Since the 1990s, mental health providers have explored how to support clients via internet-based communication [1]. Prior work suggests that young people may be particularly interested in these approaches, as digital communication feels more private and emotionally safe [2]. However, internet-based communication, particularly written communication, may have significant barriers for providers and clients, including the inability to express emotion and challenges in communicating clearly [2]. Currently, there is limited evidence on how to overcome these communication issues in counseling settings [3]. Understanding how to do so may help telehealth providers build stronger therapeutic relationships, thus improving the help-seeking process. Further, this understanding may help agencies improve services and training for providers.

Technology-based (ie, telehealth) mental health services may not be as effective as in-person services. One recent meta-analysis suggests that videoconferencing-based therapeutic relationships may be inferior to those developed during in-person therapy [4]. However, there may be times when in-person care is not accessible or feasible. Nearly half of people in the United States live in a mental health shortage area, and there are areas with less than 2 psychiatrists per 100,000 residents [5,6]. As a result, increasing access to mental health care may depend on telehealth approaches. Within telehealth studies, interventions retaining elements of human contact are more effective than entirely computer-based interventions [7,8].

Two critical aspects of the helping relationship, therapeutic relationship and social presence, may be challenged when engaging digitally. A therapeutic relationship based on mutual trust, respect, empathy, and positive regard is essential in counseling [9]. Hundreds of studies have confirmed the importance of this collaborative relationship [10]. For most help-seekers, confidence in the provider, including perceptions of empathy and expertise, is key to developing a strong relationship [11-13].

Social presence [14], the sense of connecting and being with another, is another element that may be compromised during digital communication. Social presence may also be defined as the degree to which the other person feels “real” [15]. Although it is a natural element of face-to-face counseling, telehealth providers may have to be intentional in building a social presence. When conversing with unknown entities through written technology, it is common to question whether the other person is a human or a bot [16], in part because people are not reliably able to differentiate between the two [17]. Some prior work suggests that social presence is an important aspect of digital helping relationships because it assists in building therapeutic partnerships, professional bonds, and open communication [18,19].

Much of the literature on telehealth counseling focuses on verbal communication via videoconferencing or phone [20]. Few studies examine written mental health counseling services, and there is reason to believe that spoken and written communication are substantially different. In a recent review of the digital therapeutic relationship, Bantjes and Slabbert [20] suggest practical strategies for establishing rapport in digital spaces, such as maintaining eye contact, having high-speed internet to avoid lags, and attending to lighting and microphone placement. These strategies improve the audio and visual cues, which are not applicable to written communication.

The literature on written counseling is limited [21,22]. In the 1990s, a small group of mental health providers offered therapy via email [23]. This early work identified several possible strengths and limitations. It was helpful for clients to write about their feelings, and the anonymity of email allowed them to share more openly. This asynchronous approach also increased many individuals’ sense of control, as they could choose when and where to engage with the therapist. Conversely, building a relationship and understanding nuances could be difficult without the usual social cues [23]. Two more recent literature reviews support many of the impressions formed by the early adopters, although most of the studies had very small sample sizes [22,24].

Written interventions may be challenging for the provider and patient, and both experience similar challenges. One randomized controlled trial of a chat-based cognitive behavioral therapy demonstrated reduced depression symptoms after 10 sessions [25]. In a parallel qualitative study, participants reported mixed perceptions of the experience [26]. Some reported feeling more able to share openly and process because of the anonymous platform. Others felt it was challenging to develop a relationship and express complex feelings and thoughts via writing [26]. Another study assessed differences between email-based cognitive behavioral therapy and unguided treatment. The email and unguided programs had better outcomes than the wait-list control group for some, but not all, outcomes [27,28]. Other studies, with and without in-person or telephone comparison groups, showed similarly mixed results [29,30]. In one unpublished dissertation, counselors who provided email services reported feeling substantial anxiety due to uncertainty, limited sensory information, and concerns about misunderstanding clients’ intentions [31]. The lack of visual, verbal, and social cues was particularly challenging [31]. They often focused more on the tasks and transactional aspects of helping to manage these uncertain dynamics [31]. Many also talked about needing much more time than usual to build the therapeutic relationship, although it did eventually happen for most [31].

