Users’ needs for a digital smoking cessation application and how to address them: A mixed-methods study

Data preparation

We preprocessed the gathered data by (1) using only data from sessions and the post-questionnaire if participants passed at least one attention check in the respective component, (2) using the first recorded submission for a study component if participants did the component more than once, (3) removing ratings and free-text responses for the interaction scenarios for people who wrote in their free-text responses that they could not see the video (N = 2), and (4) anonymizing free-text responses by removing potentially identifying or sensitive information. In addition, we computed the reliability of the items corresponding to the quitter, non-smoker, and physical activity identity measures. Since the reliability was sufficiently high for quitter (Cronbach’s $\alpha$ = 0.76, N = 671), non-smoker (Cronbach’s $\alpha$ = 0.69), and physical activity identity (Cronbach’s $\alpha$ = 0.89), we used the means of the items as index measures. We also reversed the scale for the TTM-stage for becoming physically active such that a higher value denotes a higher stage of change.

Analysis

We took a mixed-methods approach and proceeded in four steps to analyze the data. These steps were the thematic analysis steps described by Braun & Clarke (2006) with the addition of triangulation based on literature and quantitative results. We used two types of triangulation: method and investigator triangulation (Nancy Carter, Bryant-Lukosius & Alba DiCenso, 2014). Method triangulation was performed using data on both people’s actual behavior from their activity experiences and efforts as well as their views on possible behaviors from their free-text responses and ratings for the interaction scenarios. We also used data on user characteristics (e.g., physical activity identity) and findings from the literature. Regarding investigator triangulation, two researchers with different backgrounds were involved in all parts of the analysis. The result are the analysis steps that we now describe in detail.

Preparation of coding scheme. To create our coding scheme, the first author (NA) with a background in artificial intelligence and eHealth first familiarized herself with the data by reading all free-text responses and noting initial inductive codes. These codes were further refined deductively by looking through literature on technology acceptance and use, human motivation and behavior, and perceptions of virtual agents and robots. This included the two versions of the Unified Theory of Acceptance and Use of Technology (UTAUT) (Venkatesh et al., 2003; Venkatesh, Thong & Xu, 2012) (including their extensions with autonomy (Lakhal, Khechine & Pascot, 2013; Khechine & Augier, 2019), self-efficacy (Hewitt et al., 2019), and characteristics of the technology, situation, task, individual and other humans (Brown, Dennis & Venkatesh, 2010)), self-determination theory (Deci & Ryan, 1985), the Capability-Opportunity-Motivation-Behavior (COM-B) model of behavior (Michie, Van Stralen & West, 2011), barriers to behavior (Alfaifi, Grasso & Tamma, 2018), the findings by de Graaf, Allouch & Van Dijk (2015) on users’ experiences with a social robot, and the Ability-Benevolence-Integrity model of trustworthiness (Mayer, Davis & Schoorman, 1995). A draft coding scheme was discussed with SV who has a background in interaction design and had also read responses and formulated initial codes. The final coding scheme consisted of three levels, with four codes at the highest level, 15 codes at the second level, and 86 codes at the third level. Codes thereby captured both semantic and latent meanings of the responses (Braun & Clarke, 2012). The coding scheme is shown in Fig. S7.

Coding of free-text responses. All free-text responses were coded by NA based on the developed coding scheme. Multiple codes were used if relevant. We assessed the reliability using double coding. The second coder SV was further trained by independently coding six sets of ten responses and discussing the coding with NA after each set. Then, SV coded 100 responses. These 100 responses were chosen randomly but such that there were at least six responses per question (i.e., the 13 interaction scenarios, barriers, motivators, and activity experiences). The number of double-coded responses was selected to allow for an error margin of at most 10% to be obtained when calculating percent agreement (Gwet, 2014). We obtained moderate agreement (Cohen’s $\kappa$ = 0.41) (Landis & Koch, 1977) at the third coding level. Due to its more robust nature (Gwet, 2021), we also computed the Brennan-Prediger $\kappa$ (Brennan & Prediger, 1981) for a value of 0.97. Since participants primarily corrected spelling and grammar errors in their modifications of their activity experience answers, these modifications were excluded from further analysis.

