ASHA Ethics & AI: Informed Consent Explained

AI tools reached speech therapy faster than most policies did, and many SLPs are left guessing what their professional obligations actually are. The good news: ASHA has published specific guidance on generative AI, and the 2023 Code of Ethics already contains the rules that govern it. This article maps that guidance to the practical questions — when you must inform clients, what data you may never enter into a chatbot, and where AI-generated materials fit.

What ASHA's AI guidance actually says

ASHA maintains a dedicated resource, “Generative Artificial Intelligence (AI) for Clinicians in Audiology and Speech-Language Pathology,” with sub-pages on ethical responsibility, security, informed consent, and citing AI. The framing is neither a ban nor a green light:

“AI cannot replace the audiologist, speech-language pathologist, or assistant but could innovate personal and client-centered care when carefully considered.”

ASHA also reminds certified professionals that they have “an ethical responsibility under Principle II of the ASHA Code of Ethics to consider and evaluate any technology that they use in their work,” and cautions that there is limited legislative or regulatory oversight of AI tools today, so extra care falls on the clinician.

The informed consent rule

The core requirement comes from the Code of Ethics itself. Principle I, Rule H states:

“Individuals shall obtain informed consent from the persons they serve about the nature and possible risks and effects of services provided, technology employed, and products dispensed.”

“Technology employed” is the phrase that reaches AI. The Code defines informed consent as an agreement authorizing a proposed course of action “after the communication of adequate information regarding expected outcomes and potential risks,” and notes it may be verbal or written, as required by applicable law or policy. (A small sourcing note: ASHA's AI consent page currently labels this rule under Principle IV, but in the official 2023 Code of Ethics document the informed-consent rule appears as Principle I, Rule H. We cite the Code itself.)

ASHA's AI guidance applies the rule directly:

“Clinicians who use generative AI tools with their clients have an obligation to inform their clients, patients, or students when generative AI is being used in sessions.”

Two practical qualifiers matter. First, whether families must be formally informed can depend on state law — ASHA points clinicians to local requirements. Second, when an employer mandates an AI tool, ASHA recommends directing questions to the employer or to an attorney who knows local privacy law in health care or education. Consent is not a one-time signature: ASHA's worked examples describe verifying signed documents on file, documenting consent, and being ready to answer questions about the tool.

Privacy comes before consent

Consent conversations only matter if the data practices behind them are sound, and here ASHA is unusually blunt:

“Assume that all generative AI tools are not HIPAA compliant, unless vetted by your facility's privacy, legal, or compliance team.”

ASHA's security guidance says sensitive, proprietary, and personally identifiable information should not be entered into generative AI tools at all — including individuals' names and organization names — and notes that a paid account on a public tool like ChatGPT does not change this. For school SLPs, education records sit under FERPA, and ASHA's own worked example recommends de-identifying students before using an AI tool for IEP-related drafting.

The simplest compliant posture is the one we describe across this series: prefer workflows where the tool never needs identifying data. A worksheet generator that works from a target sound, position, age band, and theme has nothing to leak.

The other rules AI touches

Informed consent is the headline, but several other 2023 Code of Ethics rules bear directly on AI use:

Competence (Principle I, Rule A): services must be provided competently — outsourcing thinking to a chatbot does not lower the bar.
Delegation (Principle I, Rules E–F): tasks requiring the profession's unique skills and judgment cannot be delegated — to an assistant or, by extension, to a tool. Responsibility stays with the clinician.
Confidentiality (Principle I, Rules O–P): records and client information must be protected — which is exactly what pasting case notes into a public chatbot violates.
Accuracy of records (Principle I, Rule Q): AI-drafted documentation you sign is your record; it must be accurate.
Technology standards (Principle II, Rule G): technology must be used consistent with accepted professional guidelines.
Accurate public statements (Principle III, Rule E): AI-created content shared with clients or the public must be accurate — ASHA's guidance says to read, verify, and edit generated content before it goes out.

ASHA's citing-AI guidance compresses this into two cardinal rules — be transparent about AI use and cite your sources — and adds the sentence worth pinning above any desk: “the end result is still your professional work—and your responsibility.”

A practical checklist

List every AI tool you use and what data each one sees. Anything touching PHI, PII, or student records needs formal vetting by your facility or district.
De-identify by default. If a task can be done with non-identifying parameters, do it that way.
Inform clients or families when generative AI is used in sessions, and document that you did. Check your state's requirements.
For employer-mandated tools, get the privacy answers in writing from the employer.
Review and edit every AI-generated artifact before it reaches a client, a family, or a chart — accuracy is your ethical obligation, not the vendor's.
Be ready to explain, in plain language, what the tool does and what it does not decide.

Where materials generation fits

Not every AI use carries the same weight. Using AI in a live session or on client records sits at the high-obligation end: consent, documentation, vetted tools. Generating practice materials from non-identifying parameters sits at the low end: no client data enters the system, so the obligations narrow to accuracy and transparency — you still verify the material before a child practices from it.

That is the lane Ga-loo is built for. It asks for a target sound, position, and theme — never a name, diagnosis, or case history — and attaches a deterministic phonetic validation report to every worksheet, so the accuracy check required by Principle III is something you can read rather than assume. Generate a worksheet and see the report, or start from our verified word lists by sound and position.

Bottom line

ASHA's position on AI is consistent with everything else in the Code: the clinician remains the responsible professional. Inform the people you serve when AI is part of their sessions, keep identifying data out of unvetted tools, verify anything AI produces before it carries your signature, and prefer tools designed to need less trust rather than more. For the wider picture, see our guides on what's safe and what's not with AI in speech therapy and how to evaluate whether AI-generated materials can be trusted.

Sources and further reading

This article is educational and is not legal, compliance, or clinical advice.