Beyond mental health counseling, some recent work has examined written communication for brief counseling and advocacy [3,32,33]. Overall, privacy, autonomy, control, anonymity, and accessibility are seen as benefits of written services [34-36]. Building social presence and connection is an important aspect of the experience [34]. Often this professional connection builds over time, but because the help-seeker and crisis counselor or advocate do not have an ongoing relationship, it may be particularly difficult to communicate adequately and build a relationship [2,30,37,38]. Correctly understanding sarcasm, humor, and other nuanced language is particularly challenging in these brief interventions [3,39]. Like mental health counseling, the impact of written crisis and advocacy services is unclear in the current literature and may depend on geographical location, counselor training, and the help-seekers’ situations [33,40-46].

Overall, establishing a human connection based on a strong therapeutic relationship and social presence will likely result in more effective, acceptable interventions. Providing crisis services is complex, and the confines of written communication create additional challenges. Rapport-building is particularly difficult, and mistakes may cause the help-seeker to feel worse [47]. However, there are not yet best practices for building rapport in these conversations, as existing approaches to rapport-building often depend on verbal and nonverbal cues [48]. As part of a larger study focused on building best practices for written hotlines, we worked with a child maltreatment-focused text or chat hotline. This analysis aims to categorize crisis counselors’ efforts to build rapport and convey active listening in written conversations. Using these categories, we identified characteristics associated with successful conversations, as reported by help-seekers. This work provides an important foundation for how to build therapeutic relationships in written mental health and hotline services.

The data for this study are from the PACTECH (Prevent Abuse of Children Text and Chat Hotline), the text and chat arm of the Childhelp National Child Abuse Hotline [49]. Since 1982, Childhelp has offered 24/7 phone-based hotline services focused on support and resources related to child maltreatment. In 2018, the hotline expanded to include text and chat capabilities. Crisis counselors are employees rather than volunteers. Most are master-level professionals with specialized training in hotline services and child maltreatment. After conducting a quantitative pilot evaluation for 2 years, hotline leadership partnered with the lead author to use qualitative and mixed method approaches to identify best practices for services. As part of the data sharing agreement, the lead author and her research team received access to deidentified transcripts and metadata from conversations that were purposefully selected to represent a wide range of experiences and perceived outcomes.

The Purdue University Institutional Review Board approved the research protocol (IRB-2020-965). The service terms and conditions disclosed that data may be shared with researchers. As a secondary data analysis of deidentified data, additional consent from participants was not required by the Institutional Review Board. The contract teams from Purdue University and Childhelp negotiated the terms of the data sharing agreement, including data security and access. As a result of the data sharing agreement, the data may not be released publicly.

The sample consists of 314 purposely selected conversations out of the 1153 text and chat conversations during July 2020. In addition to maintaining the written transcript of the conversation for 60 days, Childhelp collects preconversation and postconversation surveys from the help-seekers. The preconversation surveys focus on help-seeker characteristics (ie, age, gender, state of residence, and referral source), while the postconversation survey focuses on their perceptions of the conversation (eg, do they feel more hopeful, less stressed, and more prepared). We used maximum variation sampling to capture diverse help-seekers and outcomes, although not necessarily in the proportions present in the overall data [50]. This approach is particularly useful when looking for diverse perspectives, as was the case for our study. We sampled based on the preconversation and postconversation surveys. In our sample, 297 (94.6%) help-seekers answered at least 1 presurvey question, and 263 (83.8%) answered at least 1 postconversation survey question. First, we selected conversations where help-seekers reported that they were satisfied, unsatisfied, or mixed. We also included some conversations without surveys to reduce survey response bias. Then, we reviewed the demographic characteristics of the selected conversations to ensure help-seekers of different ages, races or ethnicities, and genders were included in the sample. For example, most help-seekers are girls, so there were relatively few conversations with boys in our initial sample. We added additional conversations with boys to ensure the results were not only relevant to girls.

We analyzed and reported the findings from all 314 conversations. When reporting quotes, however, we were particularly interested in the 45 conversations where help-seekers reported in the postconversation survey that the hotline was a good way to seek help and that they were a lot more hopeful, a lot more informed, a lot more prepared to address the situation, and experiencing less stress. Except when specifically referencing less successful conversations, all example quotes come from these conversations, as they represent those most successful from the help-seekers’ perspectives. All quotes are reported verbatim from the conversations, including any errors.

