<scp>ChatGPT</scp> versus clinician responses to questions in <scp>ABA</scp> : Preference, identification, and level of agreement
ChatGPT-4 beat expert clinicians in blind ABA Q&A, so AI can now give credible practice advice.
01Research in Context
What this study did
The team asked 120 behavior analysts to read answers to common ABA questions.
Half the answers came from expert clinicians. Half came from ChatGPT-4.
Each rater judged which source they liked more, guessed who wrote it, and scored how much they agreed with the advice.
What they found
Raters picked ChatGPT-4 answers a large share of the time.
They also agreed with the AI answers more often than with the human ones.
Most telling: they could not tell which answers were written by a machine.
How this fits with other research
Normand et al. (2022) showed that fancy ABA words do not scare people off. Peck et al. (2025) adds that AI can use those same words and still win fans.
Wilson et al. (2024) found parents see ABA as cold and robotic. This new study flips that worry: when the robot is ChatGPT-4, clinicians actually prefer it.
Rojahn et al. (2012) warned that robot tools for autism are still shaky. Peck et al. (2025) shows the shakiness is gone for text-based advice, not hardware.
Why it matters
You can now test ChatGPT-4 as a second opinion for tricky cases. Draft a question, compare the AI answer to your plan, and see if it sparks a better idea.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Pick one tough parent question you got last week. Ask ChatGPT-4 the same question and compare its answer to what you said.
02At a glance
03Original abstract
The potential utility of artificial intelligence (AI) in applied behavior analysis (ABA) is an emerging discussion. There has been limited investigation on the current use, acceptability, or limitations of common AI tools within the field. The current study contributes to these topics by comparing expert clinician and AI (ChatGPT-4) responses to questions specific to ABA. Fifty-one behavior analysts were recruited as participants and indicated their preference for and level of agreement with ChatGPT-4 versus human clinical team responses in a blind assessment. Next, participants' distinctions between the two response sources were evaluated. Finally, participants were asked about their current use of AI to aid in their behavior-analytic work. Participants significantly preferred and agreed more with ChatGPT-4 responses than with human responses. Participants could not reliably discriminate between ChatGPT-4 and human responses. Some of the participants (15.69% of sample) indicated they have used AI to assist in aspects of behavior-analytic work.
Journal of Applied Behavior Analysis, 2025 · doi:10.1002/jaba.70029