Research on persuasive technology has been introduced relatively recently [27,28]. Briñol and Petty [29] outline persuasion as follows: “In the typical situation where persuasion is possible, a person or a group of people (ie, the recipient) receives an intervention (eg, a persuasive message) from another individual or group (ie, the source) in a particular setting (ie, the context).”

Persuasive systems may be defined as computerized software or information systems designed to reinforce, change, or shape attitudes or behaviors or both without using coercion or deception [26]. Successful persuasion takes place when the target of change (eg, attitudes or beliefs) is modified in the desired direction [29].

Changing people’s behavior is at the heart of health promotion. An individual’s behavior has a significant impact on, for example, cancer and heart disease, which are common causes of premature mortality. The Internet is transforming health care [30] and can be seen as a prime candidate for the application of key behavioral science theories and principles to promote healthier behaviors. There are several advantages in Internet-delivered interventions, for example, reducing cost and increasing convenience for users, reduction of health service costs, reduction of stigma and isolation of users, the need for timely information, and increased user and supplier control of the intervention [31-33]. Internet-based resources, in particular the Web, have many of the characteristics necessary for persuasive communication, and they may provide a channel which integrates the positive attributes of interpersonal and mass communication [34,35]. Web-based systems can give immediate feedback and match the information with the respondent's level of awareness, beliefs, and motivations at that particular time [10]. Additionally, Web-based interventions may overcome isolation of time, mobility, and geography. It has to be noted that Web-based interventions still may be no substitute for face-to-face contact [31,36].

Examples of persuasive technology can be found quite easily, as there are a variety of websites promoting healthier lifestyles. One of the strongest domains of innovation for persuasive technology in the near future will be preventive health care [37]. On a par with health behavior change, persuasive technology has the potential for significant breakthroughs in many areas of human well-being, such as education and environmental conservation. Nevertheless, the use of persuasive technology in the health arena is still in its infancy. While the field is expanding, it is evident that more research is needed to better determine how the persuasiveness of the systems affects users’ intended behavior.

According to Fogg [38], attempts to create persuasive systems often fail because many projects are too ambitious, being set up for failure. For example, a design team might select a challenging behavior as the target, for example, smoking cessation, but without having ever before created such a persuasive system, the success rate might be low. Zhang [39] stated: "A large number of health information system projects fail. Most of these failures are not due to flawed technology but rather due to the lack of systematic considerations of human and other nontechnology issues in the design and implementation processes." Thus, designing systems that aim at behavior change requires thorough understanding of the problem domain and the underpinning theories and strategies of persuasive systems design. Usually, an interdisciplinary team of professionals is also needed. The main decision points in developing interventions include defining the primary goal of the intervention, defining the target population, and selecting the messages for the intervention [40].

In the present study, the persuasive systems design model (PSD) [26] was applied as the framework for identifying various persuasive techniques that have been incorporated into the Web-based substance use interventions. The PSD model provides a recent and extensive conceptualization of technology-mediated persuasion.

We conducted electronic literature searches in five databases (Academic Search Premier [EBSCO Publishing, Herts, England], Cochrane Central Register of Controlled Trials, ISI Web of Knowledge, Medline [Ovid], and Scopus) to identify randomized controlled trials of Web-based interventions for substance use published January 1, 2004, through December 31, 2009, in English.

The following terms were used in the database searches: (1) online*, Internet*, web* and (2) intervention, self-help, treatment, trial. To identify alcohol interventions we used the following additional search terms: drink*, alcohol*. For smoking interventions we utilized the terms: smoke*, smoking, cigarette, tobacco, cessation. An asterisk (*) denotes a wildcard. We also screened the bibliographies of relevant articles, including systematic reviews [4,8,10,13,17,41,42] and meta-analyses [5-7,9] to identify potentially relevant studies. The study identification process is depicted in Figure 1.

Articles were included if they: (1) focused on Web-based alcohol or smoking (tobacco) interventions, (2) assessed behavioral outcomes or program utilization, (3) were randomized controlled trials or quasi-experimental designs, and (4) were peer-reviewed full research articles.

Follow-up studies that used data from the same cohort of participants were excluded. Both Danaher et al [43] and Severson et al [44] drew data from the same source, but they were included. Brief interventions (see [45, 46]) were excluded because of their rather limited content. Moreover, brief interventions have been commonly reported [47], even to the extent that it was deemed appropriate to put more emphasis on analyzing the less frequently reported, more complex Web-based substance use interventions. According to Moyer [45], a challenge in outlining the research literature on brief interventions is the varying definitions used in different studies. Following Babor’s [46] definition, we included the studies of interventions providing more than three sessions and/or aiming at more than 60 minutes of individual engagement with the program. Furthermore, we excluded articles explicitly using the term brief intervention in the title, abstract, or keywords. (See Multimedia Appendices 1 and 2 for excluded articles and reasons these were excluded.)

