For speech-language pathologists
Should SLPs use AI scribes and AI note-writing tools?
Published · Updated
The short answer
AI documentation tools can draft session notes, reports and goals, and ambient AI scribes do it by listening to the session. ASHA says SLPs can use them with care: keep identifiable client information out of tools your organization hasn’t vetted, tell clients and get consent when AI is used in sessions, and check every note before signing, because accuracy is your responsibility. Research on AI scribes in speech therapy is thin; trials in medicine show modest time savings and occasional errors.
What these tools do
Two kinds are common. Ambient AI scribes record a clinical encounter and quickly generate a draft note (Lukac et al., 2025). Generative AI tools draft text, synthesize speech, create images or summarize data from a prompt (ASHA).
ASHA says generative AI may help SLPs create therapy materials, correspondence, templates and checklists and, under certain circumstances, assist with documentation (ASHA). Its examples include an ambient listening tool that suggests billing (CPT) codes and an IEP goal-writing tool used with de-identified case information (ASHA). But without the clinician’s expertise, ASHA says, AI can’t be trusted to interpret diagnostic data, individualize care or replace professional judgment (ASHA).
What research with SLPs shows
Research on AI documentation in speech-language pathology is thin. We didn’t find a randomized trial with SLPs. The studies we found:
- A small before-and-after study. 12 rehabilitation professionals in Japan (physical therapists, occupational therapists and SLPs) documented as usual and then with a generative AI system developed by the authors’ organization. Documentation time didn’t change significantly, but time pressure was lower and rated documentation quality was higher with AI (Omon et al., 2025).
- A survey in brain injury rehabilitation. Of 677 physical therapy, occupational therapy and speech-language pathology providers and 270 adults with acquired brain injury, both groups reported little experience with AI and concerns about accuracy and privacy. Most providers were interested in AI for documentation; patients were more hesitant (Covington et al., 2026).
- A survey of SLPs and students. Among 107 respondents (60 clinicians and 47 students), AI tools were used infrequently for clinical purposes, mostly for administrative tasks, and participants called for institutional guidelines and training (Austin et al., 2025).
Evidence from medical settings
- A randomized trial of two scribes. 238 outpatient physicians in 14 specialties were assigned to one of two AI scribes or usual care for two months. One scribe cut time spent writing notes by 9.5% compared with usual care; the other made no significant difference. Both may have eased burnout, which the authors say needs confirming in larger trials, and clinicians reported clinically significant inaccuracies “occasionally” (Lukac et al., 2025).
- A 24-week randomized trial. With 66 practitioners, ambient AI reduced work exhaustion and cut time on notes by about 0.36 hours a day, but didn’t significantly increase professional fulfillment (Afshar et al., 2025).
- A 2026 scoping review of 21 studies found the evidence on real-world impact still limited, mostly short-term and self-reported, and concluded it doesn’t yet support widespread adoption (Lawrence et al., 2026).
These studies were in medical clinics, not speech therapy sessions, so the results may not carry over.
HIPAA: business associate agreements and de-identification
ASHA advises not entering sensitive, proprietary or personally identifiable information into generative AI tools, and assuming a tool isn’t HIPAA compliant unless your facility’s privacy, legal or compliance team has vetted it. That applies even to a paid account on a public tool such as ChatGPT. Don’t upload PHI without express direction from your facility, such as an AI tool built into a vetted electronic health record (ASHA). Details can identify someone without a name: ASHA’s example is a rare disorder that only one client in the practice has, and it advises treating clinic and program names like client names (ASHA).
In the US, a vendor that handles PHI for a HIPAA covered entity is generally a business associate and needs a business associate agreement (BAA); HHS lists an app vendor that provides transcription services as an example (HHS). See the technology checklist for SLPs for what a BAA covers and how FERPA applies in schools.
De-identification. HIPAA allows two methods: Expert Determination and Safe Harbor. Safe Harbor means removing 18 kinds of identifiers of the client and of their relatives, employers or household members, including names, places smaller than a state, all parts of dates except the year, voice prints and full-face photos, and having no actual knowledge that what’s left could identify the person. The identifiers must be removed from free text such as progress notes too (HHS).
Recording consent when a tool listens
ASHA says clinicians must tell clients, patients or students when generative AI is used in sessions and offer informed consent before using any new product. Keep signed consent on file as your state, payer and facility require, and be ready to answer questions about the tool (ASHA). In ASHA’s audiology example, the clinician explains how the audio will be processed and the right to decline; when the patient declines, the clinician uses another approach (ASHA). If your employer requires a tool, ASHA suggests taking questions to the employer or a licensed attorney (ASHA).
Recording laws also vary by state. Federal law allows a person who is a party to a conversation, or who has one party’s consent, to record it, unless the purpose is to commit a criminal or tortious act (18 U.S.C. § 2511). Some states require everyone’s consent: California bars recording a confidential communication without the consent of all parties (Cal. Penal Code § 632), and Washington requires the consent of everyone in a private conversation (RCW 9.73.030).
- Tell families before the session what’s recorded, where the audio goes, how long it’s kept and whether it’s used to train AI.
- Explain that they can say no, and have another plan ready.
- Document the consent, and say in the note when an AI tool was used if your facility requires it.
This is general information, not legal advice.
