AI for Speech Therapy Practices: Scheduling, Intake and Notes

AI for a speech therapy practice works best on the administrative edges of care: scheduling requests, intake forms, reminders and first drafts that a clinician reviews before anything becomes part of a record. It should not make clinical judgments, score evaluations or set therapy goals. A small practice gets the most value by handing an agent one bounded task with written approval rules, then widening the role only after that task runs reliably.

Key takeaways

• Scheduling is the most natural first place to start, because requests are repetitive and every proposed slot can be checked by a person before it is confirmed.
• Intake paperwork is useful for tracking missing forms and summarizing answers for staff, but the agent should never decide eligibility or interpret clinical history.
• Session-note help should mean reminders and drafts built from the clinician's own notes, with the clinician editing and signing in the record system.
• Evaluations stay with the clinician. The agent can prepare the paperwork around an assessment, not the assessment itself.
• Every agent needs scoped access, a named human reviewer and one plain success measure reviewed on a regular schedule.

Where scheduling is the right place to start

Scheduling is rarely the clinical part of a speech therapy practice, but it often takes the most staff attention. Requests arrive by phone, web form, email and messaging apps, frequently outside office hours. Each one has to be checked against clinician availability, session type, room or telehealth link, referral status and the family's preferred windows. Much of that work is matching rules that already exist in someone's head or in a scattered spreadsheet.

An agent can handle the front of that chain. It reads the request, proposes two or three slots that fit the rules the practice has written down, and drafts a reply. It should not place a new evaluation into a clinician's calendar without approval, and it should not promise a start date that depends on a referral nobody has verified. The boundary is simple to state: the agent proposes, and the front desk or clinician commits.

A useful scheduling boundary has four parts:

• The availability rules in writing, including session lengths, which clinician sees which age groups and how the waitlist is ordered
• The slots the agent may propose on its own, and the requests that always go to a staff member
• Cancellation and reschedule handling, with any fee policy applied by staff rather than by the agent
• A handoff path to a person for anything involving distress, urgent concerns or complaints

Illustrative scenario: a parent writes on a Sunday evening asking for a Tuesday slot after school for a child who already attends weekly sessions. The agent checks the clinician's recurring template, finds one open window that fits, and drafts a reply saying the slot is held until the front desk confirms on Monday morning. Nobody has to retype the request, and the family knows a person will confirm it.

Scheduling problems of this kind look similar in other service businesses. The massage clinic pattern of intake, booking and consent in one flow is a good comparison when you map your own requests.

Intake paperwork: what to prepare and what to verify

Intake for a speech therapy practice usually means a bundle: consent forms, a background questionnaire, a developmental history, a referral letter, insurance details, school contacts and sometimes release-of-information forms. The bundle arrives in pieces, and the most common problem is not a missing form but a form that sits in an inbox while the family waits.

An agent can track which items are missing for each new family, send polite reminders on a schedule the practice approves, file returned documents to the correct record, and prepare a one-page intake summary of the questionnaire answers for the clinician to read. Those are useful, bounded tasks. The agent should not decide whether a child qualifies for services, interpret a developmental answer as a diagnosis, mark a form as signed, or send a document to a third party.

This is also where data access matters most. Intake files contain health information and information about minors. Give the agent access only to the intake folder it needs, using a separate account, and log what it reads and sends. The OWASP Authorization Cheat Sheet is a practical reference for thinking about who, or what, may see which records and under which conditions. Messages from families are also untrusted input. The OWASP Top 10 for LLM Applications lists prompt injection as a risk, which means a message containing instructions should be treated as text to classify, never as a command.

Finally, confirm the rules that apply in your jurisdiction and with your compliance advisor before you connect any system. Privacy obligations for health and children's data vary, and an operations decision should not depend on a general article.

Session notes: reminders and drafts, not clinical records

Session notes are where the clinician's professional judgment lives. The helpful role for an agent is narrower than many teams expect. It can remind clinicians which notes are due under the practice's policy, pull the session list for the day, and turn a clinician's own bullet points into a structured draft that follows the practice's template. The clinician then edits, adds what the agent could not know, and signs in the record system.

The agent should not generate observations the clinician did not make, infer progress from a pattern in previous notes, or rewrite a clinical impression into more confident language. A draft that sounds polished but carries a conclusion nobody reached is worse than a blank field, because it invites approval without review.

