The frequency of messages should address the need for the program to communicate key information without overwhelming or burdening the user. Often, text messaging programs send out at least one message/day during key behavior change periods and fewer messages (eg, 3 messages/week) in less acute phases. Depending on the health behavior being targeted, fewer or more text messages per day may be appropriate. For example, Text2Quit sends 5 messages on the quit date, daily messages in the first week after the quit date, and 3 messages per week in the weeks after that [16]. For users who are frequent texters, message frequency may need to be higher so that messages stand out from the many texts they already send and receive daily. It should be noted that some programs only send out texts when a user requests information. For example, SexInfo, a sexual health information service, is a reactive service that replies when the user initiates a question/query to the system [31].

The science on message frequency and quantity has not yet identified the ideal “dose.” Poorman et al [32] suggested from a systematic review of the SMS health behavior literature that participants may be better retained if message quantity is varied over time, while Head et al [3] reported that programs with decreasing message frequency were more effective than those with constant message frequencies. A common criticism of these types of programs is that the program sends out “too many messages” [15]. Another related consideration, especially among programs targeting addictive behaviors like eating or smoking, is that the messages may act as a trigger for the behavior that is to be avoided. Abroms et al [15] found that while overall the program was effective, some participants reported that the texts were a trigger for smoking.

The timing of messages may be related to both the content of the messages (eg, what the messages ask the user to do), the daily routine of the user (eg, when the user is free to consider the text message), and the nature of the behavior change (eg, what time of day is appropriate for targeting that behavior). Consideration also needs to be given to what will trigger the messages (ie, date of enrollment or behavior change, in a weekly cycle). In Papua New Guinea, a text messaging program aimed at providers treating malaria patients was found to be most acceptable in the mornings and during work hours, to help facilitate the usefulness of the reminders [33]. While 2 messages per day were found to be acceptable, providers did not like their repetitiveness over time [33]. A study of binge drinking behaviors concentrated messages in the days leading up to and during the weekends, when binge drinking was more likely to occur [29]. Another study found that adolescent girls preferred the message timing to vary, as it was seen to be less “robotic” if the message timing was unpredictable [34].

Interaction, or bidirectional messaging, is helpful to promote engagement in a program. Shneiderman lists “Eight Golden Rules of Interface design,” of which the following may be especially relevant for SMS programs: provide consistency, offer short-cuts, offer error prevention (ie, “Did you mean this?” or “We didn’t understand you.”) and provide responsive feedback for user-initiated actions [35]. Opportunities for interaction in an SMS program can occur around surveys (eg, “Are you ready to quit? Reply 1 if you are ready or 2 if you are not ready”), with tracking (eg, “How many cigarettes did you smoke yesterday? Reply and see if you met your goal.”) and with keywords. Keywords are words that the user can send into the system at any time for additional help (eg, a user texts the keyword “CRAVE” if having a craving and receives an immediate reply with additional help). Keywords should be limited in number, so that users can easily remember them and use them as needed. By responding to a program survey or using keywords, users can expect to receive positive feedback or an actionable reply from the program. One way to promote interaction is to “gamify” the interaction, providing the user with points that the system can track over time for key types of interaction with the system. Abroms et al used a trivia game as a means of distracting the user during smoking cravings [15]. In Bauer et al’s weight management program [36], responses to check-in questions about eating and physical activity elicited an appropriate automatic response relating to encouragement, motivation, positive reinforcement, or reminders. The algorithms for an automatic text messaging program need to consider human error and include a variety of responses. For example, it might be helpful for the program to recognize common misspellings or typos, so that it will still respond appropriately rather than ignore messages. Regardless of program type, it is important to have an automated message for potential emergency texts and to have protocols in place to manage anticipated or even unanticipated events that may need to be handled by a live person. One solution is to have a real person/provider monitor messages for responses such as, “I have a headache, please help” that the program might not otherwise recognize. Another solution would be to have an automated response that says that the program does not recognize the response and to contact emergency services in the case of a medical emergency. For example, if a person replies to a computer-generated survey with a high-risk response (eg, with a high blood sugar reading or severe weight loss), the response triggers an alert to a clinician who then contacts the patient. In another example, following the release of inpatients from an eating disorders program, one study [37] had staff monitor incoming communication so that if patient responded that they were bingeing, a provider would follow-up with a personalized response rather than allow an automated reply to go through.

The source of text messages is generally the program name (eg, SmokefreeTXT or CDC), so it is easily recognized by the participant as coming from the program. However, automated messages may be supplemented with messages from a real person/counselor/clinician. Even within automated program messages, the message source can vary with some messages coming from the program and others from a specific “person” who is part of the program. For example, in Text2Quit, some messages come from a fictitious quit pal who offers social support (eg, Erika/Text2Quit). In other programs, users are paired with an actual quit buddy to interact via text and who supports their quit attempt [38].

