AI-Assisted PT Did Not Improve Shoulder Disability

New research suggests AI-generated exercise programs perform similarly to clinician-designed plans for shoulder pain.

Updated on Sept. 26, 2026 in Fitness

A yellow resistance band attached to a wall-mounted anchor in a bright, modern physical therapy clinical setting.
A randomized controlled trial found that AI-assisted physical therapy plans provide clinical outcomes comparable to those designed by human physiotherapists for subacromial pain. AI Illustration. Upload story photo >

Live Poll

Do you trust AI-generated programs to effectively manage your physical pain as well as a human?

A recent randomized controlled trial comparing AI-assisted and conventional physiotherapy for subacromial pain syndrome found no significant difference in overall shoulder pain and disability outcomes. The study evaluated exercise programs developed with the help of artificial intelligence versus those curated by physiotherapists.

Why it matters

Understanding how AI tools integrate into rehabilitation helps patients assess the role of technology in their care. As clinics adopt digital aides, this study provides evidence on whether these tools offer additional clinical benefits over standard practitioner-led care.

This randomized controlled trial of 75 participants compared three exercise programs—conventional, ChatGPT-4o-assisted, and DeepSeek-R1-assisted—over 6 weeks. While disability scores did not differ significantly, the DeepSeek-R1 group showed a 2.74 mean difference improvement in resting pain.

The players

DeepSeek

An artificial intelligence research organization that developed the DeepSeek-R1 model used in the study.

OpenAI

The developer of ChatGPT-4o, an artificial intelligence language model utilized to assist in generating exercise programming.

The details

Participants attended supervised exercise sessions three times per week for the duration of the study. The researchers aimed to determine if artificial intelligence could effectively curate exercise regimens that match or exceed the standard of care designed by human physiotherapists. While AI-assisted programming showed some difference in resting pain scores, it did not change the primary outcomes for functional disability compared to conventional methods.

Timeline

  1. August 3, 2025: The clinical trial was first posted to ClinicalTrials.gov.

  2. September 26, 2026: The research findings were published online.

Health Landscape

The integration of artificial intelligence into rehabilitation marks a shift in how exercise programming is generated for patients with musculoskeletal conditions. This study compares these emerging digital aides against the established standard of clinician-curated protocols.

If you are currently managing shoulder pain, these findings suggest that AI-assisted programs may be a supplement to, rather than a replacement for, standard physiotherapy. Discuss the appropriateness of digital exercise tools with your clinician to determine if they fit your specific treatment plan.

The takeaway

While AI shows promise in technical programming, it currently does not provide superior outcomes for shoulder disability compared to expert physiotherapist guidance. Patients should continue to prioritize comprehensive clinical assessments when selecting physical therapy interventions.

Further reading

For more on managing common exercise-related injuries and recovery protocols, visit our Fitness section.

More information

Review the peer-reviewed research article for a detailed breakdown of the trial methodology and results.

Source note: This article includes information reported by Nature.

Live Poll

Do you trust AI-generated programs to effectively manage your physical pain as well as a human?