The ‘AI gauge gap’: how patient use of AI affects musculoskeletal care

Plain Language Summary

Patients and their families are increasingly seeking advice from artificial intelligence (AI) tools such as chatbots and agents before they see a specialist. In rheumatology care, patients may use AI while waiting for an appointment, trying to understand their symptoms, or deciding whether their symptoms need urgent attention. This commentary proposes a new way to think about what AI changes before patient care begins. I call this the ‘AI gauge gap’. AI is often judged by what it can do or how much time it saves. Patients may also ask whether the advice they receive from AI helps them feel understood, changes whether they seek formal care, affects their trust in formal healthcare, or keeps human support reachable. In this commentary I propose five questions for specialist teams and their consumer partners:

1. Where might AI enter our care pathway?

2. How might AI change what symptoms mean to our patients?

3. How might AI change what our patients do next?

4. How might AI change our next patient encounter?

5. Who remains reachable and accountable after AI influences our patient’s beliefs or actions?

The aim of these questions is to help make AI use easier to discuss with patients and ensure patients still have access to human support when AI has influenced their understanding or beliefs.

Introduction

Patients increasingly use online AI and AI-supported tools before rheumatology appointments. In a 2025 national survey of people with rheumatic diseases, 27% reported health-related AI use, 58% were interested in using it, and 58% would welcome rheumatologists using AI-supported second opinions in clinical decision-making.¹ An earlier multicentre study found that 46% of patients had searched symptoms online before their appointments.² The AI tools patients may use include large language models and multimodal systems to interpret symptoms, prepare questions, seek reassurance, compare explanations, or decide whether care is needed. AI can therefore influence a patient before a clinician has listened to, examined, explained, or contextualised their story.

Several international frameworks provide a useful context for considering the influence of AI in patient care in Australia. For instance, the Organisation for Economic Co-operation and Development (OECD) AI Principles apply, because Australia is an OECD member.³ Australian policy includes the National AI Plan and the Guidance for AI Adoption, while health-specific instruments include the Australian Government’s review of health-care legislation and regulation, the Australian Commission on Safety and Quality in Health Care’s AI Clinical Use Guide, and the NSW Health AI Framework.^4–8 To date, Australia has not enacted an equivalent to the European Union’s AI Act, and the Australian Government has not yet proceeded with previously proposed mandatory guardrails for high-risk AI.^4,9 Whether more specific risk-based legislation is needed remains an open policy question. Existing instruments mainly address broad constructs, such as organisational adoption, clinical use, privacy, and assurance. This stands in contrast to the specific potential influence of unsupervised, general-purpose AI on a patient’s interpretation of their symptoms before care.

Impact of patient use of AI on care

The term ‘AI gauge gap’ is a lens to describe a possible mismatch between (A) the outcomes technical and clinical users employ to judge AI progress and (B) the outcomes affected when AI becomes involved in a patient’s reasoning. There are other gauges to consider in relation to AI. For example, a capability gauge assesses what a system can do, an efficiency gauge assesses what resources a system saves, and clinical and patient gauges focus on how a system influences the meaning of symptoms, urgency, care-seeking, stigma, trust, responsibility, and future interactions. The AI gauge gap complements governance principles by directing attention to consumer use of AI outside formal pathways. With rapid developments in AI capabilities, this could facilitate meaningful discussion, as well as potential co-design and empirical testing.

Direct evidence for an AI gauge gap in musculoskeletal care is not yet available. Rheumatology studies already show why multiple outcomes should be considered together. In a 600-patient multicentre study, 73% and 79% of participants said they would recommend the AI-based Ada and online Rheport symptom-assessment tools, respectively.² In the associated diagnostic analysis, Rheport achieved 62% sensitivity and 47% specificity for inflammatory rheumatic disease, while Ada’s top diagnosis achieved 52% sensitivity and 68% specificity.¹⁰ Diagnostic performance was therefore insufficient for safe triage, despite substantial patient acceptance. A later survey found that patients valued access to information and faster diagnosis but also worried about faulty outputs and reduced human interaction.¹ Therefore, the proposed AI gauge gap leads us to a question for study: Do technical and service metrics align with the outcomes patients value before, during, and after care?

Musculoskeletal care provides a useful test setting, because symptoms may not be visible during a consultation and can carry different meanings for patients, families, and clinicians.¹¹ For example, a person asking if movement is safe may also be asking if their body can be trusted. Or a parent asking about school attendance may be balancing protection, function, guilt, and disbelief. AI may improve access to information and help people prepare for consultations. The AI gauge gap discussion prompts (see Table) may also help patients and carers identify when human review is needed, and discuss AI-generated information without embarrassment or blame.