Triangulation with literature and quantitative results. As NA and the third author (KP), with a background in psychology, gained insights from the coded free-text responses, literature and quantitative results were used to triangulate the qualitative results. Relevant literature came from diverse research fields such as eHealth, behavior change theories, human-robot interaction, and various application domains. Moreover, we incorporated two types of quantitative data. First, we computed means and Bayesian credibility intervals for the ratings per interaction scenario. The credibility intervals we report are the 95% HDIs. Second, we computed Spearman correlation coefficients between user characteristics (e.g., physical activity identity) on the one hand and users’ activity efforts and ratings for groups of interaction scenarios on the other hand. Note that we combined scenarios about similar interactions into groups to facilitate their discussion, as shown in Table S2. We conducted Bayesian tests for the correlations using the Bayesian First Aid R-package (Bååth, 2014) and report the median values and 95% HDIs. We classified the size of the resulting correlations using the guidelines by Cohen (1992). Furthermore, we calculated the posterior probability that a positive correlation is greater than zero and evaluated the probability using the guidelines by Chechile (2020).

Search, review and definition of themes, and production of the report. NA and KP examined the results to identify overarching themes. A final set of themes was obtained using multiple rounds of discussion. To produce the report, which is the last thematic analysis step described by Braun & Clarke (2006), NA selected participant responses that illustrate the themes. Participants are referred to by numbers (e.g., P123).

Perceived usefulness

The most frequent topics both overall and for the interaction scenarios and activity experiences revolved around the perceived usefulness of the behavior (Fig. 2).

Getting help, advice or tips and learning. Thinking that they would get help, advice, tips, or learn something by engaging in a behavior was the most frequent topic overall (13.97%) and for all interaction scenario groups except for the scenario about receiving motivational messages (Fig. 2). Participants’ concerns included whether the behavior would help to reach their goals (e.g., P283), teach them how to deal with cravings (e.g., P92), or serve as a prompt to reflect in general (e.g., P274) or on their current behavior (e.g., P507). Several participants also stated that they thought a behavior was (not) helpful without providing specific reasons for this evaluation (e.g., P224, P151). For example, some participants who were against involving their General Practitioner (GP) noted that they did not see any way in which their GP could help them (e.g., P345, P639):

• No i wouldn’t [consult my GP], i don’t think my GP could do anything to help. ( Consult GP in case of smoking relapse, P345)

Obtaining information or knowledge has previously been identified as a theme in participants’ thoughts on using a self-regulation-based eHealth intervention to increase physical activity and intake of fruit and vegetables (Poppe et al., 2017). It also plays a role in the context of eHealth applications for other domains, including self-management of chronic conditions such as chronic pain (Solem et al., 2019) and type 2 diabetes (Lie et al., 2017). The scoping review of Wilson et al. (2021) also showed that the opportunity to learn new information is a motivator for the use of eHealth tools by older adults. It has even been argued that gaining knowledge is such a crucial motivation for using online activities and applications that it makes users active consumers and producers of health knowledge (Lupiáñez-Villanueva, Ángel Mayer & Torrent, 2009).

Getting motivation or encouragement. One element participants were looking for in the behaviors was help in the form of motivation or encouragement, which was with 3.82% the second most frequent topic overall for “behavior” and with 40.85% the most frequent one for the interaction scenario about receiving motivational messages (Fig. 2). Concerns about receiving motivational messages included whether the messages would be tailored to the user and situation at hand (e.g., P212, P497), help to increase or maintain self-confidence (e.g., P158), or serve as a reminder of (reasons for) quitting smoking (e.g., P6, P25):

• I don’t generally respond much to motivational messages, but in this instance, anything that can serve as a reminder for why I want and need to do this so much, is a good thing. Being asked to reflect on our reasons for quitting is definitely a good thing and something I personally would benefit from. ( Receive motivational messages, P25)

Similar findings came to light regarding using a help button for High Risk Situations (HRSs). After pressing such a help button, the virtual coach would provide support for dealing with cravings or difficulties to do planned physical activity. Here 23.65% of responses referred to getting motivation or encouragement. Participants mentioned wanting to get motivation in general (e.g., P283, P617), be reminded of (reasons for) quitting smoking (e.g., P52, P291), or get the strength to resist cravings (e.g., P79). Some participants also explicitly mentioned the importance of being motivated by someone or something else (e.g., P57, P417, P636):

• … sometimes I need a bit more motivation than what’s going on in my head and need a little kick or nudge in the right direction ( Help button for PA HRSs, P636)

These findings coincide with work by Poppe et al. (2017), which found that the opportunity to be motivated by being reminded of one’s goals was a reason for participants to prefer a mobile application instead of a website for increasing physical activity and the intake of fruit and vegetables. Moreover, Kulhánek et al. (2018) found motivation strengthening to be a frequently mentioned benefit of a conversational agent that assists with quitting smoking.