A short checklist keeps the role honest:

• Every draft is labeled as a draft and linked to the session it came from
• The clinician edits and signs inside the record system, not inside a chat thread
• Reminders name the type of note due, not the child's details, when they travel through general messaging
• Any note mentioning a safety concern is routed to the responsible clinician the same working day
• Staff can see which drafts were changed heavily, since heavy edits are a signal that the template or the input needs work

Evaluations: keep the agent on the paperwork around the assessment

An evaluation is one of the highest-stakes events in a speech therapy practice. Standardized scoring, interpretation, diagnostic statements and goal setting belong to the clinician. An agent has no business in those decisions, and a practice should say so in its written rules.

The administrative shell around an evaluation is different. An agent can hold the evaluation block on the calendar once a clinician approves it, assemble the parent questionnaire, check that the required forms have come back, confirm that materials are scheduled, and prepare a report template with clearly marked empty fields that the clinician completes. Those tasks save time without touching the assessment.

Because AI output can drift or fail in ways that are not obvious on a single day, the practice should review the agent's work on a recurring basis. The NIST AI Risk Management Framework is a useful reference here: it treats AI risk as something managed across the whole life of a system, including monitoring and governance, rather than settled once at setup. A simple recurring check is to compare the agent's draft fields with the clinician's final edits and record where they differ.

A step-by-step plan for the first engagement

A first engagement should be small enough to finish and specific enough to measure. The following sequence fits a small practice with one or two clinicians and a single front desk.

• Step one: write the scheduling rules on one page, covering session types, lengths, clinician coverage and waitlist order
• Step two: choose one intake channel, usually the web form or the main phone-to-email flow, and leave the other channels for later
• Step three: define three lists in writing: what the agent may propose, what it may send without asking, and what always goes to staff
• Step four: give the agent read access to the calendar and write access only to proposed holds, using a separate account with the narrowest scope that works
• Step five: run the agent for two weeks with staff reviewing every outgoing message, and log each correction with its reason
• Step six: choose one success measure in plain words, such as whether each request receives a proposed slot or a clear next step within the same working day, and review it weekly

The corrections logged in step five are the most valuable output of the first two weeks. They show where the written rules are vague, which is usually where the real scheduling policy has been living informally.

Choosing the first task worth handing off

The first task should be repetitive, reversible and reviewed by a person before it reaches a family or a record. Use these questions as a decision guide:

• Does a person check the output before it leaves the practice? If not, the task is not ready.
• Can the action be undone easily, such as releasing a proposed hold? Prefer those tasks.
• Does the task avoid clinical judgment, eligibility decisions and payments? Keep those out of the first phase.
• Can you name the person who owns the rules and the review? If not, the task has no owner yet.
• Does the task require data from a system you cannot yet scope narrowly? Delay it.

Be Human designs and operates this kind of setup as a managed service. A first engagement defines one useful role, the memory and tools it needs, its approval boundaries and an observable success criterion. Ongoing operations can include monitoring, recurring evaluations, integration maintenance, permission review, incident response and controlled improvements, so the scheduling role keeps its rules and access current as the practice changes. The service is remote and available in English and Portuguese.

If you are unsure which of these tasks fits your practice first, the free Be Human AI diagnostic is a practical way to find the first task worth handing to an agent.

Frequently asked questions

Can an AI agent write session notes for a speech therapist? It can remind clinicians which notes are due and draft a structured note from the clinician's own bullet points. The clinician must review, edit and sign the note in the record system. The agent should not add observations, progress judgments or clinical conclusions that the clinician did not make.

Is it safe to let AI handle intake paperwork? Some parts are safe when access is limited. The agent can track missing forms, send approved reminders and summarize answers for staff, but intake files contain health and child information. Access should be scoped, logged and reviewed, and the rules that apply depend on your jurisdiction, so confirm them with your compliance advisor.

What should a small speech therapy practice automate first? Start with one channel and one repetitive, reversible task, such as proposing appointment slots from written scheduling rules. Staff should approve each outgoing message at first, and the logged corrections will show where the rules need to be clearer. Only then should the role grow to other tasks.

Do we need a developer to run this? The first step does not necessarily require one, but someone must own the rules, the access and the review. Managed operations can cover monitoring, recurring evaluations and permission review, so the practice does not have to maintain the connections between its systems on its own.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.