Patient safety

Patient-facing AI is a communication intervention as well as an information source. Consider a family navigating school-based stigma about a child’s chronic pain. Pain-related stigma can occur across family, school, peer, and clinical contexts, including experiences of disbelief or minimisation,¹² and diagnostic certainty does not necessarily remove it.¹³ An AI conversation could legitimise symptoms, minimise them, or convert uncertainty into a confident explanation. Clinicians can invite patients to discuss the advice suggested, what seemed helpful or worrying, and which decisions still require shared judgement. Such discussion should cover benefits, limitations, uncertainty, privacy, and escalation, rather than only record a ‘yes/no’ for AI use. The safety of generative AI extends beyond factual accuracy, because people may over-trust fluent medical advice,¹⁴ and sycophantic AI models may agree too readily with a user’s initial premise.¹⁵ In musculoskeletal care, agreement with a patient’s fear, doubt, or even a reductive explanation of pain, may all influence the meaning they assign to their symptoms before more familiar risks arise, such as diagnostic delay or clinical errors.

Safety concerns and strategies require multiple gauges to be considered. For example, LupusGPT offers a patient-led, clinician-supported, multilingual information model developed through co-design, curated evidence, retrieval-based grounding, and iterative testing.¹⁶ Retrieval-augmented generation may reduce unsupported answers,¹⁷ but technical grounding does not establish how AI advice influences meaning, clinical actions, trust, or responsibility. Those outcomes still require consumer-led evaluation.

Discussion prompts

The AI gauge gap offers five discussion prompts for service-planning with patients, carers, consumer representatives, clinicians, and service leaders (see Table). In a service-planning meeting, participants might identify a pre-consultation touchpoint, work through the prompts, identify a priority benefit or risk, develop a service response, assign an owner to the service response, and nominate an outcome for review. Service responses might include waiting list information, public-facing guidance, an escalation pathway, or an invitation for patients to bring AI-generated advice into care. The overall call to action is to use these prompts to help structure co-design and evaluation plans.

Because AI may influence beliefs before a referral or appointment, a service-level response to AI use must begin by considering what happens before an initial consultation. Future co-design should test these prompts across population- and service-level communication. This could include public websites, waiting list and referral information, digital triage, health-literacy resources, and clear escalation pathways. Clinical appointments may be a relatively late opportunity in a patient’s journey to discuss how AI advice influenced them and how clinicians can prioritise human support.

For system-level AI-related implementation planning, the term ‘human in the loop’ is too imprecise. A person may be present during an AI-based interaction, or they may be nominally accountable for an AI output, or they may be reachable only if a user identifies a need for help. Patients need to know whether a suitably informed person can be reached, what that person can do, and who accepts responsibility for influential advice. Services should therefore specify the human role, response times, escalation triggers, and the limits of responsibility, rather than rely on broad labels such as keeping a ‘human in the loop’.

Professional communities can help determine which outcomes deserve the most attention. Recent rheumatology studies found differences in AI attitudes or usability by age and education.^1,2 Future co-design should therefore examine children and young people, older adults, carers, people with lower health or digital literacy, culturally and linguistically diverse communities, Aboriginal and Torres Strait Islander Peoples, and people with limited specialist access. The five prompts in this article should not be assumed to function similarly across these and other groups. Equity also requires attention to language, cultural assumptions, Indigenous Data Sovereignty, and determining who carries the risks of errors.

The AI gauge gap presented here is a conceptual proposal. Its clinical value should be tested through co-design and empirical evaluation. A durable question for musculoskeletal services and professional communities to keep on our radar is: How should care function when AI is involved before a clinician?

Implications for Aboriginal and Torres Strait Islander Health

Patient-facing AI in musculoskeletal care has implications for Aboriginal and Torres Strait Islander health because systems may reflect language, cultural assumptions, access barriers, and data practices that do not align with community priorities or Indigenous Data Sovereignty. The proposed lens has not been developed with Aboriginal and Torres Strait Islander communities and should not be applied in these contexts without Indigenous leadership, governance, and accountability. Any future co-design or evaluation should follow Australian Institute of Aboriginal and Torres Strait Islander Studies and National Health and Medical Research Council ethical principles.

AI statement

The author used OpenAI Codex to support brainstorming of additional potential gauges considered during preparation of this commentary. The author wrote, reviewed, revised, and takes full responsibility for all content.

Patient consent

No patient images, individual patient data, or case-identifiable information are included in this commentary. Patient consent was not required.

Declaration of interest

The author has received speaker fees for presentations on pain and rehabilitation and receives royalties for educational children’s books.

Funding

No specific funding supported the preparation of this commentary.