In 2019, alcohol consumption was associated with 2.6 million deaths and 116 million disability-adjusted life years lost [1]. It is a leading global health concern due to its strong association with chronic diseases, including several forms of cancer. Epidemiological evidence indicates a dose-response relationship between alcohol intake and cancer risk: the more alcohol consumed, the higher the probability of developing cancer [2]. Exceeding weekly alcohol intake thresholds of 200 g significantly reduces life expectancy by 1 to 2 years, with reductions of up to 5 years at consumption levels exceeding 350 g per week [2].

In Germany, alcohol consumption remains widespread despite modest declines over recent decades. In 2017, the average per capita consumption among individuals aged ≥15 years was 10.5 L of pure alcohol per year [3]. This places Germany among high-consumption nations, in contrast to countries such as Norway or Greece, where per capita consumption is 6 to 6.5 L of pure alcohol. Evidence-based alcohol control policies, such as taxation, restricted availability, advertising bans, and public education, are associated with reductions in population-level alcohol intake and related adverse health effects [4,5]. However, Germany continues to display a relatively permissive approach toward alcohol as a “cultural good,” contributing to persistently high consumption levels [4].

Findings in health and clinical psychology highlight that addictive behaviors, including alcohol misuse, are strongly influenced by automatic cognitive processes that operate largely outside conscious awareness [6]. According to dual-process models of behavior, human actions result from the interplay between fast, implicit (automatic) and slow, explicit (controlled) cognitive systems [7,8]. In addiction, this balance is disrupted, with excessive impulsive responses toward substance-related cues and diminished cognitive control [9,10]. Consequently, individuals may continue substance use despite consciously recognizing its negative consequences, as automatic approach biases drive craving and relapse [11].

The approach-avoidance task (AAT) has been developed to measure automatic approach and avoidance tendencies toward specific stimuli [12]. In this paradigm, participants push or pull a joystick in response to images of various orientations, thereby simulating avoidance or approach movements. Reaction time (RT) differences between push and pull trials indicate an underlying bias, that is, faster push or pull reactions dependent on image content. The AAT has been validated across multiple domains, such as fear of spiders [12] and social anxiety [13], and later adapted for addiction research; alcohol-dependent individuals typically exhibit stronger approach biases toward alcohol-related stimuli than toward neutral cues [14], while similar biases have been observed for cannabis-related [15], nicotine-related [16], and heroin-related [17] stimuli.

The AAT paradigm has been successfully extended into therapeutic applications. Such AAT-based neurocognitive training, also known as cognitive bias modification, aims to retrain automatic action tendencies by repeatedly pairing alcohol-related stimuli with avoidance responses. Clinical studies have demonstrated that such training, when added to standard inpatient treatment for alcohol dependence, significantly reduces relapse rates for up to 12 months after discharge [18]. One limitation of these existing studies is, however, that they rely on laboratory-based setups, requiring special response joysticks. This increases participant burden, restricts accessibility, and ultimately limits scalability.

To address these limitations, recent developments have focused on adapting AAT-based training for mobile platforms, enabling real-world use and repeated self-administered interventions. Mobile AAT app versions using touch screen gestures [19,20] or motion sensors [21] have demonstrated technical feasibility and user engagement. A mobile neurocognitive training app may therefore represent a scalable, low-threshold tool for measuring and modifying automatic approach biases in both clinical and preventive contexts, potentially broadening access to evidence-based behavioral interventions beyond traditional treatment settings.

The aim of the present feasibility study was to determine whether automatic approach tendencies toward alcohol stimuli can be reliably measured “at home” using an app-based AAT and whether systematic avoidance training with alcohol-related stimuli can modify these tendencies. Specifically, the study sought to assess the reliability of the app’s measurements, the influence of image content on automatic action tendencies in both clinical and nonclinical samples, changes of action tendencies with repeated training, and the usability of the app.

The current pilot study investigated the feasibility of using a smartphone-based AAT (alcohol AAT) to measure and modify automatic response tendencies toward alcohol-related stimuli. The study demonstrated several key findings. First, the smartphone’s motion sensors provided reliable and stable measurements across sessions, supporting the feasibility of the app-based assessment. Second, the app detected systematic reaction tendencies in clinical patients, with faster push-away responses to alcohol-related images (indicative of avoidance) and higher acceleration when pulling alcohol-related images toward themselves (indicative of approach tendencies). Third, similar systematic tendencies were observed in a nonclinical sample of soccer fans with overweight and risky alcohol consumption. Fourth, app training increased avoidance tendencies toward alcohol-related stimuli, particularly among participants with risky baseline alcohol consumption as measured by the AUDIT-C.

