Artificial intelligence (AI) is finding its way into health apps, including tools designed to help people stop smoking. But an important question remains: does adding AI actually help more people quit cigarettes?
In our new systematic review and meta-analysis, published in the International Journal of Medical Informatics, we examined evidence from randomised controlled trials to find out. The results are promising in places, but they also show why enthusiasm for AI should not get ahead of the evidence.
Artificial intelligence-enabled interventions for smoking cessation: a systematic review and meta-analysis of randomised controlled trials
Read Full Paper →How is AI being used to help people quit smoking?
Not every app that sends a motivational message uses AI. Conventional digital programmes can follow fixed schedules or rules without learning from users.
The interventions in our review went further. Some used machine learning to recommend messages based on participants' responses. Others used conversational agents to interpret what people typed, smartwatches to detect smoking-related gestures, or predictive models to identify moments when someone might be at risk of smoking again.
What did our review examine?
We identified 14 randomised controlled trials involving 101,331 participants. Eight trials provided sufficiently comparable smoking-abstinence results for at least one meta-analysis.
The total number of participants needs context: one large commercial-app trial accounted for 97,164 of those randomised. Most other trials were much smaller, and not every trial measured long-term quitting in the same way.
We focused on abstinence at six months or longer and at approximately three months. We also examined whether results differed when quitting was confirmed using biological measures, such as breath carbon monoxide tests, rather than self-report alone.
Did AI improve smoking cessation rates?
In the main analysis, results tended to favour AI-enabled interventions, but the evidence was not statistically conclusive.
At six months or longer, the pooled risk ratio was 1.33 (95% confidence interval 0.82–2.15), based on four trials with 2,694 participants. At approximately three months, it was 1.36 (95% confidence interval 0.63–2.93), based on four trials with 715 participants.
Risk ratios above 1 favour AI-enabled interventions, but each confidence interval also includes no benefit.
These estimates suggest a possible benefit, but the ranges of uncertainty also include no benefit. The studies differed considerably in the AI technologies they used, the support given to comparison groups and how they measured quitting.
Why did some other analyses look more encouraging?
When we restricted the analyses to trials that biologically verified smoking abstinence, the results favoured AI: a risk ratio of 1.53 at six months or longer and 2.03 at approximately three months.
However, each analysis contained only three trials. Restricting the evidence to verified outcomes also excluded large trials with less favourable results, so the stronger estimates cannot be attributed to verification itself.
We carried out another check using two completed trials whose results were available only in trial registries. Adding them moved the longer-term estimate closer to no difference (risk ratio 1.20, 95% confidence interval 0.91–1.59).
Is AI better than ordinary smoking-cessation apps?
This may be the most important question. AI-containing programmes more consistently appeared helpful when compared with usual or minimal care. But that does not tell us whether the AI itself made the difference: the programme might simply have offered more support.
When AI tools were compared with active, non-AI digital interventions, results were inconsistent. Our review therefore could not establish that AI adds a reliable benefit beyond well-designed existing support.
What about ChatGPT-style quit-smoking assistants?
Despite growing interest in generative AI, none of the randomised trials in our review evaluated a generative large language model intervention. The evidence should therefore not be used to claim that ChatGPT-style assistants have been proven to help people stop smoking.
We also found that technical details and monitoring of potentially harmful AI responses were often poorly reported. More transparent studies are essential.
What should people take away?
AI may eventually make smoking-cessation support more responsive and accessible. For now, though, the certainty of the evidence is low to very low. Larger trials are needed to test sustained, biologically verified quitting and to compare AI against equally well-designed non-AI support.
For people trying to quit smoking today, AI tools should not be regarded as proven replacements for established smoking-cessation treatments.
Read our open-access study:
Adebisi YA, Alhur AA, Ogunkola IO, et al. Artificial intelligence-enabled interventions for smoking cessation: a systematic review and meta-analysis of randomised controlled trials. International Journal of Medical Informatics. 2027;223:106788. https://doi.org/10.1016/j.ijmedinf.2026.106788
Comments
Loading comments...
Leave a Comment
Share your thoughts on this article.