Improving feelings or mood, novelty, and comparison with others. Besides getting help and motivation, several other topics related to the perceived usefulness of the behavior. This included, for instance, whether engaging in the behavior improved one’s feelings or mood (e.g., P50, P244), whether the behavior was novel (e.g., P236, P464), or whether the behavior made a comparison with others possible (e.g., P310). Giving users the option to compare their performance to others, for example, can be an effective motivation strategy in persuasive games (Orji, Vassileva & Mandryk, 2014), just as novelty can be motivating (Croes & Antheunis, 2021; Sadeghi et al., 2021). According to this novelty effect, people are initially curious about a new system or technology and have high expectations regarding its usefulness. Yet, these perceptions decrease over time as the system’s limitations become apparent. The novelty effect was implied by one participant who gave the following response when asked to describe their motivators for doing their preparatory activities:

• curiosity at first, but that waned ( Motivators, P338)

Recommendations for addressing the perceived usefulness of the behavior. The central role perceived usefulness plays for the acceptance of a system has been modeled by the UTAUT (Venkatesh et al., 2003), which posits that the extent to which a person thinks that using a system will lead to personal performance improvement influences the intention to use that system. This effect on behavioral intention has been illustrated in studies of technologies in diverse contexts such as an app for insomnia treatment (Fitrianie et al., 2021b) and a socially assistive robot (Fridin & Belokopytov, 2014). One approach to increase the perceived usefulness is to tailor advice or content in general to users’ preferences. Previous work has, for example, attempted to automatically optimize suggested activities for older adults (Costa et al., 2016), physical activities for adults (Rabbi et al., 2015), or breast cancer screening recommendations (Ayer et al., 2016). Yet, future work on such automated approaches should take the novelty effect into account. An alternative is to use a participatory approach in which potential users can contribute to the design (Davis et al., 2021). Yet, our participants often drew conclusions about the usefulness of an intervention component without having tried it. This suggests that other elements such as telling people that the content is meant specifically for them (Dijkstra, 2016) or explaining in more detail why and how something helps them (Horsch et al., 2015) may be necessary.

Lastly, since getting motivation or encouragement was frequently sought out in our study, special attention should be given to how participants can be motivated or encouraged. Receiving motivational messages was overall seen positively in our study (Fig. 3), but we saw earlier that several people were concerned about the format of these messages. In addition, participants were also looking for motivation in other interactions, such as a discussion with their virtual coach about their repeated failure to reach their physical activity goals (Fig. 2).

Timing and intensity

The timing of the behavior is interesting in that it appears to play a role primarily for one type of interaction, namely, regular planning and reflection for HRSs. There appeared to be no consensus regarding an ideal time for this. For instance, some participants liked the proposed day (e.g., P470, P487) for physical activity- or time of the day (e.g., P121, P493, P627) for smoking-related planning and reflection. But others found the suggested timing for planning and reflection to be inappropriate:

• Maybe [I would make a plan with my virtual coach] on a different day, e.g friday for the week ahead starting on Monday. I do like the planning of the week and think ahead of the hurdles to avoid them. ( Plan for PA HRSs on Sundays, P262)

• I think it’s great idea [to make a plan with my virtual coach,] just not sure if I would have enough time in the morning to do so (I start work at 6.30 a.m. so that would have to be quick plan) ( Plan for smoking HRSs in the mornings, P113)

Besides timing, participants were also concerned about the frequency (e.g., P158, P178) and duration (e.g., P362, P552) of (potential) application components.

Finally, while some participants favored the regularity of the planning and reflection interactions (e.g., P462), others expressed their preference for on-demand rather than regular support:

• I think consulting the virtual coach might be helpful – but the option to let the app know whenever I have a craving would be more convenient. ( Reflect on smoking HRSs in the evenings, P100).

Recommendations for addressing the timing and intensity of the behavior. Previous work on needs, barriers, and facilitators for older adults to use eHealth applications found instant availability of help through such applications to be a facilitator (Vergouw et al., 2020). However, given the diversity of user needs and preferences, there appears to be a need for tailored timing and intensity of intervention components. The importance of personalizing these elements of health behavior change applications has previously been pointed out by Dijkstra (2016). Approaches to achieving such personalization include both letting participants choose themselves and automatically determining opportune moments (Horsch et al., 2017). Letting participants choose acknowledges that users may know best what time suits them and that supporting autonomy and competence may increase motivation and performance (Ryan & Deci, 2000). In contrast, determining opportune moments for the user aims to account for the fact that people’s self-knowledge is generally not very accurate (Vazire & Carlson, 2010; Dunning, 2012). A middle ground may be to support users in systematically finding out what suits them (Karkar et al., 2016).

