers, and Social Security numbers. Expert Determination uses a qualified expert to document very low re-identification risk. A distinctive story can still identify someone, so keep narrative details generic.

Do clients need to consent to AI-assisted case notes?

 

Requirements vary by funder, state, and program, so check your contracts and policy. NASW’s technology standards call for explaining to clients how technology is used to gather, manage, and store their information. A one-sentence plain-language disclosure in intake paperwork is a practical place to start.

Can AI write case notes without a person reviewing them?

 

No. AI can invent details, miss context, and state risk more firmly than the facts support. The case manager remains responsible for every word in the record. Compare each draft with your own observations and records before signing, and treat the AI as a drafting assistant only.

What is the safest way to use AI for case notes?

 

The safest option is AI built into the case management system your organization already uses, because client data stays inside a platform covered by your existing security controls and agreements. Next is an enterprise AI plan with a signed BAA. Free chatbots should only see generic, de-identified text.

What is the best first step?

 

Write a one-page policy on AI for case notes and pick one approved tool before anyone experiments on their own. That sequence gives staff a safe default and gives you a record to point to if a funder or auditor asks. If you are still choosing a system, our guide to case notes software covers what to look for, and the LiveImpact team can show you how AI works inside a case management platform built for nonprofits.

Last updated October 2026. This guide is general information and not legal advice.