SLPcentralAI note taker for speech therapists: what to look for
Where your clients’ information goes, which agreements matter, and how to tell a draft from a finished note.
An AI note taker for speech therapists drafts your session documentation for you to review, and before you pick one, the questions that matter most are where your clients’ information goes, whether the company signs the agreement your setting needs, and whether the draft comes from your real session data or from what you paste in or record. The writing quality is the easy part. Every tool still produces a draft that you check and sign.
In short: ask which companies receive client information and whether they’ll sign a Business Associate Agreement (health care) or your district’s data privacy agreement (schools). Ask what “de-identified” means in their marketing. Find out whether it records sessions, whether it knows your data or you paste it in, and whether you approve every note before it’s saved. Your district or employer decides what’s approved.
Hi, I’m Ryan, an SLP grad student. I built one of these tools, SLPcentral, so I have a stake in this. The checklist below is the one I’d use on any tool, mine included, and SLPcentral’s answers are near the end so you can hold it to the same list.
What is an AI note taker or AI scribe for SLPs?
The label covers a few different designs, and they raise different questions:
- Session recorders (ambient scribes). They record the therapy session’s audio, transcribe it and draft a note from the transcript. The questions: consent, what happens to the audio, and whether talk in a session captures trial-by-trial data.
- Paste-in drafters. A general chatbot or a note generator: you type or paste a summary, and it drafts a note. The questions: what you pasted, where it went, and whether the numbers survived.
- Data-based drafters. You record data in the app during the session, and the AI drafts from that data. The questions: what the AI receives, under what agreements, and whether it sticks to your numbers.
Where does client information go?
Ask the company to name every service that receives client information: the database host, the transcription service if there is one, and the AI model provider. Then ask where the data is stored, how long each one keeps it, and whether any of them may use it to train their models. A good privacy policy answers all of this in plain words. ASHA’s AI security guidance (opens in a new tab) says not to enter personally identifiable information into generative AI tools, not to upload protected health information unless your facility has directed or vetted the tool, and to assume generative AI tools don’t meet HIPAA’s requirements.
Will the company sign a BAA or your district’s privacy agreement?
In health care, a company that stores or processes protected health information for a covered entity is a business associate, and HIPAA requires a written Business Associate Agreement with it (HHS guidance (opens in a new tab)). Ask two things: will the company sign one with you, and does it have its own BAAs with the services it sends your data to? In schools, HHS says the HIPAA Privacy Rule generally doesn’t apply, because student health information in education records falls under FERPA (HHS (opens in a new tab); FERPA (opens in a new tab)). A district or contractor that bills Medicaid electronically may be different, so ask your district. There, the usual instrument is your district’s student data privacy agreement, and your district decides which tools are approved.
What does “de-identified” actually mean?
Be careful with this word in marketing. Under HIPAA, data counts as de-identified only by one of two routes: an expert determination, or removing 18 kinds of identifiers, which include names, geographic details smaller than a state and every part of a date except the year, with no actual knowledge that what’s left could identify the person (45 CFR §164.514(b) (opens in a new tab)). Swapping a name for a code is a good protection, but on its own it isn’t that standard, because session dates and other details usually still go along. So ask exactly what the AI receives, and whether that’s covered by an agreement anyway.
Does it record your therapy sessions?
If a tool records sessions, ask what consent it expects you to get, whether your district or employer allows recording, whether the audio is stored and for how long, and who can hear it. Also ask whether a transcript of a session can give you trial counts and cue levels. Many SLP goals hinge on correct out of total and the support each response needed, and that’s easier to record as it happens than to reconstruct from talk.
Does it know your data, or do you paste it in?
A tool that starts blank only knows what you give it, so every number passes through a copy step, and every copy step is a chance for an error. A tool that holds your caseload and session data can draft from the actual trials. Either way, check whether it keeps accuracy goals as percentages and frequency goals as counts, and whether it reports the cue level you recorded rather than a plausible one.
Is every note a draft you approve?
ASHA’s AI guidance (opens in a new tab) says AI isn’t a substitute for clinical expertise and that clinicians should review AI-assisted documentation carefully. Look for a tool that labels output as a draft, never saves or submits a note on its own, and makes it easy to edit before you save. Then check each draft the same way every time: numbers, cue levels, invented details and identifiers.
Questions to ask before you buy
- Which companies receive client information, and where is it stored?
- Will you sign a BAA with me, or our district’s data privacy agreement? Do you have BAAs with your own AI and transcription providers?
- Does any provider keep the data, or use it to train models?
- What exactly does the AI receive: names, codes, dates, free text?
- Does the tool record sessions? If it records anything, is the audio stored?
- Does it draft from data I recorded, or from what I paste or say?
