AI-Assisted Coaching: Practical Applications for RBT and Paraeducator Supervision

Overview & Clinical Significance

The supervision demands on BCBAs in school and clinic settings have never been greater. Sarah Heiniger's presentation addresses a resource constraint that most behavior analysts recognize immediately: there are not enough hours in a BCBA's day to provide the individualized, consistent, high-quality coaching that RBTs and paraeducators need to maintain procedural fidelity and continue developing as practitioners. AI tools, applied thoughtfully and within clear ethical boundaries, offer a partial solution to this structural problem.

The clinical significance of the problem Heiniger addresses is real. Supervision quality is a primary determinant of implementation fidelity, and implementation fidelity is a primary determinant of client outcomes. When BCBAs are stretched across multiple roles — clinical supervisor, behavior consultant, crisis responder, IEP team member, documentation manager — coaching quality is often the first thing to degrade. RBTs who receive inconsistent or insufficient coaching implement procedures with lower fidelity, develop skills more slowly, and experience higher rates of burnout-driven turnover.

AI tools do not replace the clinical judgment and relationship-building that effective supervision requires. What they can do is reduce the administrative and cognitive load associated with supervision tasks that are important but not irreducibly human: generating individualized SMART goals, creating first-draft coaching scripts, summarizing performance data patterns, and producing training materials tailored to individual RBT competency profiles. This load reduction frees the BCBA's attention and time for the clinical and relational dimensions of supervision that cannot be automated.

The ethical dimension of this application is central to Heiniger's framework. Introducing AI tools into supervision creates genuine risks — privacy risks, quality risks, risks of supervisors accepting AI-generated content without critical review — that require explicit ethical analysis. BCBAs who adopt AI tools without this analysis are not practicing within the spirit of the Ethics Code's requirements.

Background & Context

Artificial intelligence tools — particularly large language model-based applications — have entered nearly every professional context, and behavior analysis is not exempt. The current generation of AI tools can generate written content, analyze patterns in structured data, produce customized training materials, and engage in rudimentary clinical reasoning when provided with well-structured prompts. Their capabilities are real and expanding; their limitations are equally real and must be understood before clinical application.

For supervision applications specifically, the most relevant AI capabilities include: generating individualized SMART goals aligned with BACB task list competency areas when provided with performance data about a specific RBT; producing coaching scripts for specific behavioral procedures that a supervisor can review, edit, and use as a feedback guide; creating study materials tailored to identified knowledge gaps; summarizing performance trends across data collection systems; and generating scenario-based practice questions for competency development.

The concept of 'prompt engineering' — the skill of writing effective queries to AI systems to produce useful outputs — is directly relevant for BCBAs using AI in supervision contexts. Heiniger's presentation teaches practitioners to develop AI prompts that provide sufficient behavioral specificity to produce clinically useful outputs rather than generic responses. The quality of AI output is highly dependent on the quality of the input: a prompt that describes a specific RBT's performance profile, the behavioral procedure at issue, and the supervision context in behavioral terms will produce a more useful coaching script than a vague request for help with RBT training.

The school-specific context of this presentation is important. BCBAs serving in school settings — particularly those functioning as the only behavior analyst across multiple buildings — face supervision demands that would exceed any individual's capacity to address through traditional supervision models alone. Paraeducators in school settings often receive even less specialized coaching support than clinic-based RBTs, yet their behavioral implementations with students with disabilities are equally critical to student outcomes.

Clinical Implications

For BCBAs using AI to support RBT and paraeducator coaching, the clinical implications begin with clarity about what AI can and cannot do well. AI can produce SMART goal drafts that align with task list competency areas — but the BCBA must review them against the specific RBT's current performance profile, the clinical context, and the feasibility of the implementation environment. AI can produce coaching script drafts — but the BCBA must ensure they accurately describe the specific procedure being coached, that the script matches the implementation context, and that the feedback language is appropriate for the individual supervisee.

The AI-assisted goal generation process has a specific clinical benefit beyond efficiency: it can surface competency areas that supervisors, under time pressure, might overlook. If a BCBA provides an AI system with an RBT's task list performance summary and asks for a prioritized set of SMART goals, the output may identify skill gaps in less frequently coached competency areas that are nonetheless clinically significant. The AI functions as a systematic checklist rather than a creative replacement for clinical judgment.

Data analysis applications are particularly promising for high-caseload supervisors. AI tools can process session note data, data sheet records, and performance checklists to identify patterns — declining fidelity on specific procedures, increased behavioral incidents following schedule changes, inconsistent reinforcement delivery for specific client programs — that might escape notice in manual review across large datasets. The clinical decision-making based on those patterns still requires BCBA judgment; the pattern identification can be assisted by AI.

For paraeducators in school settings, AI-generated training materials — practice scenarios, knowledge checks, procedure summaries — can supplement the limited direct coaching time available. Paraeducators who have access to well-designed self-study materials between BCBA coaching visits maintain implementation skills better than those who rely entirely on intermittent direct coaching. AI can make the production of these materials feasible within realistic time constraints.