The rationale for this investigation is grounded in the recognition that many addictive behaviors, including alcohol use, are strongly influenced by automatic cognitive processes, which operate largely outside conscious control [6]. These automatic processes can drive consumption despite awareness of negative health consequences, a phenomenon that has been conceptualized using dual-process models of behavior [7-9]. Previous clinical applications of the AAT paradigm have primarily focused on patients with alcohol dependence, where neurocognitive training was used as an adjunct to standard therapy. In these studies, patients were trained to avoid alcohol-related stimuli, which resulted in measurable changes in approach-avoidance tendencies and reductions in relapse rates over 12 months [18]. However, these interventions were limited to stationary computer setups using joysticks, restricting applicability outside clinical or laboratory environments. Mobile or app-based implementations have been largely limited to touch screen or keyboard simulations that fail to capture genuine approach and avoidance movements [19,20]. The current study addresses this limitation by using a mobile app capable of measuring actual arm movements, thereby extending the technical and conceptual scope of the AAT paradigm [21]. Despite ongoing debates regarding the precise underlying mechanisms of approach and avoidance tendencies, there is a consensus that approach motivations prompt individuals to reduce physical distance to a stimulus, whereas avoidance motivations drive them to increase it [24,25]. In contrast, RT, the sole measurement in previous studies, is defined as the time between picture onset to the first movement of the smartphone; some type of effect anticipation is needed to use RT as an implicit measure of approach or avoidance. It might be conceptually clearer to assess both RT and acceleration, as many AAT studies intermix these 2 response dimensions by defining RT as the time between picture onset and movement completion.

Empirically, it remains unclear what this specific mobile acceleration task app measures. Response force has been shown to be closely related to motivation strength in humans [26] and is traditionally used to measure approach-avoidance motivation in animals [27-29]. The results of the present study suggest that it might indeed be related to approach or avoidance reactions, as are results of other studies [21,30]. However, the small correlation between the two measures (RT and acceleration) suggests that they may not simply represent 2 different ways of measuring the same construct. The validity test (study 2) with inpatients in treatment for AUD did not find an alcohol approach bias with the RT measure, which contrasts with many previous studies [18,31,32]. This can be a measurement artifact but could also be a reflection of active coping, motivational state, and/or task implementation (eg, using a relevant vs irrelevant feature task). Other studies also found an alcohol avoidance bias in early treatment inpatients using a relevant feature task [33], which was explained by interference between conscious motivation to stop drinking and automatic processes. This interference might be smaller with the acceleration measure, although such an assumption needs further empirical validation.

Significant limitations of this proof-of-concept study are, of course, the small sample sizes and the highly selective convenience samples. No control groups were recruited, and no sham training was included. Some of the results were unexpected and need to be treated with high caution, for example, faster RTs if instruction was “push alcohol.” Furthermore, while RT-based tendencies have a long history on approach-avoidance tendencies, acceleration-based tendencies are a relatively novel measure; therefore, these results should be interpreted as exploratory and potentially unstable. Another limitation relates to the assignment of training dose in study 3, which occurred at the club rather than individual level, potentially introducing clustering and selection effects. Most importantly, the present proof-of-concept study did not assess changes in actual drinking behavior; therefore, the app cannot be seen as clinically relevant or effective.

While the pilot study does not yet impact clinical practice, the smartphone-based approach offers a promising way for preventive interventions targeting larger populations. In nonclinical samples, the app could potentially be used for primary prevention to maintain low-risk drinking behaviors and abstinence or for selective prevention and early intervention to reduce risky alcohol consumption.

In conclusion, the present study provides initial exploratory evidence that a smartphone-based AAT can reliably measure and modify alcohol-related automatic response tendencies. By enabling scalable, home-based delivery of AAT training, this approach has the potential to complement traditional prevention and intervention strategies and extend their reach to individuals at risk of developing alcohol-related problems. Future research should focus on randomized controlled trials to evaluate whether modifications in automatic approach-avoidance tendencies translate into sustained behavioral change and reduced health risks in diverse populations.