We used qualitative content analysis to process the conversations. We used both inductive and deductive processes to develop the codes. The first draft of the coding frame was based on our work with child maltreatment–related conversations within the Crisis Text Line [51-53]. Then, we revised the framework based on the content of the conversations.

Our development process followed the adaptation of grounded theory described by Schreier [54]. The lead author and her graduate research assistant reviewed all the conversations. During a second review of the conversations, we took notes on commonalities within the conversations, emphasizing material not captured in the first draft of the codebook. As we refined the codebook, all team members met weekly to discuss emerging materials and define and develop codes. After completing the framework and definitions, we coded 30 conversations and met to compare the code applications. We discussed differences in coding and refined the framework with the entire team. Then, we coded 30 additional conversations and assessed the coder agreement. After the second round of pilot coding, we reached 95% agreement on the codes and moved to code the full data set. In sum, we had 127 codes in the codebook, which were applied 22,326 times. After coding all the conversations, we reviewed the materials within each code. This process followed the segmentation process described by Schreier [54], where coded materials are decontextualized and reviewed to identify commonalities and themes. Through this process, we also assessed whether we met saturation, which occurs when all categories have been identified in the data set. Schrier’s [54] definition of qualitative content analysis saturation is different from other forms of qualitative methods. In other forms of qualitative analysis, saturation refers to the point at which reviewing additional material does not provide new information. We informally assessed this type of saturation by examining whether all codes were used if we considered only half of the sample. We found that all codes were used when we reviewed 2 different randomly selected split samples, which suggests that few new insights would be gained if we added additional conversations to our sample. After conducting these checks, we categorized the conversations by the outcomes and focused on similarities and differences across the groups.

For this analysis, we focused on the codes related to rapport building and active listening conversations. There were two main types of approaches used by crisis counselors: (1) counseling approaches and (2) evaluation-based language. Active listening skills, otherwise known as attending skills, are how counselors build connections with clients, express empathy, and convey that they are listening [48]. These skills may be defined slightly differently; asking questions, paraphrasing, reflecting feelings, and interpreting or summarizing the situation are generally recognized skills. We added validation [55] and unconditional positive regard [56], which are also commonly incorporated into helping relationships. Evaluation-based language, such as praise and apologies, is commonly used by adults when talking with children [57]. These statements differ from other approaches because the counselor’s evaluation of the situation is included.

We also examined how these approaches differed between the help-seekers most satisfied with the conversation (ie, answered all after-conversation survey questions as “Yes”) and those who were least satisfied with the conversation. We intended to define the least satisfied as those who answered all the after-conversation survey questions as “No.” However, only 4 people fit that criterion, so we included all help-seekers who answered most of the questions negatively.

The research team included the lead author, a graduate research assistant, and 2 collaborators at Childhelp. The lead author is a family violence prevention researcher with a PhD in public health and an MA in counseling. She has experience conducting qualitative analyses of written hotline transcripts. The graduate research assistant was a master of public health student and had worked on the lead author’s research team for 3 years. She had experience with qualitative child maltreatment research. The Childhelp collaborators have substantial experience in hotline counseling and leadership. One has an MS in counseling psychology. The second has an MS in family and human development and an MEd in guidance counseling.

Overall, our sample of help-seekers was generally similar to Childhelp’s overall text and chat users (Table 1) [58,59]. Help-seekers tended to be female, young, and seeking help for themselves. Overall, they were generally at least a little more hopeful, informed, and prepared to deal with the situation after the conversation (Table 2).

There were some differences in active listening skills, other counseling skills, and evaluation-based language between successful and less successful conversations. Although the sample of successful and less successful conversations was too small for formal statistical analysis, some commonalities emerged. First, although conversations were approximately the same length, less successful conversations tended to have more statements that attended to rapport building. Second, there were also differences in how the counselors applied these approaches. Unlike the preceding sections, this section includes quotes from both successful and less successful conversations.

