AI in Occupational Therapy: Funding Letters and Parent Updates

A private occupational therapy practice can use AI to turn therapist-approved, de-identified goals and observations into a draft funding letter and a separate parent update. First map every recommendation to an actual assessment record, then check the funder's current wording and evidence rules. The treating therapist must verify, amend and sign both outputs before they are shared.

The boundary is professional judgement. AI can reorder notes, expose gaps and simplify sentences; it cannot decide what support a child needs, whether progress is clinically meaningful or whether a funder will approve a request. The Royal College of Occupational Therapists expects practitioners to recognise errors and bias, protect confidentiality, use AI transparently and take responsibility for the decisions they make in practice.

Follow me on Instagram@sagnikteaches

Start with a goal-to-evidence sheet, not a blank prompt

A funding request often fails on the connection between a goal, the observed barrier, the intervention tried and the support recommended. Build a compact evidence sheet from the existing record before any drafting. Give each goal an identifier. For each one, note the functional task, baseline observation, relevant context, intervention, observed change, remaining barrier and therapist-approved recommendation. If a field is missing, write "not recorded" rather than asking the model to fill it.

Connect on LinkedInSagnik Bhattacharya

For an illustrative private practice, goal G1 is for a child to fasten a school coat independently. The record shows success on two of five observed attempts at baseline and four of five after six sessions, with a visual cue still needed. These are fictional figures for teaching the method, not a client story or an effectiveness claim. The therapist can describe the functional change and the remaining cue requirement. The model must not turn four of five into "independent in all settings" or infer a diagnosis from the task.

Subscribe on YouTube@sagnikteaches

Goal G2 might be a classroom transition routine. The therapist's record has a parent report that mornings are harder, but no direct school observation. That is valuable context, yet it is a different evidence type. Label it "parent report" and avoid writing "school staff observed". If the letter needs school evidence, the therapist should obtain and review it through the normal consent route. A polished paragraph cannot replace a missing source.

Use the approved documentation process for storage. The OT documentation audit checklist is relevant once drafts enter the record. RCOT's revised professional standards, announced in July 2026, add guidance on responsible AI use: practitioners should understand how AI systems generate outputs, recognise errors and bias, protect confidentiality and take responsibility for professional decisions. That means the evidence sheet needs a named author and date, and the final letter needs a traceable review, not merely an AI-generated note that it was "checked".

Remove identifiers while preserving the logic of the case

De-identification is a deliberate step, not a find-and-replace operation. Remove names, exact dates of birth, addresses, school or employer names, contact details, record numbers and unusual events that are unnecessary to the writing task. Keep only the functional observations and goal structure needed for a draft. A child's age range may be relevant to the wording, but an exact birthday rarely is. Keep the re-identification key inside the practice's controlled record, away from the AI prompt.

For example, "[first name], aged 8, attends [school] and lives with [specific relatives]" is still too much for a generic drafting prompt if those details do not affect the argument. A safer abstract input is "school-aged child; goal G1; coat fastening observed at two of five attempts initially and four of five after six sessions; visual cue still required." Even this needs the practice's own approval and data assessment. A rare combination of circumstances can remain identifiable despite a missing name.

If the practice has a contracted AI system approved for identifiable health data, that may allow a different data route. The owner should verify the precise service, agreement, retention and access settings with a qualified adviser, rather than assume all accounts of a vendor behave the same way. For a first pilot, fictional or properly de-identified cases are easier to govern. The vendor security questions help when evaluating a tool that touches records.

A third example concerns a draft that quotes a parent verbatim from a session note. The quote may include family circumstances irrelevant to the funding request. The therapist should decide whether a quote is necessary, whether it was recorded accurately, and whether the intended recipient is authorised to receive it. If not, replace it with a factual, attributed summary or remove it. AI's ability to preserve a quote is not a reason to disclose it.

Draft the funder version against its current criteria

Fetch the funder's current form or guidance before composing the letter. Check required headings, decision criteria, evidence period, professional credentials, signature requirements and any prescribed terminology. Some funders ask for goals and functional impact; others want previous interventions, alternatives considered or a costed service schedule. Do not use the same template indiscriminately. The therapist should make the recommendation first and ask AI to make the reasoning readable.

Where the NDIS is the funding route, its plan reassessment report guidance for allied health providers works as a ready-made review list. It asks whether the report summarises the supports provided, evidences previous therapies trialled, measures how functional ability is progressing, documents barriers and explains any outcome not achieved. It then asks for clear, evidence-based recommendations linked to the participant's therapy outcomes and goals, with service details such as type of service, approach, duration, frequency, intended outcome and the justification. Read the current version of that page before each submission, because scheme guidance changes. An insurer may use different contractual language. Check the current destination before reusing the AI draft; do not claim that one letter meets every scheme's requirements.

