Innovation or exploitation: the ethical challenges of AI in speech therapy

Artificial intelligence in speech therapy can feel like a superpower: instant practice prompts, real-time speech feedback,
and documentation that doesn’t require a second cup of coffee the size of a goldfish bowl. But here’s the thing about
superpowersthey’re only heroic if they don’t quietly set your house on fire.

Speech therapy (often delivered by speech-language pathologists, or SLPs) is intimate work. It involves voices, faces,
learning histories, disabilities, family dynamics, and sometimes school records. When AI enters that spacelistening,
scoring, prompting, translating, summarizingthe ethical stakes aren’t theoretical. They show up in who gets helped, who
gets harmed, and who profits along the way.

This article explores the ethical challenges of AI in speech therapy with a practical lens: what’s truly innovative, what
veers into exploitation, and how clinicians, schools, and families can demand better guardrails without tossing the tools
into the nearest digital dumpster.

What “AI in speech therapy” actually looks like right now

AI isn’t one productit’s a toolbox. In speech-language pathology, today’s AI-enabled features often fall into a few buckets:

• Speech recognition feedback: apps that listen to practice words/sentences and score accuracy or intelligibility.
• Language support: tools that suggest prompts, generate practice materials, or simplify clinician-written instructions.
• Telepractice enhancements: transcription, automated notes, and session summaries for remote therapy.
• Assistive communication (AAC) improvements: predictive text, smart phrase suggestions, or faster access methods.
• Screening and analytics: systems that detect patterns in speech samples (rate, pauses, phoneme patterns) to flag “possible concerns.”

In many settings, these tools are used alongside professional judgmentnot as a replacement for it. That’s the ideal. The
ethical tension begins when convenience turns into overreach: when “helpful assistant” starts acting like the clinician,
the evaluator, the marketer, or the data broker.

Where the innovation is real (and worth protecting)

1) Access for people who are far from services

AI can support telepractice and at-home practice between sessions. For rural families, homebound patients, or communities
with few SLPs, that can mean fewer missed opportunities and more consistent practice.

2) More personalized, more engaging practice

When AI is well-designed, it can adapt practice difficulty, vary prompts, and give immediate feedback that keeps a child
(or adult) from zoning out after the third repetition. Done right, it’s like having a patient, tireless practice buddy
who never says, “Ugh, again?”

3) Reduced documentation burden

Notes, progress summaries, and session documentation are essentialbut they can also swallow evenings. AI-assisted
transcription and drafting can lighten the load and free time for actual care, if privacy and consent are handled correctly.

4) Better support for multilingual and family communication

Translation and simplified explanations can help families understand goals, home practice, and progressespecially when
clinicians and families don’t share a language. This can improve follow-through and trust, provided the tool is accurate
and clinicians review content before it reaches families.

These benefits are real. And that’s why ethics matter: if we don’t protect the “good” use cases, we risk losing trustand
losing accesswhen the “bad” ones inevitably make headlines.

So where does exploitation creep in?

“Exploitation” doesn’t always look like a villain twirling a mustache while monetizing phonemes. Often, it’s quieter:

• Data extraction: collecting voiceprints, recordings, and behavioral data beyond what therapy requires.
• Overpromising: marketing an app as a diagnosis tool (or a therapist substitute) without clinical validation.
• Paywall pressure: locking core functions behind subscriptions that families feel forced to buy.
• Automation without accountability: using AI outputs as “objective truth” when they’re probabilistic guesses.
• Equity gaps: tools working better for some accents, dialects, or devices than othersthen blaming the user.

The ethical goal isn’t “never use AI.” It’s “never let AI quietly rewrite the rules of care, consent, and fairness.”

The big ethical challenges (and what they look like in real life)

1) Privacy and confidentiality: your client’s voice is not “just data”

Speech therapy data can include identifiable audio/video, health histories, disability status, school records, and family
information. That’s sensitive even before you add the fact that voices can be uniquely identifying.

Ethical red flags include vague privacy policies, unclear data retention, “we may use your data to improve our models”
without meaningful limits, and third-party tracking that links therapy usage to advertising profiles. In healthcare-adjacent
apps, a major trap is assuming that “health-related” automatically means “HIPAA-protected.” Many consumer apps aren’t
covered entities under HIPAA, which changes what rules applyand how exposed users can be.

The practical question is not “Do you have a privacy policy?” (everyone has a privacy policy). The question is: “What
exactly do you collect, what do you share, and what can a user realistically refuse without losing care?”