In total, 23 studies were included in the review and coded. The methodological quality of the included studies was evaluated by applying the CONSORT (Consolidated Standards of Reporting Trials) 2010 checklist [48]. There were some concerns with the quality of 4 studies [44,49-51], but they were still included. Overall, the information presented in the selected articles was thoroughly examined and evaluated. The first author coded all the included articles using a predefined form (devised by the authors) for evaluating persuasive systems. In addition, the abstracted data included various study characteristics (see Table 2). The resulting entries were checked and commented on by the second author. Any disparities were resolved through discussion.

Information technology always influences people’s attitudes and behavior in one way or another [52,53]. Oinas-Kukkonen and Harjumaa [26,52] have conceptualized a framework for designing and evaluating persuasive systems, known as the persuasive systems design (PSD) model. The PSD model builds on multiple theoretical constructs, such as goal-setting theory [54], elaboration likelihood model [55], and theory of reasoned action/planned behavior [56]. The PSD model is described in full detail elsewhere [26].

Although the PSD model is yet relatively unknown, we consider its use to be justified for this context. In our view, it is the most sophisticated persuasive design and evaluation method available. Many of the principles in the PSD model have been adopted and modified from the seminal work of Fogg [27]. We acknowledge the existence of similar endeavors, such as Ritterband’s behavior model for Internet interventions [22] and Abraham and Michie’s [24] taxonomy of behavior change techniques used in interventions. Despite the similarities and potential overlap, these approaches are quite different.

The PSD model presents a way to analyze, design, and evaluate the persuasion context and related techniques. Persuasion context analysis includes recognizing the intent, the event, and the strategy for persuasion. Dey [57] defines context as follows: “Context is any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and applications themselves.”

In the PSD model [26], the categories for persuasive system principles are primary task support (supporting the user’s primary task), dialogue support (supporting the interaction between the user and the system), system credibility (the more credible the system is, the more persuasive it is), and social support (the system motivates users by leveraging social influence).

The intent includes the persuader and the target behavior change type that the system is to cause in the user. The persuader is the initiator for the development of the system. The event contains the use, user, and technology contexts.

The use context refers to the problem domain dependent features. The use context, in particular problem domain dependent features, is relevant to the persuasion event. The use context can be determined by answering the following questions [58]: Who are the users as a group? What problem-domain dependent features are to be addressed by the design? Who (or what) else is competing for attention in this space?

The user context refers to the individual users characteristics. The user context includes, but is not limited to, users’ (patient, research participant, consumer [22]) characteristics, goals, abilities, and cultural factors. The user context can be clarified by addressing the following questions [58]: What is specific for the users with regard to what they are to be persuaded of? Why is persuasion needed? What constrains their decision?

There is an obvious need for depicting the technology context when describing a Web-based intervention. The technology context refers to the features and requirements of the technological platform and/or application.

The strategy in the PSD model emphasizes two elements, namely the message and the route. The message refers to the form and/or content selected to deliver the intended transformation (eg, behavior or attitude change). The content could be, for instance, statistical data about the health risks of drinking, but the information could be presented to the user in plain text, streaming video, or it could be embedded in a game. The route for persuasion can be direct, indirect, or both. A direct approach provides one or a few solid and convincing arguments, whereas an indirect route is based on a number of facts rather than a single strong argument (compare central and peripheral routes in ELM). Both routes may be in use simultaneously. A system might represent rational arguments while employing design patterns, which in themselves have been proven persuasive. As an example, an avatar with a specific voice type (angry vs soft spoken or male vs female) to present the message might make the delivery of the content more persuasive for a user; thus, users would be persuaded through the design choices made by the designer.

The persuasive design dimensions and principles are discussed and exemplified below.

The characteristics of the included studies are presented in Table 1. Of the 23 articles, 16 targeted smoking, 5, problem drinking, and 2, smokeless tobacco use.

The functionalities in the primary task category support the carrying out of the user’s primary task. Persuasion techniques in this category [26] include reduction, tunneling, tailoring, personalization, self-monitoring, simulation, and rehearsal (compare Fogg [27]). These were reported relatively well in the reviewed studies.

Dialogue support defines the key principles in keeping the user active and motivated in using the system and helping the user to reach the intended behavior. The principles in the dialogue support category are praise, rewards, reminders, suggestion, similarity, liking, and social role. Surprisingly, dialogue support was an underreported area in the reviewed papers.