Accuracy, bias and your responsibility for the record
ASHA says it’s the clinician’s ethical responsibility to make sure any AI-created content is accurate, and that generative AI can produce inaccurate or misleading content. Before signing or sharing an AI-drafted document, read the tool’s terms of use and privacy provisions, understand its limits and intended use, read the draft to check nothing is missing, and correct errors and add specific data (ASHA). Clinicians must review AI-assisted documentation for accuracy and for compliance with legal and payer requirements, including AI-suggested billing codes (ASHA).
Speech-to-text has known weak spots that matter for the people SLPs serve:
- Invented text. In a study of a widely used speech-to-text model, about 1% of transcriptions contained whole phrases or sentences that weren’t in the audio, and 38% of those included harmful content. Hallucinations were more common for speakers with aphasia and linked to longer pauses (Koenecke et al., 2024).
- Racial disparities. Five commercial speech recognition systems had an average word error rate of 0.35 for Black speakers and 0.19 for white speakers (Koenecke et al., 2020).
- Children. Speech recognition is significantly less accurate for children than for adults (Shivakumar & Georgiou, 2020).
Check any numbers in an AI-drafted note, such as percent accuracy or number of trials, against your own session data before you sign.
Questions SLPs ask
Can SLPs use ChatGPT to write session notes?
ASHA advises against entering protected health information or other identifiable details into public generative AI tools, even with a paid account, unless your facility has vetted the tool. You remain responsible for the accuracy of anything you sign.
Do I need consent to use an AI scribe in therapy sessions?
ASHA says clinicians should tell clients when generative AI is used in sessions and obtain and document informed consent. State recording laws vary, and some require every party’s consent.
Is an AI scribe HIPAA compliant?
Not automatically. A vendor handling PHI for a covered entity is generally a business associate and needs a BAA, and ASHA suggests assuming a generative AI tool isn’t HIPAA compliant until your privacy, legal or compliance team has vetted it.
Do AI scribes save time?
In medical trials, results are mixed: one randomized trial found a 9.5% cut in note time with one scribe and no significant change with another. There’s little research with SLPs.
Sources
- Lukac, P. J., Turner, W., Vangala, S., Chin, A. T., Khalili, J., Shih, Y. T., Sarkisian, C., Cheng, E. M., & Mafi, J. N. (2025). Ambient AI scribes in clinical practice: A randomized trial. NEJM AI, 2(12).
- ASHA: Generative artificial intelligence for clinicians.
- ASHA: Artificial intelligence (AI): Considerations for CSD professionals.
- ASHA: Artificial intelligence (AI) considerations: Prioritizing security.
- Omon, K., Sasaki, T., Koshiro, R., Fuchigami, T., & Hamashima, M. (2025). Effects of introducing generative AI in rehabilitation clinical documentation. Cureus, 17(3), e81313.
- Covington, N. V., Friedman, S. E., Klukas, C., Becker, N., Vruwink, O., & O’Brien, K. H. (2026). Open notes in brain injury rehabilitation part 2: Provider and patient preferences for artificial intelligence solutions. Archives of Physical Medicine and Rehabilitation, 107(5), 843–852.
- Austin, J., Benas, K., Caicedo, S., Imiolek, E., Piekutowski, A., & Ghanim, I. (2025). Perceptions of artificial intelligence and ChatGPT by speech-language pathologists and students. American Journal of Speech-Language Pathology, 34(1), 174–200.
- Afshar, M., Baumann, M. R., Resnik, F., Hintzke, J., Sullivan, A. G., Wills, G., et al. (2025). A pragmatic randomized controlled trial of ambient artificial intelligence to improve health practitioner well-being. NEJM AI, 2(12).
- Lawrence, R., Herbert, K., Georghiou, T., Mehta, R., Ng, P. L., Rolewicz, L., Morris, S., & Shand, J. (2026). The use of ambient voice technology (AVT) for clinician-patient consultations in healthcare practice: A rapid systematic scoping review. BMJ Digital Health & AI, 2(1), e000120.
- US Department of Health and Human Services, Office for Civil Rights: Business associates (content last reviewed July 30, 2026).
- US Department of Health and Human Services, Office for Civil Rights: Guidance regarding methods for de-identification of protected health information in accordance with the HIPAA Privacy Rule.
- ASHA: Artificial intelligence (AI) considerations: Obtaining informed consent.
- 18 U.S.C. § 2511: Interception and disclosure of wire, oral, or electronic communications prohibited.
- California Penal Code § 632: Eavesdropping on or recording confidential communications.
- Revised Code of Washington 9.73.030: Intercepting, recording, or divulging private communication: Consent required.
- ASHA: Artificial intelligence (AI) considerations: Ensuring ethical responsibility.
- Koenecke, A., Choi, A. S. G., Mei, K. X., Schellmann, H., & Sloane, M. (2024). Careless Whisper: Speech-to-text hallucination harms. Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency, 1672–1681.
- Koenecke, A., Nam, A., Lake, E., Nudell, J., Quartey, M., Mengesha, Z., Toups, C., Rickford, J. R., Jurafsky, D., & Goel, S. (2020). Racial disparities in automated speech recognition. Proceedings of the National Academy of Sciences, 117(14), 7684–7689.
- Shivakumar, P. G., & Georgiou, P. (2020). Transfer learning from adult to children for speech recognition: Evaluation, analysis and recommendations. Computer Speech & Language, 63, 101077.