Importance of and motivation to change

Importance of change. When it comes to physical activity, several participants pointed out that becoming (more) physically active was not important to them. Reasons included already being physically active (e.g., P143, P533) and not seeing how physical activity helps to quit smoking (e.g., P657). Some participants, for instance, thought that working simultaneously on becoming physically active and quitting smoking was too difficult:

• It’s a bit condescending. Giving up smoking is hard enough without having to do a fitness regime also. ( Follow PA program while quitting smoking, P464).

Notably, this was even though participants were informed about the potential positive effects of physical activity on quitting smoking at the start of the study as well as every time they were assigned an activity for physical activity increase. Participants who were in favor of physical activity-related behaviors, on the other hand, frequently pointed out the benefits of physical activity both in general (e.g., P245, P335) and for quitting smoking specifically:

• I really think that physical activity could help me to quit smoking forever ( Motivators, P543)

The TTM provides a framework for assessing participants’ views on the importance of changing a behavior. Stage one, the precontemplation stage, is one in which people do not aim to change their behavior within the next six months and are often un- or under-informed about the consequences of their behavior (Prochaska, Redding & Evers, 2015). As people progress through the stages, they become increasingly aware of the consequences of their behavior(s) and ready to take action. The relationship between TTM-stage and taking action was observable in our study. More precisely, we observed a small correlation of 0.21 between people’s TTM-stage for becoming physically active and the effort they spent on their activities (Table 1). Hence, participants in higher stages overall spent somewhat more effort on their activities. Notably, we find this association between the TTM-stage for becoming physically active and the effort spent on activities even though only half of the activities were targeted at physical activity.

Motivation to change. Once participants have become aware of the importance of changing their behavior, they also need to be motivated to do so. In our study, people’s motivation to reach their goals was with 50.00% the most commonly reported motivator for completing their activities, making it the overall most frequently mentioned topic for the “user” unit (Fig. 2). Goals which motivated participants included quitting smoking (e.g., P403, P463) and becoming more physically active (e.g., P126, P142), but also more general goals such as improving one’s health (e.g., P224, P390) and very individual ones such as being able to hold one’s breath for longer (P624). Sometimes, participants explicitly linked these other goals to quitting smoking or becoming more physically active:

• I was motivated by exploring all the positive health, family and psychological benefits that comes with quitting. And it fuels my drive to do the assigned activities. ( Motivators, P521)

• my current health is not good so to try to improve this i need to give up smoking and do more exercise. This is my main motivation ( Motivators, P493)

At the same time, not wanting to quit smoking was mentioned by some participants as a barrier to doing their activities (e.g., P455, P591). Moreover, several participants were not confident in their ability to reach their goals (e.g., P111). These findings resonate with ones by Milcarz et al. (2019), according to which difficulties in quitting smoking and a lack of willingness to quit were the most commonly mentioned barriers.

Besides conscious or reflective motivation, automatic motivation also plays a role. Automatic motivation consists of automatic processes involving, for example, emotional reactions and reflex responses such as feeling excited at the prospect of going running in the evening (Michie et al., 2014). One way to capture people’s automatic motivation is to look at their self-identity (Michie, Van Stralen & West, 2011; West, 2013), which has been shown to predict people’s (intention and motivation to perform a) behavior (Priebe et al., 2020; Wolstenholme et al., 2021). In our results, we find some influence of self-identity on behavior based on a small correlation of 0.25 between physical activity identity and the effort people spent on their activities (Table 1). Thus, it appears that participants with stronger physical activity identities spent more effort on their activities. In addition, physical activity identity was also positively related to people’s willingness to follow a physical activity program besides quitting smoking, receive motivational messages, and involve their GP in the quit smoking process (Fig. 4). This is interesting, as some of these behaviors do not involve physical activity. One explanation could be that people with stronger physical activity identities felt more involved in the intervention as a whole due to the combination of quitting smoking and becoming physically active. One participant, for instance, pointed out the importance of being willing to become more physically active to be a suitable participant for the intervention:

• I’m fairly positive about this question because considering some of the activities were physical it would be important that the person was willing to be more physically active. ( Follow PA program while quitting smoking, P204)

This suggests that participants may need to be separately motivated to reach all goals an intervention puts forward. Notably, the association between quitter or non-smoker self-identities and the effort participants spent on their activities is much weaker than in the case of physical activity identity (Fig. 4). This likely is the case because participants had to be in either the contemplation or the preparation stage of the TTM for quitting smoking to be eligible for the study. As such, their identity as quitter or non-smoker may have been quite weak because not (yet) fitting with their (readiness to change their) smoking behaviors.