Credibility is a persuasive element (eg, [78,83,84]). Harris et al [85] suggested that perhaps even seemingly superficial design elements of a website can influence responses to health-risk information. In their study, credibility cues affected the engagement with the site and influenced subsequent health behavior and cognition. According to Briñol and Petty [29], confidence in one’s thoughts is likely to be undermined if the received message is not credible.

The PSD model describes seven principles for supporting system credibility: trustworthiness, expertise, surface credibility, real world feel, authority, third party endorsements, and verifiability. The analysis of these is more a continuum than a dichotomy. For this reason, explicit numbers and percentages are not given here. However, based on the textual descriptions of the interventions, we suggest that most of them incorporated—to a notable degree—expertise (system provides information demonstrating knowledge, experience, and competence), verifiability (system provides means to verify the accuracy of site content via outside sources), authority (system quotes an authority, such as a statement or norms by an authoritative health institute), and trustworthiness (system provides truthful, fair, and unbiased information). Trustworthiness is crucial in website engagement as users will engage with sites they perceive trustworthy and navigate away from those they mistrust [78].

Naturally, evaluating perceived credibility is more or less subjective. People make initial assessments of the system credibility based on a firsthand inspection. This principle is called surface credibility. A persuasive system should provide information of (and means to communicate with) the organization and/or actual people behind its content and services. This feature is called real-world feel. To fully evaluate both surface credibility and real-world feel would require a hands-on approach on the actual implemented interventions.

A persuasive system could also provide endorsements from respected and renowned sources, for example, a recommendation by an authoritative health organization, an award for excellence in usability, or a privacy seal to ensure confidentiality. However, none of the articles reported utilizing third party endorsements.

According to Uchino [86], social support may refer to the aspects of the social network (groups or familial ties), specific behaviors (eg, emotional or informational support), or our perceived availability of support resources that may be shaped early in life. In the PSD model, the social support category describes how to design the system so that it motivates users by leveraging social influence. The model operates with the following persuasion techniques: social learning, social comparison, normative influence, social facilitation, cooperation, compensation, and recognition.

The social learning principle means that individuals may be more motivated to perform a target behavior if they can observe others performing the behavior while using the system. A closely related principle is social comparison: users will be more motivated to perform the target behavior if they can compare their performance with the performance of others. Users are also more likely to perform target behavior if they are able to observe others performing the behavior or are being observed by others. This principle is called social facilitation.

The most common means to providing social support were asynchronous peer discussion forums and synchronous chat rooms. A variety of other online social support features were also utilized. For a full description of studies regarding social learning, social comparison social facilitation, and normative influence, see Table 2.

A system can apply normative influence, in other words, positive peer pressure to enhance the likelihood that an individual will adopt a target behavior. Buller et al [63] and Matano et al [66] reported that users were working individually with the program to create a sense of privacy so they would unfold their smoking intentions.

In addition to the aforementioned ways for providing social support, a system can persuade users to adopt a target attitude or behavior by leveraging human beings’ natural drive to compete and cooperate. Competition was reported in 2 and cooperation in 1 of the analyzed papers. In An et al [59], the intervention site actively promoted a “Quit and Win” contest and included links to the online sign-up for this contest. An example of cooperation was found in Woodruff et al [76], as the students were working as a group in the virtual environment.

By offering public recognition to an individual or a group, a system can increase the likelihood that a person and/or group will adopt a target behavior. Only 1 of the articles presented functionality countable as recognition. In Woodruff et al [76] a message, “Guero's [a person’s name] been quit for a month!” was presented on billboards throughout the virtual world.

Of the 23 articles included in the review, 20 primarily measured health behavior outcomes. Of these articles, 12 reported statistically significant differences between groups. Overall, 3 articles assessed program utilization, and 2 of them reported positive findings (see Table 1).

The primary task support components were reported relatively widely in the reviewed studies. Reduction, self-monitoring, simulation, and personalization seem to be the most used ways to support accomplishing a user’s primary task. This is an encouraging finding because reduction and self-monitoring can be considered as the key elements of primary task support. The utilization of tailoring was surprisingly low. The lack of tailoring may imply that the interventions are targeted for too broad an audience. It would be reasonable to assume that different approaches were needed for different kinds of user groups. Elements of dialogue support were mostly underutilized in the interventions. Leveraging reminders [82] was the most common way to enhance the user-system dialogue.

Credibility issues are crucial in website engagement as users will engage with sites they perceive credible and navigate away from those they do not find credible. Based on the textual descriptions of the interventions, we cautiously suggest that most of them were relatively credible. Nevertheless, it seems to be, as Danaher and Seeley [19] eloquently put it, that “credibility is in the eye of the beholder.”