The limitations of AI for supervision are clinically important. AI cannot observe behavior, build therapeutic relationships, navigate complex family dynamics, make ethical judgments, or respond adaptively to the real-time unpredictability of clinical settings. BCBAs who treat AI-generated supervision content as finished rather than as a first draft requiring clinical review are introducing quality risks that may affect both supervisee development and client outcomes.

Ethical Considerations

Heiniger identifies three primary ethical consideration domains for AI use in supervision: confidentiality, oversight, and informed use. Each maps onto specific BACB Ethics Code requirements.

Confidentiality under Code 2.04 and Code 2.05 requires that client information be protected and that data be used only for clinical purposes with appropriate consent. When BCBAs use AI tools to process clinical data — including session notes, behavioral data, or supervisee performance information — they must ensure that identifiable information is not submitted to AI systems without appropriate authorization. Many current AI tools process inputs through cloud-based systems; submitting identifiable client data to these systems without understanding their data handling practices may constitute a confidentiality breach. BCBAs should use de-identified data when leveraging AI for analysis tasks, and should review any AI platform's data handling policies before using it for clinical data processing.

Code 2.01's requirement for evidence-based practice applies to AI-generated clinical content. AI systems can produce convincing text that contains clinical inaccuracies, misapplied behavioral terminology, or procedural recommendations that are inconsistent with the evidence base. BCBAs bear full clinical and ethical responsibility for all supervision content they use, regardless of how it was generated. Reviewing and critically evaluating AI outputs before use is not optional — it is an ethical requirement.

Code 1.05 on competence applies directly to AI use. BCBAs who use AI tools in supervision without understanding the tools' capabilities and limitations are practicing outside their competence in this specific domain. Developing competence in ethical AI use is now a relevant professional development need for BCBAs in supervisory roles — not because AI is required but because its use requires competent application to be safe.

Code 4.04's requirement that supervision genuinely develop supervisee competency is relevant here as well. AI-generated content that is accepted uncritically and passed to supervisees without clinical review does not fulfill the BCBA's supervisory obligation — it substitutes efficiency for genuine investment in supervisee development.

Assessment & Decision-Making

Deciding which supervision tasks are appropriate for AI assistance requires a competency analysis of the task itself. Tasks that require clinical judgment — diagnosing the function of a performance problem, deciding whether to modify a treatment plan, determining whether a supervisee is ready for less intensive supervision — are not appropriate for AI assistance. Tasks that are primarily administrative, generative, or pattern-recognition in nature are candidates for AI support.

The prompt development process requires deliberate practice. Effective supervision prompts for AI tools typically include: a description of the supervisee's current competency level in behavioral terms, the specific skill or task area being addressed, the performance objective and criterion, the clinical context (setting, population, procedure type), and any relevant constraints (the supervisee's learning history, available supervision time, organizational resources). Prompts that include this level of specificity produce outputs that require less revision than generic prompts.

Decision-making about AI output quality requires critical review criteria. Before using an AI-generated coaching script, the BCBA should assess: Is the behavioral procedure described accurately and in sufficient detail? Is the feedback language specific and behavioral rather than vague and evaluative? Is the content appropriate for the supervisee's current skill level? Does the script reflect any knowledge gaps or errors that must be corrected? Does the suggested approach align with the evidence base for the specific procedure?

Data analysis applications require particular caution around AI-generated pattern interpretations. AI systems identify patterns in data but do not understand the clinical context that gives those patterns meaning. A trend identified by AI as 'declining performance' may reflect a clinical decision — a more challenging goal phase, a new implementation context — rather than a performance concern. BCBA interpretation of AI-identified patterns against the clinical record is always necessary.

What This Means for Your Practice

Start with one specific supervision task that consumes disproportionate time relative to its clinical complexity: generating draft SMART goals for upcoming supervision cycles, creating a procedure summary document for a new RBT, or drafting a practice scenario bank for a paraeducator training. Use an AI tool to produce a first draft, then review, edit, and refine it before use. Track the time savings and the quality outcome. This first experiment establishes both your personal sense of AI utility and your initial prompt-writing skills.

Develop your prompt library. As you create effective prompts for specific supervision tasks, save them. A library of well-designed prompts for common supervision applications — SMART goal generation, coaching script drafting, knowledge check creation — becomes a durable efficiency resource that improves with iteration.

Establish a clear personal policy about what types of supervision content you will and will not use AI to generate. Write it down. This policy should address: data de-identification requirements before AI submission, review criteria for AI-generated outputs, and the types of clinical decisions that remain purely within BCBA judgment. Having an explicit policy protects against the gradual scope creep that occurs when AI use is governed only by implicit habits.

For school-based practitioners specifically: identify the paraeducator training tasks that are most repetitive across multiple buildings and that could be most efficiently standardized with AI assistance. Procedure summaries, visual supports for implementation steps, and practice scenario banks are high-value targets. The efficiencies compound across multiple paraeducators — a one-hour AI-assisted training material development session may serve 20 implementers across multiple schools.