Recommendations for addressing the perceived importance of change and motivation to change. Our results show that as a first step, people need to be able to link a behavior to a desired outcome. For example, in this study, the benefits of physical activity for quitting smoking may have needed to be clearer. Our findings resonate with ones by Poppe et al. (2017), who showed that participants did not always agree with provided information on the positive effects of physical activity and fruit and vegetable intake. Participants who are not aware of the consequences of their behavior are often described as being in the precontemplation stage of the TTM. Important processes of change in this stage include receiving information about the behavior, evaluating how the problem affects one’s environment, and experiencing emotions about one’s situation and problems (Prochaska, Redding & Evers, 2015). Once people have learned about the consequences of their behavior, a next process of change may be self re-evaluation or realizing that the behavior change matches one’s identity better (Prochaska, Redding & Evers, 2015). Several participants of our study mentioned, for example, that an activity asking them to think about their feared or desired future self concerning quitting smoking or becoming more physically active was a major motivator (e.g., P243, P263):

• The task where I was to think about what I would be in the future gave me a huge boost for the rest of the tasks ( Motivators, P592)

Thinking about oneself in this way may weaken identity related to a problem behavior and strengthen identity related to one’s non-problem behavior (i.e., non-smoking if the problem behavior is smoking). Consequently, as identity represents a strong form of automatic motivation, such a change in identity can lead to successfully changing (health) behaviors such as smoking (Williams et al., 2002; Park et al., 2021). In addition, people need to gain confidence to change their behavior. Ways to increase confidence include allowing users to make small wins (Amabile & Kramer, 2011), letting them observe a relatable other succeed (Bandura, 1977), verbally persuading them (Strecher et al., 1986), or improving their mood (Kavanagh & Bower, 1985).

Autonomy

A topic that appeared several times in participants’ responses was autonomy (Fig. 2). For example, one participant perceived asking for help in a physical activity HRS as being a sign of excessive dependence on others, especially since they regarded the situation as not serious enough to require help:

• If I needed help like this just because I wasn’t exercising one evening I would feel very concerned that I could no longer stand on my own two feet in life. This is going too far. We are adults and this kind of hand holding is not healthy or helpful. ( Help button for PA HRSs, P133)

While a need for autonomy was put forward in several different interactions, the percentage of times this topic appeared was with 8.39% notably high in the context of involving one’s GP (Fig. 2). Participants mentioned wanting to decide for themselves whether to involve their GP instead of following the advice of the virtual coach (e.g., P429, P528). They also mentioned wanting to keep trying to change their behavior themselves (e.g., P472, P660), or only contacting their GP in severe cases (e.g., P85, P539). One participant also emphasized that seeing a GP is not helpful unless one is also motivated to quit smoking:

• … I believe that quitting smoking comes entirely from self motivation. Even if I would go to my GP, if I’m not personally committed it would be just a waste of time and money ( Consult GP at start of quit attempt, P624)

According to self-determination theory, people need to feel in control of their behaviors and goals to initiate behavior (Ryan & Deci, 2000). As such, the proposal to involve a GP likely violated this need. Besides our study, other eHealth interventions have identified autonomy and independence as needs for long-term weight maintenance (Asbjørnsen et al., 2020), healthy living for cardiovascular disease prevention and rehabilitation (Breeman et al., 2021), and self-management of chronic pain (Solem et al., 2019). The latter study also explicitly reported the difficulty of finding a balance between asking for help and being independent. Interestingly, however, our participants’ overall stance on involving the GP was also much more negative than for the other proposed interactions, with the mean of the credible interval even being less than zero for contacting the GP at the start of the quitting process (Fig. 3). This contradicts findings of previous work, which show that smokers tend to accept unsought conversations about smoking with their GP (Lems, 2006). Therefore, it was likely not the involvement of a GP alone that concerned participants in our study, but rather the way and personal situation in which involving a GP was proposed.