The prevalence of social support in the reviewed interventions was encouraging (see Table 2). Social support provided by the systems is often based on peer support. According to Eysenbach [36], there is not yet very strong evidence for what type of peer-based social support the systems really should provide. In their systematic review, Shahab and McEwen [17] argued that while chat forums could potentially aid in stopping smoking through the providing extra social support, it still seems to be that entirely automated interventions are even more effective. Bennett and Glasgow [14] claimed that there are no examples of trial designs that would enable a systematic investigation of the potential benefits of social networking. In this review, we were not able to confirm the effect of social support components on health behavior outcomes. This is mainly due to the fact that only 1 of the included articles [72] explicitly studied the effect of adding a social component (virtual community) within the intervention. Furthermore, many research and design issues still need to be resolved about the type of online support, for example, whether it should be expert-led versus user-driven, moderated versus unmoderated, synchronous versus asynchronous, or open access versus restricted access, among other issues.

Evaluating the effectiveness of specific persuasive features within Web-based interventions is difficult since the features are not usually explicitly tested. According to Kypri and Lee [87], the descriptions and analyses of how interventions are developed are often absent from scientific literature due to space constraints. However, we think that the presentation of detailed information about the theoretical basis, functionality, content, and structure of a Web-based intervention helps to interpret the results and conduct evaluations as on a more finely grained level. Ahern [88] points out that randomization to experimental groups or conditions remains the gold standard for evaluating intervention efficacy but may not provide the most relevant information for dissemination. Furthermore, methodological challenges and latent scientific foundations in researching Internet-based interventions are acknowledged by many researchers (eg, [19]). Glasgow [89] stated that the diversity in content area, disciplines involved, and publication outlets is the reason that the consistency in how Internet-based solutions have been conceptualized, reported, and evaluated has been low. Finally, the brevity of intervention descriptions makes it more difficult to draw generalizable results (see [8,13]).

Due to these issues, the application of the PSD model turned out to be relatively laborious. In the present review, we relied on textual descriptions of the interventions, thus being able to provide only a limited synthesis. Regardless of its wide coverage, the PSD model is not an exhaustive list of persuasive features, and also some of the features are overlapping (eg, social learning/comparison/facilitation and liking/similarity) and thus rather difficult to analyze. New persuasion techniques may also be identified in the future. The PSD model has been built in such a manner that it may evolve, but even as it stands now, it is an important asset for any health behavior change system developer.

Analyzing persuasive design is a challenging task. When conducting an analysis such as described in the present review, potential bias lies in the interpretation of the articles. Nevertheless, in extracting and categorizing persuasive features, we rigorously observed if the authors clearly stated the described variables. Obviously, the articles did not necessarily follow the very same terminology as found in the PSD Model. Thus, the analysis was based on interpretive categorization.

This systematic review focused on randomized controlled trials, thus excluding potentially meritorious studies. (Quasi-experimental studies were not found in the search process.) A meta-analysis was not conducted due to the heterogeneity of the studies. Overall, there are already several reviews on Web-based (or similar) interventions. To our knowledge, the present review is the first systematic review to address persuasive system features in Web-based interventions for substance use.

In this review we examined the persuasive system features of the included Web-based interventions. We think that this type of novel approach is useful for current and future research for recognizing what kind of tactics in present systems have been utilized to motivate people in achieving better health.

However, at this point, linking specific persuasive features to outcomes is difficult, relying only on brief textual descriptions of the interventions. Also, it is not possible to determine the (perceived) credibility of a Web-based intervention based on reading an article. We acknowledge that many studies have examined, for example, the role of tailoring in health behavior change interventions [10,11], but the persuasiveness of a particular component/application/system is a more complex issue and has yet to be tackled in future endeavors.

We are not implying that the mere presence of persuasive features is enough. The development of Web-based and other similar interventions is a highly elaborate and a multifaceted issue. Still, it is relevant to consider the technological aspects since the Web and related technologies are being used as a delivery channel. Atienza and colleagues [90] pertinently remarked that “health information technology does not occur in a vacuum, but rather technologies exist within social systems.” In order for widespread adoption, dissemination, and extended use of technology-enabled health behavior change interventions to take place, it is necessary to investigate not only how the interventions affect individuals, but also how individuals interact with technology and each other [90]. Further research is also warranted to increase our understanding of how and under what circumstances specific persuasive features (either in isolation or collectively) lead to positive health outcomes in Web-based health behavior change interventions across diverse contexts and populations.