A decision has to be made as to whether the program will run as a single generic program, with all users receiving the same content, whether there will be different versions (or protocols) for different types of users, or even whether individually tailored versions of the program may be offered. In general, creating extra protocols or tailoring to individual characteristics can be more expensive and therefore should be carefully thought through ahead of time. Tailored protocols may be considered because they can result in higher readership, higher message recall, perceptions of higher personal relevance, and in some cases, greater behavior change [39]. Some evidence exists to support the use of tailored text messaging programs [3,40]. Kharbanda et al [40] found it was helpful to add a child’s name to a text message about immunizations and follow-up vaccines, because it was more salient for parents. In another study, adolescent participants expressed that it would be more appealing if they were asked how they were feeling and the responses were tailored directly to them [34]. Logic can be built so that participants receive different message content based on their stage of change, readiness, and timing around key behaviors of interest (ie, quitting or birth of baby). This logic can be assigned from the outset, or can be modified based on responses to messages received by the program.

Models of mobile programs include programs that stand alone (eg, SmokefreeTXT) [41] or programs that supplement other existing programs such as face-to-face group counseling [42], phone counseling at quitlines, or those that supplement interactive websites. In cases in which texting supplements a larger program, mobile communication can be thought of as additional touch points to reinforce messaging from counseling sessions or Internet programs. To date, most text messaging programs for behavior change that have been evaluated have been stand-alone programs [16,17], and evidence to support their use as a supplemental platform (in addition to other in-person or mobile/Web-based components) is mixed [43,44].

Text messaging is not a secure technology and, therefore, has risks associated with it for transmitting personal health information. This is especially a sensitive issue when programs will include personal health information and originate from a Health Insurance Portability and Accountability Act (HIPAA) covered entity, within the US context [7], or meet other health care-related privacy standards internationally. One way to make risks more transparent to participants is to disclose them in clear language at the time of signing up. Privacy risks are detailed in the terms and conditions of program use and explicitly state that text messaging is not secure, and that by signing up, the subscriber agrees to the risks. One example is how the NCI describes their terms of service online for the use of SmokefreeTXT [45]. A similar disclosure will be included in the roll-out of the Veteran Administration’s Annie text messaging system, which is currently being pilot tested and will perform a variety of functions including the collection of blood pressure readings [9].

Messages can be based on many behavioral approaches including providing information or advice, asking users to track behaviors, providing feedback on goals, offering reminders, providing positive reinforcement, or providing social support (see Table 1 for examples). Chib et al [46] used texting to ask quiz questions about HIV transmission and testing. We recommend limiting each message to a single topic or actionable item.

Assuming the source is a credible health entity, the language should fit the professional character of the organization and not be inclined to use slang, too many abbreviations (eg, “how r u doin?”), or informal punctuation (eg, “well done!!!!”). Our experience is that users find this type of language to be unprofessional coming from a credible health source. According to Ranney et al [34], even teens expressed that they did not like slang used in health messages.

Once the message library is drafted, check the literacy demands associated with the messages. This can be done using the Flesch-Kincaid Grade Level Test or Flesch Reading Ease test tool to determine the readability of the messages. In general, shorter words and sentences have lower literacy demands. Generally, for an adult audience, an 8th-grade reading level is considered the maximum recommended reading level [13], and in some cases, a lower reading level may be necessary. Further, with populations that may have low English literacy, but higher literacy in another language, a text messaging program can offer a translated version. For example, the program Text4Health provided parents with the option of switching the program language as follows: “To receive messages in Spanish, text ESPANOL” [40].

Given that mobile phone penetration is at 64% among adults in the United States [25], and there are up to an estimated 1.75 billion users globally [47], users are likely to be reading text messages on mobile phones. This means that text messages can be seamlessly linked to Web content to expand their content in the form of mobile Web pages, videos, audio, games, and social media. Increasingly, text messaging programs are being built in conjunction with mobile phone apps. The large-scale texting program, text4baby, now has an app that can be downloaded to support the text platform, as well as mobile Web pages offering additional content to supplement approximately 50% of the text messages [48]. However, because this adds additional development costs, it is important to study whether the addition of apps confers benefits to the end user. Additionally, social media provides another low-cost extension for an SMS intervention. In a weight-loss intervention for college students, the enhanced group received Facebook content reinforced with SMS content [44].

The message library can be written to include periodic check-ins about the program’s success. These may take the form of surveying users about behavior change of interest (eg, “Text2Quit: Have you smoked a cigarette over the past 7 days? Reply 1 if yes or 2 if no”). These offer the opportunity to redirect program messaging for the user (eg, If reply is 1: “Sorry to hear you slipped; Here’s what you can do to get back on track.”), as well serve the purpose of providing valuable data on participant engagement and a program’s success [15]. Reminder text messages can be effective in prompting users to reply, in order to increase response rates for key program evaluation metrics. One consideration related to response rates is that there is likely to be more missing data with the increasing frequency of check-ins. Thus, the importance of higher response rates will need to be balanced against providing multiple opportunities to check-in with users and obtain updates on their status.