AI and ADHD — Clinical Quick Reference

Key frameworks and questions from "AI and ADHD: A Clinical Look at Promise and Peril"

Clinician reference Print-ready May 2026

Educational reference for clinicians. Not therapy, supervision, diagnosis, legal advice, or a substitute for clinical judgment with a specific client.

Three intake questions worth adding now
Question 1
"Are you using any AI tools regularly — for work, organization, decision-making, or emotional support?"
That last category — emotional support — is the one that opens the most. Don't skip it.
Question 2
"Has AI changed how you think about yourself or your diagnosis?"
This catches the self-diagnosis thread, the way AI can reinforce or destabilize someone's existing self-concept, and the impact of being told by a machine that your experience is valid or diagnosable.
Question 3
"What happens when the tool isn't available?"
This is your dependency screen. If removing access causes significant functional decline — increased anxiety, inability to initiate, loss of organization — that's clinical information. The same question you'd ask about any other coping strategy.
Reading AI use patterns — what they reveal clinically
Pattern
What it looks like
Clinical signal
Endless validation-seeking
Needs every angle considered before acting. Returns to AI to re-check decisions already made.
Perfectionism anxiety
Constant revision
Draft after draft. Nothing gets sent. Output quality is secondary to the act of refining.
Completion avoidance (thoroughness costume)
AI as primary relationship
Prefers AI conversation to human contact. Uses it as a confessional. Arrives pre-processed.
Vulnerability avoidance — discomfort with being genuinely seen
Self-diagnosis via AI
"ChatGPT told me I have autism." Running characteristics lists. Personalized, responsive validation.
Identity instability + need for external confirmation of internal experience
Polished but unexplained
Client produces articulate, organized output in session. Struggles to say where it came from.
Masking amplification — AI smoothed the signal before you could read it
Two guardrails you can give clients directly
The One-Action Rule
End every AI interaction with one single, concrete, measurable action you can take in the real world within the next 48 hours.
Building a plan is not an action. Gaining insight is not an action. Sending the email is an action. Trying the new routine once — just once — is an action.
Converts AI from a loop amplifier into a change scaffold. If the client can't name a real-world action, the session may have been a productive sideshow.
Write First, Prompt Second
Before opening AI, write your own ideas first. Five minutes. A notes app, paper, anything. Process-write the thought that drove the impulse to use AI.
Protects voice. Gives the client something to bring to AI rather than letting AI define the starting point. Preserves the signal before elaboration drowns it.
Especially useful for ND clients who use AI for writing, creative work, or processing between sessions.
The Presence Gradient — where AI belongs
In-Person
Full presence — nonverbal, somatic, bidirectional
→
Video
Reduced spatial; visual and verbal intact
→
Phone
Voice only; para-verbal remains
→
AI
Text only; no mutual transformation
AI belongs on this spectrum. It just belongs at one end. Knowing where it sits protects the rest of the spectrum from being displaced.
The Productive Sideshow Check — Adler's frame applied to AI
Is this activity goal-directed, or does it fail to address the actual target?
  • Activity feels productive. Includes genuinely productive aspects. But nothing has changed in the real world.
  • Motion without movement. Effort without progress.
  • AI can identify these patterns. Name them with impressive accuracy. But it is structurally incapable of interrupting them.
  • That interruption is the clinical work.
Informed consent consideration
If a client is using AI as a significant therapeutic supplement — processing between sessions, drafting responses to things that came up, working through decisions — does your treatment plan account for that? Not as a reason to intervene, necessarily. As a reason to know. The client who arrives having already processed extensively with AI is arriving with a different readiness than the client who has been sitting with something raw all week. That affects where you start. It affects what the session can do.