2) Informed consent: consent can’t be buried in a 37-page scroll

In speech therapy, informed consent isn’t a checkbox; it’s an ongoing conversationespecially when technology is involved.
People deserve to know when AI is listening, what it’s doing with recordings, whether data is used to train models, and
what alternatives exist.

In pediatrics, consent gets extra complicated: parents/guardians may consent, but children still deserve age-appropriate
explanations and respectful boundaries. In schools, privacy obligations can involve education records and student health
information, which raises additional compliance expectations beyond typical consumer-app norms.

A good consent practice sounds like: “Here’s what the tool does, here’s what it stores, here’s who can see it, here’s
what happens if you opt out, and here’s the human plan if the tech fails.”

3) Bias and linguistic diversity: AI can mistake “different” for “wrong”

Speech and language are deeply shaped by dialect, accent, culture, disability, and context. Automated speech recognition
systems have documented accuracy gaps across different speaker groups, and those gaps can translate into unfair therapy
feedback: lower scores, more “errors,” or misleading progress graphs.

Imagine a child who speaks African American English being told by an app that they’re “incorrect” over and overnot
because they can’t communicate, but because the model wasn’t built to understand their dialect. That isn’t just a bug.
It’s a fairness failure.

Ethical practice means treating “model performance across populations” as a safety issue, not a marketing feature. If a
vendor can’t explain how their tool performs across accents, dialects, ages, and disability-related speech differences,
the tool isn’t ready to judge anyone’s speech.

4) Clinical validity and safety: “helpful” feedback can still cause harm

In speech therapy, a tool that gives confident but wrong feedback can frustrate learners, reinforce errors, or create
anxiety about speaking. Even worse, some tools drift into quasi-diagnostic language (“signs of disorder,” “risk level”)
without appropriate clinical testing.

AI models can also change over time (updates, retraining, “improvements”), meaning performance today may differ from
performance next month. That’s not inherently badbut it demands monitoring, evaluation, and clear boundaries on what the
tool is allowed to decide.

A safe posture is: AI can support observation and practice, but a clinician (or qualified team) remains responsible
for interpretation, diagnosis, and treatment decisions.

5) Transparency and accountability: who’s responsible when the AI is wrong?

If a tool mis-scores speech, produces an inaccurate session summary, or generates a misleading recommendation, who owns
that mistake? The clinician? The school? The vendor? Everyone, magically, and therefore no one?

Responsible AI use requires clear accountability pathways:

• Clinician oversight: AI outputs are reviewed before they affect care plans or family communications.
• Auditability: you can trace what the AI used, when it updated, and why it produced an output.
• Human fallback: clients can receive care without being forced into an AI-only channel.

Transparency also means being upfront with clients when AI contributes to messages or records. If families are receiving
AI-generated explanations, they should knowand they should know how to reach a human for clarification.

6) The “consumer app” gap: health-adjacent tools that act like therapy but aren’t regulated like it

Some AI speech tools live in a gray zone: they influence health behaviors, collect sensitive data, and market
“therapy-like” outcomes, yet operate outside traditional healthcare compliance structures.

This is where exploitation risks spike: targeted advertising, “quiet” policy changes, and unclear breach notifications.
The ethical burden falls on clinics and schools to vet vendorsand on vendors to stop treating trust as optional.

7) Equity and the digital divide: innovation that widens gaps isn’t progress

AI tools can expand access, but only if people can actually use them. Families without stable internet, newer devices, or
private spaces for telepractice can get left behind. If AI tools become the default pathway to services (or homework),
inequity can deepen.

Ethical deployment means planning for low-tech alternatives, offering device-agnostic options, and ensuring that “opt out”
doesn’t translate into “lose services.”

8) Professional integrity: AI can’t replace competence

Ethical speech therapy requires skill, judgment, and individualized care. If AI tools are used as shortcutscopy-pasted
goals, generic plans, auto-generated progress notes without reviewthe risk isn’t just “sloppy paperwork.” It’s the
erosion of clinical responsibility.

The most ethical framing is “augmented care”: AI helps with drafts and patterns, while humans make decisions, ask
follow-up questions, and notice the nuance that doesn’t fit a model.

A practical “trust checklist” for AI tools in speech therapy

Whether you’re an SLP, a clinic director, a school administrator, or a parent trying to evaluate an AI speech therapy app,
ask these questions before you adopt anything:

Data and privacy

• What data is collected (audio, video, transcripts, device identifiers, location)?
• Is data used for model training? If yes, can users opt out without losing functionality?
• How long is data kept, and can it be deleted on request?
• Is data shared with third parties (analytics, advertisers, affiliates)?