- Does it keep accuracy and frequency apart, and report the cue level I recorded?
- Is every note a draft until I save it? Can it ever submit something on its own?
- Can I get the note into my district’s IEP system or my clinic’s records easily?
- What does it cost after the trial, and what happens to my data if I stop paying?
How SLPcentral answers this checklist
Full disclosure again: this is mine, so read it with that in mind. SLPcentral is a data-based drafter for SLPs, mostly school-based:
- It drafts from the trials you tapped. During the session you tap ✓ or ✗ per trial, with a cue level when a response needed support, or count occurrences for frequency goals. Ollie, the app’s AI assistant, drafts SOAP notes and progress reports from that data.
- It doesn’t record your sessions. The only audio is the optional Brain Dump, including a voice message you send Ollie with the microphone: you talk in your own voice, usually after the session. Google Speech-to-Text turns it into text, and the audio is then discarded. It is never stored.
- Saved names are swapped for a code first. The client names you’ve saved are swapped for a code like CLT_X92J before anything goes to the AI. Names stay on your device unless you turn on encrypted name sync. A name you type into a note, or say in a Brain Dump, can still be sent, so keep names out of free text. The code isn’t de-identification: the AI also receives goal wording, trial counts, cue levels, session dates, attendance codes, screener results and anything you type or dictate, which is why it all runs under BAAs. The full list is in our Privacy Policy.
- Signed BAAs. Clinical records are stored, and drafts are made, under signed Business Associate Agreements with Google Cloud and Anthropic, in the United States. Neither may use your data to train its models. HIPAA covered entities, such as practices that bill insurance electronically, can ask us for a BAA, and districts can send us their student data privacy agreement.
- Drafts, never finished notes. Every draft is labeled as one, and nothing goes into a client’s notes until you review it, edit it and tap Save. You copy the final note into your district’s or clinic’s system.
Where it’s weaker: if you want a full transcript of every session, SLPcentral doesn’t do that, by design. It isn’t an EHR or an IEP system, so the final note still gets copied into yours. Price: licensed SLPs start with a 14-day free trial, then $19.99 a month or $149.99 a year (pricing). How it compares with other SLP software is on our comparison page, and there’s more about Ollie in Meet Ollie. If you’re weighing a general chatbot instead, read Can SLPs use ChatGPT for SOAP notes? first.
Frequently asked questions
How do I choose an AI note taker for speech therapy?
Start with privacy and agreements, not writing quality. Ask which companies receive client information, whether the company will sign a BAA or your district’s data privacy agreement, what the AI receives, whether sessions are recorded, whether it drafts from your recorded data, and whether you approve every note before it’s saved.
Can SLPs use AI scribes?
Often yes, if your district or employer approves the tool and you review every note. ASHA’s guidance says to keep personally identifiable information out of generative AI tools, and protected health information out unless your facility has vetted the tool, and that AI isn’t a substitute for clinical expertise.
Does an AI note taker need a BAA?
In health care, yes, when it stores or processes protected health information for a covered entity such as a practice that bills insurance electronically. In schools, student records usually fall under FERPA, and districts use a student data privacy agreement instead. Ask which applies to you.
Is a client code the same as de-identified data?
Not by itself. Under HIPAA, de-identification means an expert determination or removing 18 kinds of identifiers, including names and every part of a date except the year. A code in place of a name is a useful protection, but dates and other details usually still go along.
Can an AI note taker write my notes for me?
It can draft them. You’re still responsible for the note, so check every number against your data, check cue levels and anything the draft added, and save it yourself.
Sources
- ASHA, AI Considerations: Prioritizing Security (opens in a new tab) (keep PII out of generative AI tools, and PHI out unless your facility has vetted or directed the tool; assume tools don’t meet HIPAA’s requirements)
- ASHA, AI Considerations for CSD Professionals (opens in a new tab) (review AI-assisted documentation; AI isn’t a substitute for expertise)
- U.S. Department of Health and Human Services, Guidance on HIPAA & Cloud Computing (opens in a new tab) (business associates and BAAs)
- U.S. Department of Health and Human Services, Does the HIPAA Privacy Rule apply to an elementary or secondary school? (opens in a new tab)
- U.S. Department of Education, Student Privacy Policy Office, What is FERPA? (opens in a new tab)
- HIPAA Privacy Rule, 45 CFR §164.514(b) (opens in a new tab) (the two ways health information counts as de-identified)
- SLPcentral Privacy Policy (what is sent to Google and Anthropic, Brain Dump audio, BAAs)
Drafted with AI help from the sources listed, then reviewed and edited by Ryan Lillard. No client information is used in this article. This article is general information, not legal advice; follow your district’s or employer’s policies. Ollie is an AI assistant; he drafts, you decide.