Draft a funding letter from the de-identified evidence below.
Use only the supplied facts and the therapist-approved recommendation.
For each requested support, connect functional goal, baseline,
observed change, remaining barrier, proposed support and rationale.
Label parent report separately from direct observation.
Mark absent facts as [THERAPIST TO COMPLETE].
Do not invent a diagnosis, prognosis, eligibility finding,
price, service frequency or professional opinion.
Return a source map: sentence -> evidence-sheet field.
[Paste approved evidence sheet and current funder headings.]

The source map helps the therapist find unsupported prose. It is especially useful for adjectives such as "severe", "substantial" and "essential". A model may add them to strengthen a request even when the evidence sheet contains no basis. The therapist should replace them with measured observations and their own professional reasoning, or remove them. Persuasive language cannot substitute for evidence and may undermine credibility if challenged.

For an insurer, the first review is often whether the letter answers the exact question asked. If the insurer requests the reason a particular support is needed now, a long history of every session may bury the answer. The therapist can add a short "request" paragraph followed by the relevant observation and rationale. If the insurer requests a costed schedule, the administrative team should supply the approved fee and frequency from its current records; the model should not estimate a price or choose a billing code. Where policy interpretation is disputed, obtain advice from the appropriate qualified person.

An illustrative insurer asks why the existing home programme alone is insufficient. The record shows the family completed practice on four of six recorded weeks, but the child's coat-fastening task still required a visual cue in clinic. The letter may accurately describe both facts and the therapist's reason for further support. It must not claim the family "failed" the programme or that further sessions are guaranteed to remove the cue. This distinction protects the family relationship and keeps the funding argument tied to evidence.

An illustrative weak sentence reads, "The child urgently requires twelve months of weekly therapy to become fully independent." The evidence sheet may only support that a visual cue remains after six sessions and that the therapist recommends a further review period. A defensible replacement would state the observed task, the remaining cue and the therapist's actual recommendation, with the proposed frequency and review date only if documented. The qualified therapist decides the recommendation; AI edits its expression.

Walk one fictional funding request through review

Consider an illustrative three-therapist practice. One therapist has six sessions of records for a school-aged child with two goals. G1 is coat fastening, described above. G2 is following a three-step packing routine: the baseline record shows one of three steps completed without cueing, while the latest observed session shows two of three. The therapist has documented home practice and a parent report that the routine is still difficult on busy mornings. These invented figures illustrate traceability, not a real patient's progress.

The therapist's own plan proposes four further sessions over an eight-week review period, plus one parent check-in, subject to the funder's current rules. AI receives the de-identified evidence sheet and a list of required headings. Its first draft uses "independent coat fastening" for G1, even though the visual cue remains. The therapist corrects that line. It also turns the parent's account of mornings into a clinic observation; that attribution is corrected. Finally, the model adds a sentence that the recommended sessions "will eliminate" the difficulty. The therapist deletes the promise because the record cannot support a guaranteed outcome.

The revised letter has five functional parts: request and purpose, goals and baseline, intervention and observed response, remaining participation barrier, and therapist recommendation with review point. The therapist verifies the source of every statement, checks the funder's actual form, adds the appropriate identifying information only inside the authorised document system, signs and records the version sent. A colleague can later reproduce the reasoning from the evidence sheet and session record without needing access to an AI chat history.

Suppose drafting from a blank page previously took 55 minutes and the AI draft takes 12 minutes plus 28 minutes of checking in this illustrative trial. The apparent saving is 15 minutes on that one letter, but it is not the success measure. The more important result is whether the final request is accurate, complete, appropriately disclosed and easy for the decision-maker to follow. If review takes longer because the model repeatedly invents progress, the practice may drop the workflow.

Write the parent update for a different reader

A parent update should not be a shortened funding submission. It should say what was practised, what the therapist observed, what the family reported if relevant, and what happens next. Avoid unexplained professional shorthand. State uncertainty honestly: a skill seen in clinic may not yet appear at home or school. Ask the therapist to specify whether home practice advice is already approved in the care plan; the model must not create a new exercise or change treatment instructions.

For the same fictional child, an illustrative parent update might say: "We practised the coat fastening sequence in today's session. With the visual cue, the final four of five attempts followed the sequence. We will keep checking whether the same steps work in different settings. You mentioned that mornings can be rushed; we will discuss what makes practice manageable at the next review." The therapist should amend the wording to match the actual record and the family's communication preferences before sending.

Compare that with an unsafe AI draft: "Your child has mastered coat fastening and should practise independently every morning." It upgrades supported performance to mastery and prescribes a routine absent from the plan. A parent may treat it as a clinical instruction. The fix is not merely changing the adjective; the therapist must check the intended home-practice instruction against the care plan. This is why the parent update has its own review, even if it uses the same evidence sheet as the funding letter.

Another illustrative update concerns a child who declined a task during one session. The model may frame this as "poor motivation". The record may only say the child chose a different activity after a break. Use observable wording and the therapist's interpretation, if documented. Be careful with family-sensitive details, especially when more than one parent or carer receives updates. Confirm recipient permissions and the communication channel under the practice's normal consent process.