Recommendations for addressing users’ autonomy. Our results suggest that when a virtual coach recommends help for participants, their need for autonomy can be violated. Thereby, it appears to be crucial to not only consider what is recommended but also how and when. Interesting work in this regard has been conducted by Tielman et al. (2019). Their model for referring patients to human care is based on a combination of willingness to see a human and severity of the situation. If the situation is not severe and a patient is not in favor of seeing a human, it may be better not to actively try to persuade a person to see a human. The reasoning behind this is based on social judgment theory (Sherif & Hovland, 1961), which posits that any recommendation made to a person who is against a suggestion will likely fail and make the user more opposed to the idea. Besides considering people’s willingness to contact a human and the severity of the situation, it may also be useful to use a different formulation for the recommendations. For example, more emphasis could be placed on persuading people using, for instance, testimonials (Dijkstra, 2016). Alternatively, one could explain how the recommendation is in line with a user’s values (Dijkstra, 2016), or formulate utterances more carefully, as suggestions and less as commands (Free et al., 2009).

Personal characteristics

Besides people’s self-identity and TTM-stages, we also looked at the effects that their smoking frequency, weekly exercise amount, age, education level, or personality may have (Fig. 4). While previous work suggests a relationship between socioeconomic status and smoking (Fernández et al., 2006, Hiscock et al., 2012) and physical activity behaviors (Scholes & Neave, 2017), as well as between age and smoking behaviors (Kviz et al., 1994; Li, Hsia & Yang, 2011), our results do not suggest a strong effect of education level and age on people’s activity efforts and ratings of the interaction scenarios (Fig. 4). What we did observe is a moderate correlation of 0.41 between smoking frequency and willingness to discuss repeated failures to reach physical activity goals (Table 1). One explanation for this observation could be that as heavier smokers are more likely to seek help in quitting smoking (Chaiton et al., 2007), they may be more open to receiving support. In addition, we find several small to moderate correlations between people’s Big-Five personality dimensions and their activity efforts and interaction scenario ratings (Fig. 4). For example, there is a small correlation of 0.20 between conscientiousness and activity effort (Table 1). This matches the observation that the code “conscientiousness” appeared several times in relation to the preparatory activities (e.g., P320, P380):

• I approach the activities with clear focus and intense dedication and discipline ( Activity experience, P521)

Moreover, extraversion correlated with people’s willingness to involve both their social environment and GP as shown in Table 1. This is in line with the observation that, especially with regards to involving their social environment, several participants expressed their desire for privacy (Fig. 2). While some participants wanted privacy in general (e.g., P495, P573), others were concerned explicitly with not wanting to be seen failing:

• I would prefer to keep it to myself until im confident that i can kick this habit for good ( Tell SE about quit attempt, P111)

• … I have done that [i.e., tell my social environment about my quit attempt] before and felt pressure to stop. Then you feel like a failure if you dont succeed ( Tell SE about quit attempt, P79)

Overall, however, the effects of personality were relatively small. This is in line with previous work, which found primarily small effects of personality on physical activity (Rhodes & Smith, 2006; Rhodes & Boudreau, 2017) and smoking behaviors (Terracciano & Costa, 2004; Munafo, Zetteler & Clark, 2007).

Recommendations for addressing users’ personal characteristics. Our results are indicative of personal characteristics playing a role in people’s preferences for using an eHealth application. The consequence is that it may be harder to convince some participants of the merits of certain behaviors, such as involving their social environment. However, our observed effects were primarily small, and the topics “want for privacy” and “conscientiousness” appeared relatively infrequently in the free-text responses (Fig. S9). Therefore, rather than directly tailoring application components to people’s characteristics, the most straightforward approach may be to simply acknowledge that people differ in their preferences and leave room for these differences. Interestingly, we find that, even when not explicitly asked to, people tend to follow their preferences. For example, several people modified a preparatory activity in which they were asked to visualize smoking or becoming more physically active as a fighting match. Instead of a fighting match, they imagined a bike race (P104), a soccer match (P539), or a verbal fight (P87). While not all people provided reasons for doing so, P539, for instance, mentioned imagining a soccer match because of being a soccer fan.

Difficulty of integrating (health) behaviors into people’s busy lives

People are busy with other things. The most frequently mentioned topic with regards to the environment was with 5.04% whether participants had enough time to perform a behavior, and especially to complete their preparatory activities (Fig. 2). Notably, 18.80% of barriers to completing the activities involved the availability of time, making it the most commonly mentioned barrier (Fig. 2). It turned out that participants tended to be busy with their daily lives, including work, child care, and daily chores, and that these tasks left no time (e.g., P111, P432), caused people to be too tired (e.g., P66, P140, P262), or made it difficult for people to focus on their preparatory activities (e.g., P495, P600, P642):

• I run out of time with home life taking a focus so didnt get time to complete ( Activity experience, P432)

• It was hard to plan the exercise in my daily planning, as the days are full and i’m very tired at the end of the day ( Activity experience, P262)

• I thought about it [i.e., the person I would like to be once I have successfully quit smoking] for a couple minutes with the intention to write it down, but got distracted with other things. It did remind me that I did want to quit though. ( Activity experience, P642)

When participants had no time to do their activities, they reported having spent their time on other priorities such as home life in the example of P432. A similar phenomenon was found in Lie et al. (2017) where people dropped out from an eHealth intervention for self-management of type 2 diabetes because the daily life took the front stage. Similarly, a scoping review by Wilson et al. (2021) saw that the inability to incorporate a behavior into one’s routine was a barrier to using eHealth tools. In addition, the statement by P262 (i.e., being too tired to do a behavior) suggests that the participant expected the behavior to require a considerable amount of effort, which was also one of our codes for the “behavior” unit (Fig. 2). The expectation of effort or effort expectancy is one of the predictors of the intention to perform a behavior according to the UTAUT (Venkatesh et al., 2003), and several studies have confirmed this relationship in the context of eHealth tools (e.g., Boontarig et al. (2012), Quaosar, Hoque & Bao (2018), Fitrianie et al. (2021b)).

Prompts and triggers are helpful. The second most frequent topic for “environment” was with 3.49% of all responses whether participants had prompts or triggers for doing a behavior, and especially to complete a preparatory activity (Fig. 2). Participants commonly reported that they forgot to do (part of) their preparatory activity (e.g., P269, P331). Reasons included being busy (e.g., P43, P527), and that some activities (e.g., tracking one’s smoking behavior) required one to remember to do something at specific times (e.g., P182, P227):

• I approached the activity with a positive thought but found myself forgetting to record the timea i had a cigarette. Thia was mainly due to smoking when i had opportunity for a quick break so was always rushing. ( Activity experience, P182)

Participants also commonly started an activity but then stopped to do something else and forgot to get back to it:

• I watched 1 minute but I started different activity and forgotten about it. ( Activity experience, P281)

The importance of prompts and triggers is further illustrated by the fact that several participants who did complete their activities reported making use of them. Participants mentioned completing their activity right after the session had ended (e.g., P186, P480), once they received the reminder message we sent after the session on Prolific (e.g., P180, P393, P417), or based on a reminder they had set themselves (e.g., P7, P346):

• I set a remind on my smartphone to recall me the activity, so yesterday, in my bed before sleeping, I thought about who I want to be once I have quit smoking ( Activity experience, P346)

A scoping review by Wilson et al. (2021) found a lack of reminders or alerts to be a barrier for older individuals to use eHealth tools and their presence to be a motivator. Similarly, Horsch et al. (2015) saw that people favored the use of reminders to help with forgetting in the context of insomnia treatment. Participants of the study by Horsch et al. (2015) also emphasized, however, that users themselves should set reminders. This links to the earlier discussed topic of autonomy in that users want to be in control of application components.

Recommendation for addressing the difficulty of integrating (health) behaviors into people’s busy lives. According to the COM-B model of behavior (Michie, Van Stralen & West, 2011), one predictor of behavior is whether people have the opportunity to perform the behavior. This includes sufficient time and prompts or triggers to remind them. Our results suggest that both of these factors tend to be lacking. A straightforward way to help people who lack time is to create action plans. Action plans specify where, when, and how one plans to do something to create a link between a cue and a behavioral response (Hagger & Luszczynska, 2014). Action plans have been effective in changing behaviors such as physical activity, smoking, and alcohol consumption (Sniehotta et al., 2005a; Hagger & Luszczynska, 2014). To further account for sudden barriers to doing a behavior such as being tired, one can specify coping plans to create a link between a possible risk situation and a feasible way of coping with it (Sniehotta et al., 2005b).

Another strategy could be to prompt participants to create reminders themselves or determine suitable times for sending automatic reminders (Horsch et al., 2017). This was actually recommended by one participant ( Activity experience, P247). It turns out that reminders are already one of the most common persuasive components of eHealth applications (Lentferink et al., 2017). However, it is important to keep in mind that a high effort expectancy and other more relevant priorities likely also play a role for somebody who is too tired or has no time to do a behavior. This shows that characteristics of the environment can be intertwined with ones of the behavior and the user.

Lastly, the topics “having enough time” and “having prompts or triggers” primarily appeared in participants’ statements about their actual behavior (i.e., their activity experiences and barriers) rather than their views on possible behaviors in the form of the interaction scenarios (Fig. 2). Thus, these factors are less evident to people when they are just asked about their views on possible behaviors. One likely explanation for this is the optimism bias, according to which people tend to be overly optimistic about themselves and their future (Weinstein, 1980). For example, the study of Horsch et al. (2015) showed that people tended to be rather optimistic about their future adherence to an eHealth application for insomnia treatment. Reasons may include relying too much on future willpower and ignoring things that could go wrong (Horsch et al., 2015). This underlines the importance of having participants interact with a system to get a thorough assessment of their needs (Canada Health Infoway, 2013). On the other hand, other topics, namely ones related to the other party, primarily appeared in the views on the interaction scenarios rather than statements about actual behavior (Fig. 2). Thus, combining data on actual and potential behaviors offers a clear benefit.

Helpfulness of support from social environment

We have already touched upon the role of the social environment in the context of characteristics of the other party that influence people’s views on interaction scenarios. However, our results also suggest the general importance of support from one’s social environment. In fact, 6.60% of participants reported support from their social environment to be a motivator for doing their preparatory activities (Fig. 2). The social environment supported these participants in their wish to reach their goals such as quitting smoking (e.g., P212, P475) and helped them to complete the activities:

• Wrote down on a list last night and discussed with partner. Helpful and motivating. ( Activity experience, P133)

An important form of support is not just verbal but also behavioral. Some participants felt less motivated because their social environment did not live according to their own behavioral goal of not smoking (e.g., P127, P207, P315), or because they did not feel part of a group that worked toward the same behavioral goal (e.g., P262):

• Probably the people watching the ad will think about the fact that usually smokers need motivation from other people, but often the motivation is not there because smokers tend to surround themselves with other smokers. ( Receive motivational messages, P207)

• … As well as the motivation of other participants. When feeling you are a part of a group that want to achieve the same thing i feel that this is more motivated. ( Barriers, P262)

Previous work by Willemsen et al. (1996) found social pressure from the social environment of Dutch employees to be a predictor of intention to quit smoking, and Breeman et al. (2021) concluded that involving the social environment was a desired core attribute for an eHealth application for healthy living. In addition, Meijer et al. (2016) note that the support and social norms present in social environments can shape identities, which in turn can affect behavior as discussed previously. For example, according to the social identity model of recovery (Best et al., 2016), a person’s recovery identity in the context of addiction can become stronger if it is shared with other people who favor recovery.

Recommendations to address the helpfulness of support from one’s social environment. We find that support from one’s social environment can be motivating and can help to perform activities that are part of a behavior change intervention. A straightforward way to include social support in an eHealth intervention is to prompt participants to either tell their social environment about their behavior change process or discuss with a significant other about how they can support it. Both of these elements were generally seen positively by our participants (Fig. 3). However, it is important to keep in mind the earlier discussed personal characteristics such as a want for privacy and characteristics of the social environment that may influence whether people want to involve their social environment. Taking a similar approach to Tielman et al. (2019) and taking people’s willingness to involve a human and situation severity into account may be beneficial. Another approach may be to connect people who work toward the same behavioral goal. Promising results can be obtained with relatively simple solutions such as a WhatsApp group (Simons, van den Heuvel & Jonker, 2018). Such a group has the advantage that it is easy to implement and accessible due to the omnipresence of WhatsApp (Simons, van den Heuvel & Jonker, 2018).

Diversity of other environmental factors

Several other environmental factors such as not having access to the Internet on certain days (P233), not using one’s phone when at home or on specific days (e.g., P376, P624), not being able to access the videos contained in the activities due to one’s location (P668), poor weather (P468), or restrictions related to the COVID pandemic (e.g., P4, P631) were mentioned. Given the diversity of environmental factors that can play a role, it is likely difficult, if not infeasible, to specifically address all of them. Since such other environmental factors appeared relatively infrequently (Fig. S9), the most important insight may thus be to design an eHealth application in such a way that it leaves room for individual barriers and gives users resources to try to cope with these barriers themselves.