Clinical quality

• What evidence supports the tool’s claims? Is it validated for the population using it?
• How does the tool handle dialects, accents, and atypical speech patterns?
• What happens when the model is uncertaindoes it communicate uncertainty or pretend it’s sure?

Transparency and control

• Can clinicians review and override AI outputs easily?
• Are updates disclosed, with notes about what changed?
• Is there a human support channel, not just a chatbot sending “thoughts and prayers”?

Fairness and access

• Does the tool work on older devices or low bandwidth?
• Are there accommodations for disability access needs?
• Is the pricing model designed for careor for extracting ongoing payments?

If a vendor can’t answer these clearly, that’s your answer.

What responsible innovation should look like

Ethical AI in speech therapy is not a vibe. It’s a set of practices:

• Minimum necessary data: collect only what’s needed for therapy, not what’s profitable “later.”
• Meaningful consent: plain-language explanations, real choices, and no punishment for opting out.
• Bias testing and reporting: evaluate performance across dialects, ages, and disability-related speech differencesand publish results.
• Clinician-in-the-loop design: AI supports, clinicians decide.
• Security by default: strong safeguards for stored audio/transcripts and careful vendor management.
• Accountability: audit trails, error reporting, and clear responsibility when harm occurs.

In other words: if an AI tool wants to be in the therapy room, it should behave like a professional guestquietly helpful,
transparent about what it’s doing, and respectful of everyone’s privacy.

Experiences from the front lines (composite stories)

To make the ethical challenges less abstract, here are a few composite scenarios based on common real-world dynamics in
clinics and schools. Names and details are generalized, but the dilemmas are very familiar.

1) The “miracle app” that turned practice into shame

A middle schooler uses an AI articulation app at home. The app repeatedly marks their productions as “wrong,” even when
their SLP hears meaningful improvement in session. The student starts refusing practice, saying, “Why bother? The app
says I’m terrible.” The ethical issue isn’t just accuracyit’s harm. An uncalibrated tool undermined motivation, confidence,
and the therapeutic relationship. The fix wasn’t more app time; it was human context, adjusted expectations, and selecting
tools validated for that student’s speech profile.

2) Telepractice made access possible, then privacy got messy

A rural clinic expands services through telepractice. To reduce paperwork, they add an AI transcription feature that drafts
session notes. It works beautifullyuntil a parent asks, “Where do those recordings go?” Suddenly the team realizes the
vendor stores audio for “quality improvement” and the opt-out process is unclear. The clinic pauses the feature, renegotiates
vendor terms, and updates consent language. The lesson: efficiency is great, but “Where does the audio go?” should be asked
before the first session, not after.

3) The school pilot that forgot families exist

A district launches an AI-supported language practice program for K–2 students. Teachers love the dashboards; administrators
love the charts. Families receive almost no explanation beyond “new learning tool.” Later, a caregiver discovers the app
includes third-party analytics and wants deletion options. The district scrambles. The ethical failure wasn’t the pilotit
was the silence. Transparency, consent, and clear parent communication should have been part of the rollout, not an afterthought
stapled onto a PR email.

4) The bilingual family and the “pretty good” translation

An AI tool translates home practice instructions into a family’s preferred language. It’s mostly correctuntil one subtle
mistranslation flips a cue and the family practices the wrong target for two weeks. Nobody did anything “wrong,” but the
system lacked a safety net: clinician review, plain-language confirmation, and a quick feedback loop. AI can help bridge
language gaps, but “pretty good” is not a clinical standard when families rely on instructions to support progress.

5) The billing note that tried to write itself

A busy clinic adopts AI note drafting. A clinician skims, signs, and moves on. Months later, a chart review finds repeated
inaccuracies: wrong goals, mismatched minutes, and generic language that doesn’t reflect skilled services. It wasn’t malicious;
it was fatigue plus automation. Ethical AI use requires “review” to mean reviewnot “glance and hope.” If a system can produce
fluent text, it can also produce fluent mistakes.

Conclusion: keep the human, earn the trust

AI can be an incredible tool in speech therapyexpanding access, improving engagement, and reducing busywork. But because
speech therapy deals with identity, disability, and deeply personal communication, the ethical bar must be high.

The dividing line between innovation and exploitation is simpler than it sounds: innovation improves outcomes while
protecting dignity; exploitation extracts value while shifting risk onto clients, families, and clinicians. If an AI speech
therapy tool can’t clearly explain what it collects, how it performs across diverse speakers, and how humans remain in control,
it hasn’t earned a place in care.

The future doesn’t need “AI vs. SLP.” It needs AI that behaves like a trustworthy assistantcompetent, transparent, and
accountableso clinicians can do what only humans can: listen for meaning, build confidence, and help people be heard.