Clinical Takeaway
A smartphone-based AI relational agent (Woebot for Substance Use Disorders) was tested in a randomized controlled trial against a psychoeducational comparator to evaluate its ability to reduce problematic substance use in adults. The intervention is designed to provide accessible, scalable digital support for tracking and reducing substance use without requiring traditional clinical contact. Results from this trial add to earlier feasibility data by assessing whether treatment gains hold at one-month follow-up.
#18 A relational agent for treating substance use in adults: A randomized controlled trial with a psychoeducational comparator.
Citation: Prochaska Judith J et al.. A relational agent for treating substance use in adults: A randomized controlled trial with a psychoeducational comparator.. Journal of substance use and addiction treatment. 2026. PMID: 42035882.
Design: 5 Journal: 0 N: 2 Recency: 3 Pop: 2 Human: 1 Risk: -2
- Preclinical only
Abstract: BACKGROUND: Highly accessible and scalable, digital mental health interventions can reduce barriers associated with traditional treatment. Woebot for Substance Use Disorders (W-SUD) is a smartphone application in development that uses Woebot, a guided self-help relational agent intended to offer support to track and reduce problematic substance use. Two formative studies supported W-SUD’s feasibility, acceptability, and efficacy to reduce substance use at end-of-treatment (EOT). OBJECTIVES: In a randomized trial with a 1-month post-treatment follow-up, we aimed to evaluate W-SUD’s efficacy in reducing substance use occasions relative to a psychoeducational control. METHODS: U.S. adults (N = 258) with problematic substance use (CAGE-AID≥2) were recruited online and randomized to W-SUD or an email-delivered psychoeducational control. Primary and secondary outcomes were change in past-month substance use occasions from baseline to 8-weeks EOT and baseline to 1-month follow-up, respectively. RESULTS: The analytic sample (N = 202; 107 = W-SUD; 95 = psychoeducation) averaged 38.3 years of age (SD = 10.4) and was 53.5% (108/202) female and 71.3% (144/202) White. At baseline, participants averaged 33.1 (SD = 18.3) past-month substance use occasions with problematic substance use reported as alcohol (73.3%, 148/202), cannabis (30.7%, 62/202), stimulants (17.3%, 35/202), and cocaine (12.4%, 25/202). Anxiety (32.7%, 66/202) or depression (23.3%, 47/202) co-occurred. W-SUD participants sent a median of 322.5 (interquartile range: 59.8, 924.0) in-app messages during the intervention. There was no significant treatment effect on the primary (beta = -0.675, p = 0.741; 95%CI = -4.96-3.49) or secondary outcomes. Baseline to EOT within-group change in past-month substance use occasions were -13.6 (SD = 15.4, d = -0.879) and -12.9 (SD = 15.3, d = -0.847) for the W-SUD and psychoeducation conditions, respectively, indicating moderate to large changes in both, which was sustained at 1-m
What This Study Teaches Us
A smartphone app called Woebot designed to help people reduce substance use showed no significant benefit compared to standard psychoeducational materials in a randomized trial of 202 adults. Both groups improved somewhat over 8 weeks, but the app provided no added advantage.
Why This Matters Clinically
Digital interventions are increasingly promoted as scalable solutions for substance use disorders, but this rigorously designed trial found the relational agent approach didn’t outperform basic education. Clinicians considering recommending such apps to patients should know the evidence doesn’t yet support superiority claims.
Study Snapshot
| Study Design | Randomized controlled trial with 8-week intervention period and 1-month follow-up |
| Population | N = 202 U.S. adults (107 intervention, 95 control), mean age 38.3 years, 53.5% female, 71.3% White, with problematic substance use (CAGE-AID ≥ 2); 73.3% with alcohol use, 30.7% cannabis, 17.3% stimulants, 12.4% cocaine |
| Intervention | Woebot for Substance Use Disorders smartphone app (W-SUD) designed to provide guided self-help support for tracking and reducing substance use over 8 weeks versus email-delivered psychoeducational control |
| Primary Outcome | Change in past-month substance use occasions from baseline to 8-week end-of-treatment |
| Key Result | No significant difference between W-SUD and psychoeducation control (beta = -0.675, p = 0.741; 95% CI = -4.96 to 3.49) |
Where This Paper Deserves Skepticism
The abstract is truncated and doesn’t report within-group changes, making it impossible to assess whether either intervention actually worked or just engaged participants (median 322 in-app messages suggests engagement, but outcomes data are incomplete). Online recruitment of self-selected participants may skew toward higher motivation or internet savvy, limiting generalizability to real clinical populations. The heterogeneous substance use mix (73% alcohol but substantial cannabis and stimulant use) makes it unclear whether the null finding applies equally across drug types.
Dr. Caplan’s Take
This is a sobering but important null result from a reasonably powered trial by serious researchers. The fact that Woebot participants engaged substantially (over 300 messages sent) yet achieved no better outcomes than simple psychoeducation tells us that engagement alone and conversational warmth from an AI agent don’t translate to better substance use outcomes. We should be cautious about the narrative that smartphones and chatbots are magic bullets for substance use disorders, even when they’re well-designed and users find them acceptable.
Clinical Bottom Line
A well-designed trial found no advantage of a sophisticated relational agent app over standard psychoeducation for reducing substance use, despite good user engagement. Clinicians should not yet promote such apps as superior treatment options pending evidence of genuine added benefit.
|
Have thoughts on this? Share it:
