Retrospective audit of consecutive patients presenting with anxiety, depression, or stress (April–August 2025) at J-Shalom Hospital, a primary care and psychiatry clinic in Ibadan, Nigeria. The clinic operates part time psychiatric services and therefore records a relatively low volume of mental health consultations during the audit period. Of 38 mental health presentations, 28 completed screening.

Per Nigeria’s National Code of Health Research Ethics (NCHRE) [9], clinical audits evaluating existing care practices without experimental intervention are exempt from full Health Research Ethics Committee review. Data were collected as part of routine clinical assessment and subsequently anonymised for audit purposes. Screening questions related to technology stress were incorporated into routine clinical interviews conducted by treating clinicians. No identifiable patient information was retained. Permission to review anonymised clinic records for audit purposes was granted by J-Shalom Hospital administration. The clinic operates under the purview of the Federal Ministry of Health guidelines; no formal data transfer agreement or external institutional approval was required for internal audit of routinely collected data.

Consent refusal was treated as an exclusion criterion consistent with standard clinical practice; no experimental procedures requiring formal research consent were conducted. Of 38 eligible patients, 7 declined routine clinical data use or were excluded on clinical grounds, yielding a final sample of 28.

Two clinicians conducted screening using AIAS-adapted questions incorporated into routine clinical interviews, covering general technology stress and four AI dimensions (learning/interaction, job replacement, sociotechnical, and ethical/configuration anxiety). Suicidality was assessed via the Columbia–Suicide Severity Rating Scale (C-SSRS) as part of standard intake. These instruments were used as clinical tools, not research instruments administered outside usual care.

95% CIs were calculated using Wilson’s score method appropriate for small samples (n<30) [10].

AI-related stress affected nearly half of digitally distressed patients, transitioning from theoretical constructs to clinical realities. The finding that three patients reported worsened suicidal ideation following chatbot interactions represents a critical safety signal warranting further investigation.

Pre-existing anxiety and depression may amplify perceived chatbot rejection through negative cognitive biases. In several cases, patients described chatbot responses that they perceived as dismissive or invalidating, which intensified existing feelings of isolation or distress, consistent with the known role of perceived social rejection in suicidality. Digital literacy modulates AI encounter interpretation; lower literacy heightens anthropomorphisation and distress.

Unlike passive social media consumption [1], conversational AI creates bidirectional communication illusions. When systems hallucinate or fail to recognize distress, vulnerable individuals may interpret this as social rejection, a known suicidality risk factor representing a novel form of iatrogenic harm distinct from traditional technostress [8].

Findings align with AIAS dimensions [4] and recent technostress evidence [5,6], and resonate with generative AI safety reports [7,8]. Clinicians should screen for AI interactions in patients with unexplained mood deterioration. Future intake forms should incorporate validated AIAS items.

A small sample (n=28; n=12 AI-stressed) produces wide confidence intervals. Single-center findings may not generalize beyond similar resource-limited settings. Self-reported AI-symptom attribution lacks psychometric rigor, and causal inference is not possible from this retrospective, cross-sectional design. Selection bias is possible. AI stressor categorisation may oversimplify overlapping distress presentations. Despite these limitations, this audit provides preliminary evidence warranting rigorous prospective investigation with validated instruments and larger samples.

AI stressors are clinically relevant psychiatric presentations. Unregulated AI ’therapist’ tools pose iatrogenic harm demanding urgent regulatory attention. Future research should examine AI stress across larger, cross-national samples using validated instruments to establish prevalence, risk factors, and outcomes essential for evidence-based screening protocols and regulatory frameworks.