Home » Insights » Ask, Audit, Apply: A Simple Way to Think With AI in Practice

Ask, Audit, Apply: A Simple Way to Think With AI in Practice

What the Ask, Audit, Apply framework proposes

Andrew Jenkins, Katherine Eisenberg and Roy Ziegelstein published the Ask, Audit, Apply framework in the Journal of General Internal Medicine in October 2026 (Jenkins et al., 2026). It gives clinicians a short routine for working with AI decision support tools. It is a conceptual framework, and it has not yet been tested in clinical trials or in real workflows.

The authors focus on automation bias. People tend to lean on automated output and check it less, especially when it sounds authoritative. Generative AI adds to the problem because it writes fluently. The framework breaks the general idea of human oversight into three things a clinician does.

Ask: frame the question before you open the tool

Before typing anything, state the clinical question, the relevant context and what you want from the tool. Vague prompts produce broad or off-target answers. The authors also see this step as a check on anchoring, because writing the question down slows your thinking at the right moment.

Audit: check what the answer rests on

First, decide whether the response is simply information or an actual recommendation. Then look at the evidence behind it. Does the tool draw on curated clinical sources, or on general text from the internet? Nobody reviews these responses before you see them, so the appraisal falls to you. When the tool says nothing about its sources, verify the content elsewhere.

Apply: make the decision yours

AI does not make decisions. Clinicians do. You weigh the output against the patient’s values, the competing risks and your own judgement, and then you record your reasoning in your own words. Responsibility cannot pass to an algorithm, even when you follow its suggestion.

Why Ask, Audit, Apply matters for psychologists

Psychologists may not ask AI for a differential diagnosis every day. However, many already use it for session notes, report drafts, research summaries and psychoeducation material. The same three steps fit those tasks well. A report drafted by AI still needs an audit, and the formulation inside it still needs to be yours.

The American Psychological Association reached a similar conclusion in its 2025 ethical guidance. Psychologists should critically evaluate AI content before using it in clinical work, and they remain responsible for final decisions (American Psychological Association [APA], 2025). The guidance also calls for informed consent that explains the purpose, benefits and risks of any AI tool used in a client’s care.

There is a second reason to pay attention. Your clients are using AI too. In the APA’s 2026 Chatbots and Mental Health Survey, 77% of psychologists said their patients report using AI, and 39% had patients who used it to self-diagnose (APA, 2026a). Furthermore, 94% of respondents said chatbots cannot treat mental health conditions with the nuance those conditions need.

 

How the framework lines up with HPCSA Booklet 20

South African practitioners already have local guidance to work with. The HPCSA published Booklet 20, its ethical guidelines on the use of artificial intelligence, in September 2025 (Health Professions Council of South Africa [HPCSA], 2025). It rests on three pillars: ethics, legality and technical standards.

 

Each step of the framework sits comfortably next to those pillars. Ask fits the ethics pillar, because a well framed question keeps the client’s needs at the centre. Audit fits the technical pillar, which expects tools to be reliable and safe and warns against informal or unvalidated ones. Apply matches the booklet’s core message that AI may never replace a practitioner’s own expertise and judgement.

The legal pillar adds something the American paper does not cover. Client information you enter into an AI tool falls under POPIA. Before you paste session notes into a chatbot, check where that data goes and whether your client agreed to it. The booklet also states that patients who decline AI may not be disadvantaged or refused access to health services.

PsyCampus offers an AI+ Healthcare certification for practitioners who want formal training in this area. Recent graduates can also check whether they qualify for a graduate discount on CPD.

A version of Ask, Audit, Apply for everyone else

The framework was written for doctors, but the logic works for anyone who asks a chatbot about their health, their mood or their child. In 2026 the APA released a public guide to AI-generated advice that makes many of the same points (APA, 2026b). Here is how the three steps translate.

  • Ask: Be specific about what you want. “Help me list questions for my psychologist” is a better use of AI than “What is wrong with me?”
  • Audit: Treat the answer as a starting point. Check any health information with a qualified practitioner, and remember that AI produces text without reasoning the way a clinician does.
  • Apply: Keep the big decisions with people. The APA advises against using AI to diagnose, interpret assessments, make major life decisions or support you through a crisis.

If you already see a psychologist or counsellor, tell them you use AI. They can help you separate the useful parts from the misleading ones. And if you are in crisis, speak to a person you trust, your practitioner or the South African Depression and Anxiety Group (SADAG).

What the framework cannot tell us yet

Ask, Audit, Apply is a sensible idea, not a proven intervention. The authors say future studies must test whether it changes how clinicians check AI output and whether patients end up safer (Jenkins et al., 2026). Until then, treat it as a habit worth building rather than evidence of better outcomes.

It also says little about psychotherapy itself. The relationship, the cultural context and what happens in the room are hard to capture in a prompt. That gap is a useful reminder of where AI belongs: a tool you question, not a colleague you defer to.

Start with one task this week

Try the three steps on the next AI task you do, even a small one like drafting a psychoeducation handout. Write the question first, check where the content comes from, then edit until the work is yours. After a few rounds it becomes routine.

Our HPCSA accredited CPD courses include further reading on ethics and technology in practice.

References

Shopping Basket
Scroll to Top