There is a useful separate handout for practical exercises when the therapist has approved them. A parent update may briefly point to it, but should not smuggle new instructions into a progress paragraph. The home-practice handout workflow is from a related discipline; its separation of approved exercise instructions from progress reporting is helpful here too. The occupational therapist still decides whether any exercise is appropriate for this child and verifies every instruction before sharing it.

Use a sign-off checklist that catches clinical drift

The treating therapist should review the draft while the source record is open. Check each number, date, attribution, goal, level of assistance, setting and recommendation. Read any claim of necessity or likely outcome twice. Confirm that no clinical conclusion appears solely because it made the letter sound stronger. Check recipient, consent and attachment. For a funding letter, verify the current form or insurer terms. For a parent update, verify that any suggested activity was already approved. Record who reviewed and when.

RCOT's revised standards leave responsibility for professional decisions with the practitioner, whichever tool helped with the wording. Treat the final signed letter as the professional document; keep the draft and source map according to the practice's record and retention policy. If AI was used in creating a record or report, document that use in the way your practice policy and professional body require. Ask a qualified adviser when the practice is uncertain about confidentiality, disclosure or scheme-specific obligations.

Check whether the drafting process improves the service

For a small practice, measure rejected drafts, missing-evidence flags, corrections after sign-off, parent clarification questions and time spent on review. Sample perhaps five consecutive letters after the pilot, not the five that turned out best. Read the final version against the source record. If AI regularly adds unsupported claims, change the input format or stop using it for funding letters. If the input sheet repeatedly has empty fields, improve assessment documentation rather than prompt design.

A final illustrative test is a request to reuse last month's letter for a new funding round. The prior letter may have old goals, an old frequency and a different form. Instead of asking AI to "update it", build a new evidence sheet from current records and compare changes explicitly. A copied paragraph can survive editing because it sounds plausible. The reviewer should be able to point to the current observation supporting every retained sentence.

Keep the two outputs under separate filenames and approval dates. A funding letter can be approved while a parent update still needs a recipient check, and vice versa. The practice should be able to answer a simple question later: who approved this statement, from which evidence, for which reader and under which version of the funder's rules? If it can, AI has helped with drafting without taking over the therapist's judgement.

One practical way to test that trace is to have a colleague review a fictional letter without seeing the AI prompt. Give them the final text, the evidence sheet and the current funder headings. Ask them to mark every factual statement they cannot locate in the evidence, and every required heading that the letter has missed. If the reviewer finds three unsupported claims, the right response is to repair the draft process and repeat the exercise, not to ask the model to sound more cautious. A letter that is easy to audit is more valuable than one that merely sounds persuasive.

Also check accessibility for the parent version. A family may prefer shorter sentences, a translated version or a phone call to clarify the written summary. AI can help rephrase approved content, but a translation or simplified version still needs human checking for meaning. For example, "requires a visual cue" might be simplified to "does the step when shown the picture"; that may be clearer, but the therapist must confirm it accurately describes the observation. Keep both versions aligned and record which one was sent.

Questions about OT drafting boundaries

Can AI choose the support hours to request?

The therapist should determine and justify the recommendation from assessment, goals, progress and funder requirements. AI can organise the therapist's approved reasoning into clear prose and flag missing fields. It should not invent a dosage, claim that a service meets funding criteria, or convert a parent's wish into a clinical recommendation.

Is removing a child's name enough to de-identify a prompt?

No. Dates, school, rare circumstances, family details and combinations of observations can identify someone. Use a practice-approved de-identification method and the minimum facts needed to test wording. If real records must be processed, review the vendor's data terms, permissions and professional obligations before using the service.

Should a parent update repeat the funding letter?

Usually the purposes differ. A funding letter explains evidence and recommendations to a decision-maker; a parent update explains what was practised, what was observed and what the family can expect next. Both must agree on goals and facts, but the parent message should not become a dense funding argument or imply guaranteed progress.

Further reads

• AI Scribe for Occupational Therapists: A Buyer's Checklist — Assess a tool that handles session information.
• Best AI Tools for Occupational Therapists: School, Clinic, Home Visit — Compare tools by the setting where the OT works.
• AI for Speech Therapy Reports: Draft, De-Identify, Review — See a related evidence-to-report workflow.
• AI for Speech Pathology in Australia: NDIS Reports and Scheduling — Compare evidence headings for another therapy funding route.
• AI for Disability Services: Service Agreements and Incident Logs — Check documentation boundaries in a related service.
• NDIS Occupational Therapy Report Template: A First Draft With AI — Report headings, an evidence table, a de-identified drafting method and a review checklist for an AI-drafted NDIS OT report you stay accountable for.
• How Can AI Help an Occupational Therapy Practice? Seven Admin Jobs — Seven admin jobs for an occupational therapy practice, from report drafts to billing checks, with clinical judgement left with the therapist.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Royal College of Occupational Therapists, news release on the revised Professional standards for practice, conduct and ethics (July 2026); NDIS, Plan reassessment reports guidance for allied health providers. Check the current funder's own form and criteria before submission.