Generative AI in Counselor Education: A Scoping Review of Applications, Ethics, and CACREP Competency Mapping (2015-2026)

Synthesis: Van Patten-Freeman and Rath (2026) scoped peer-reviewed literature on generative AI in counselor education from January 2015 to June 2026 and found 14 qualifying sources — three conceptual and eleven empirical — all published between 2023 and 2026. Five application clusters emerged, dominated by AI-simulated clients. Mapping to the 2024 CACREP curriculum domains showed that 85.7% of sources addressed Counseling Practice and Relationships or Assessment and Diagnostic Processes, while Career Development and Group Counseling were absent entirely. The review's sharpest finding is evidentiary: none of the empirical studies measured trainee competency, used comparison conditions, or employed longitudinal designs, so the literature has established what GenAI does and how trainees perceive it but not whether AI-assisted training produces better counselors.

Key Findings

1. The field appeared suddenly. 14 peer-reviewed articles appeared within 32 months after none before 2023, with the corpus concentrated in 2025 (11 sources).
2. Simulated clients dominate. Five application clusters emerged — simulated clients, supervision support, assessment and documentation, study aids, and curriculum administration — with AI-simulated clients the most common.
3. Mapping is narrow. 12 of 14 sources focused primarily on CACREP 3.E (Counseling Practice and Relationships, 10 sources) or 3.G (Assessment and Diagnostic Processes, 2); career development and group counseling were absent, and ethics and culture appeared mainly as framing.
4. Cultural neutrality is the default. All 14 sources discussed algorithmic bias, but three converged on an actionable finding: GenAI defaults to culturally neutral personas unless the prompt specifies culture, so competence developed through unstructured practice is bounded by the prompt writer's cultural imagination.
5. Crisis and risk assessment are unexamined. No source evaluated AI performance in a crisis or risk-assessment scenario, which the authors name as the most consequential safety gap.
6. Evidence does not reach competence. Six studies analyzed AI outputs, four assessed trainee perceptions, and one reviewed institutional documents; none measured counselor competence, used validated pre/post skills instruments, evaluated transfer to practicum, or linked training to client outcomes.
7. The proposed CE-AIIV framework targets the gap. It organizes future work around comparative, longitudinal, competency-referenced clinical investigation rather than further description of tools.

What the review covered

The search covered PsycINFO, ERIC, Scopus and PubMed, supplemented by hand-searching a special issue of the Journal of Counselor Preparation and Supervision, and screened 1,247 records down to 14 sources after excluding work focused on populations outside counselor education, work that did not address generative AI, service-delivery rather than training studies, non-peer-reviewed items and non-English publications.

The included corpus is thin in a specific way. Three sources are conceptual, eleven empirical; the empirical ones mostly analyze model outputs or trainee perceptions. Designs lacked random assignment, comparison conditions, longitudinal data, and cross-platform or cross-institutional replication — only one study compared platforms. That profile means the field can describe plausible uses but cannot yet say that a given use improves clinical skill.

Applications and their limits

Simulated clients are the center of gravity: trainees practice intake, couple counseling or difficult conversations with a model that can be prompted to adopt a persona. Related work uses AI for supervision support, documentation and assessment, study aids, and curriculum administration.

Three problems recur across these applications. First, cultural defaults: ChatGPT did not spontaneously generate racial or ethnic backgrounds, trainees reported a culturally neutral persona absent explicit prompting, and cultural dynamics appeared only when written into the vignette. Second, dependency and empathy: trainees in one study identified overreliance and diminished empathy as risks. Third, safety: no study tested crisis or risk-assessment scenarios, so the highest-stakes use is the least examined.

The theoretical base is also uneven. Experiential and adult learning theory was the most common framework (five sources, 35.7%), four sources used no formal framework, technology acceptance literature was absent, and no source employed a counselor development model. The authors treat that absence as a design flaw: without a developmental theory, it is hard to specify what a trainee should be able to do at each stage or what AI practice is meant to add.

The CACREP mapping

Mapping to CACREP domains shows where the field has concentrated and where it has not. Domain 3.E (Counseling Practice and Relationships) was primary in 10 sources and 3.G (Assessment and Diagnostic Processes) in two. Domains 3.A and 3.H were primary for one source each; 3.B, 3.C, 3.D and 3.F were primary in none. Secondary mappings tell a different story — 3.A and 3.B appeared in 11 sources — which the authors read as ethics and culture functioning as obligatory framing rather than objects of investigation. Career development and group counseling were entirely absent, a gap that matters because both are core CACREP domains and both involve relational dynamics that simulated clients could plausibly support.

What this means for practice

• Treat simulated clients as a supplement, not a substitute. The literature consistently positions AI as supplementary to supervised practice, but no study has tested that positioning.
• Write culture into the prompt deliberately. The default is neutrality, so cultural competence will not emerge from unstructured practice.
• Do not deploy AI for crisis or risk assessment without evidence. This is the least studied and most consequential scenario in the corpus.
• Measure competence, not perceptions. The field needs validated pre/post skills measures and transfer to practicum before claims about training effectiveness can be made.
• Use a counselor development model to design sequences. Specifying what AI practice adds at each developmental stage would make studies comparable.

Limitations

• The review covers 14 sources, all published in a 25-month window, so it describes an emerging literature rather than a mature evidence base.
• The search was limited to four databases plus one journal's special issue and to English-language peer-reviewed work, so dissertations, conference proceedings and non-English work are absent.
• Coding and charting were conducted by two authors without a reported second independent coder or agreement statistic.
• The CACREP mapping classifies primary and secondary domains, and the authors acknowledge that many sources address multiple domains, which makes counts sensitive to classification decisions.
• The proposed CE-AIIV framework is presented as an agenda, not tested in the review.

Citation

Van Patten-Freeman, A. R., & Rath, S. K. (2026). Generative AI in Counselor Education: A Scoping Review of Applications, Ethics, and CACREP Competency Mapping (2015-2026). The Interdisciplinary Journal for Research on Pedagogy and Learning Design, 1(1).