Given the limited evidence on digital cognitive behavioral therapy (CBT) for chronic subjective tinnitus, particularly interventions supported by large language models (LLMs), this two-arm, 30-day randomized controlled trial will aim to evaluate the efficacy, safety, and usability of an LLM-based conversational CBT intervention compared with a digital education control. Participants with chronic subjective tinnitus will be randomly assigned to either the intervention group, which will receive daily AI-supported CBT sessions through the Fudan Tinnitus Doctor (FTD) system, or the control group, which will receive static tinnitus education materials matched for duration and platform interface. The FTD system will be powered by a multi-agent large language model and will deliver personalized CBT-style dialogues, including psychoeducation, cognitive restructuring, relaxation, and mindfulness guidance. Outcomes will include tinnitus severity, sleep quality, mood symptoms, global improvement, and user adherence and experience.

Chronic subjective tinnitus is a common and often distressing auditory condition that frequently co-occurs with insomnia, anxiety, and depression. Cognitive behavioral therapy (CBT) remains the most evidence-based treatment to reduce tinnitus-related distress, but traditional face-to-face CBT is limited by accessibility, cost, and availability of trained professionals. Digital CBT interventions have demonstrated potential benefits, yet their efficacy varies, and engagement and personalization remain major challenges.

Recent advances in large language models (LLMs) have enabled the development of interactive conversational systems capable of delivering psychologically informed support. The Fudan Tinnitus Doctor (FTD) is a multi-agent conversational AI platform specifically designed to provide CBT-based tinnitus management. It integrates psychoeducation, cognitive restructuring, relaxation, and mindfulness guidance within natural language conversations. The system will be powered by a multi-agent large language model with retrieval-augmented generation (RAG) and safety moderation to ensure content validity, consistency, and user safety.

This single-centre, parallel-group, open-label randomized controlled trial will evaluate the efficacy, safety, and usability of the FTD system compared with a digital education control. Participants with chronic subjective tinnitus meeting the inclusion criteria will be randomly assigned (1:1) to either the FTD intervention group or the control group. The intervention group will receive 30 days of AI-supported CBT sessions through the secure hospital web platform, while the control group will receive static tinnitus education materials matched for platform and duration but without interactive AI components.

The primary outcome will be the change in tinnitus severity measured by the Tinnitus Handicap Inventory (THI) from baseline to Day 30. Secondary outcomes will include sleep quality (Pittsburgh Sleep Quality Index, PSQI), anxiety (GAD-2), depression (PHQ-2), and overall perceived improvement (Patient Global Impression of Change, PGIC). Exploratory outcomes will assess usability (System Usability Scale, SUS), satisfaction (Net Promoter Score, NPS), and engagement metrics.

All assessments will use validated Chinese versions of the respective questionnaires. Safety monitoring will be conducted throughout the study period. Adverse events and any AI-related performance errors will be automatically logged and reviewed by the research team.

This study will provide preliminary evidence on the efficacy and safety of an LLM-based conversational CBT system for tinnitus management, offering a scalable and patient-centered approach that could enhance accessibility and adherence in digital mental health care.

Inclusion Criteria:

1. Diagnosis of subjective tinnitus confirmed by a qualified investigator.
2. Tinnitus duration ≥ 3 month prior to screening.
3. THI score of ≥ 18
4. Aged between 18 and 70 years
5. The ability to use a smartphone: Both groups will receive therapeutic support and disease management primarily through a mobile application or web-based platform. This requires functional ability to operate the provided software and willingness to engage with the digital intervention protocol.
6. Perceptibility: Tinnitus is present and recognizable by the participant under ordinary daily conditions (not only in quiet environments).
7. Compliance and consent: Participant is assessed by the Investigator as able to understand study procedures, voluntarily agrees to participate, is willing to comply with all trial-related requirements, and provides signed informed consent prior to performance of any protocol-specified procedure.

Exclusion Criteria:

1. Acute-phase tinnitus (duration < 3 month or tinnitus secondary to acute otologic events).
2. Pulsatile or objective tinnitus.
3. Unstable symptoms: Fluctuating tinnitus loudness at screening or baseline.
4. Psychiatric contraindication: History or current presence of severe psychiatric disorders (e.g., major depressive disorder with recent suicidality, schizophrenia) that, in the opinion of the Investigator, may interfere with study participation or data validity.
5. Testing limitations: Inability to adequately complete tinnitus testing procedures or comply with study assessments.
6. Prior therapy exposure: Previous history of sound therapy specifically for tinnitus management.
7. Concurrent treatments: Ongoing psychotherapy for conditions such as anxiety or depression, or any other psychological treatment that may interfere with the study endpoints.
8. Concurrent trial participation: Active participation in another interventional research study that, in the opinion of the Investigator, could affect tinnitus symptoms or interact with study procedures.
9. Investigator judgment: Any condition (medical, surgical, psychiatric, or social) or factor which, in the Investigator's judgment, renders the subject unsuitable for participation, places the patient at higher risk, or may interfere with compliance or scientific validity of the trial.

Deidentified individual participant data (IPD) collected for this trial, including outcome measures and baseline characteristics, will be made available upon reasonable request to qualified researchers. Data sharing will require prior approval from the Institutional Review Board of the Eye & ENT Hospital of Fudan University. The DeepSeek-R1 671B model used to support the intervention is open-source and publicly accessible at https://github.com/deepseek-ai/DeepSeek-R1.