Overall, counselors in less successful conversations tended to be vague or to directly repeat what was said by the help-seeker. These differences were particularly apparent when counselors were paraphrasing, asking open questions, or apologizing. For example, paraphrasing in less successful conversations tended to be either very vague (eg, “It sounds like you are being hurt already”) or very specific (eg, “I am hearing you have some future plans to get a job and earn your own money...”). In the last example, the help-seeker used the same phrasing in their previous statement. Conversely, successful conversations tended to be specific without direct repetition (eg, “Sounds like they are something to help you cope”). Similarly, less successful conversations tended to include open questions that were either broad (eg, “What’s happening?”) or focused on clarifying how the crisis counselor could help (eg, “How are you hoping that I can help?”). Some successful conversations also included questions clarifying how the crisis counselor could help, but it was more common to ask more specific questions, like “What is it that you would like to vent about?” and “What are their thoughts on CPS involvement?” Finally, crisis counselors used generally vague apologies about the situation in less successful conversations. Saying things like “I am so sorry this happened to you” or “I’m sorry to hear that” was common. Although some successful conversations also included these types of apologies, it was more common to pair the apology with a specific reason, such as “I’m so sorry that you have been experiencing this for so long” or “I am sorry to hear Mama is sick.”

Overall, our study suggests that it is possible to build therapeutic relationships via a text and chat hotline with individuals seeking child maltreatment–related information and support. Approximately 15% (n=45) of our sample reported that the hotline was a good way to seek help and that they were a lot more hopeful, a lot more informed, a lot more prepared to address the situation, and experiencing less stress. However, our sample was intentionally selected to represent a wide range of help-seeker perceptions, so this does not indicate that 15% of the hotline’s help-seekers felt this way. Based on the 2022 Childhelp data report, about 85% of help-seekers reported getting the information they needed, 80% of help-seekers reported feeling more hopeful after the conversation, and 75% reported feeling better prepared to deal with the situation [59]. These percentages suggest that the hotline provides a well-received service.

Generally, counselors built rapport through active listening skills, other counseling techniques, and evaluation-based language (ie, apologies, praise). Through active listening skills and other counseling techniques, counselors often expressed that they were listening, wanted to understand the help-seekers, and cared for them. They expressed their approval or disapproval of the help-seekers and aspects of their experiences through evaluation-based language. Although there is nothing inherently wrong with using apologies and praise, they tend to be avoided in many therapeutic approaches. Praise may undermine intrinsic motivation (ie, internal drive) and reduce engagement in the process [63-65]. Further, these types of evaluation-based language are rooted in control, as they are given based on something that another individual (ie, the crisis counselor) deems desirable [63]. As a result, the help-seeker might seek praise by giving answers that they believe the crisis counselor wants to receive instead of accurate answers, which may reduce the benefit of the conversation. However, praise and compliments may be a quick way to build encouraging feelings [66]. As it is challenging to build relationships via writing, praise may be one way to build a relationship quickly. Additional research into the impact of evaluation-based language is necessary to understand its role in written crisis counseling.

There were some differences between successful and less successful conversations. Surprisingly, less successful conversations tended to include more attending language than successful conversations. However, there were differences in the ways that crisis counselors apply these techniques. Overall, the crisis counselors in successful conversations tended to be more specific and tailor their responses to the help-seekers. Possibly, counselors who gave tailored responses built rapport more quickly; thus, fewer attending statements were required. If this is the case, they could move to problem-solving more quickly, which may also contribute to help-seekers’ perceptions that they were more prepared to address the situation and were more informed. These tailored responses may increase help-seekers’ perceptions that the crisis counselor is invested in the conversation. Several help-seekers explicitly asked if they were speaking with a bot in this sample. Having tailored responses may increase crisis counselors’ social presence and reduce help-seekers’ concerns about whether they are “real.” As organizations consider using large language models and chatbots in these types of services, careful attention should be given to help-seekers’ perceptions about the service and its appropriateness for the audience. As the National Eating Disorders Association learned when its wellness chatbot began providing diet information, large language models trained on outside data may not be a good fit for conversations with help-seekers [67].

Our work has several limitations, including some inherent to secondary data analysis. First, we could not speak with the help-seekers or the counselors about the conversations. Although we were able to identify similarities across well-received conversations, it is possible that other aspects of the conversations contributed to help-seekers’ perceptions. Second, we do not know how these conversations shaped long-term outcomes. Moreover, it is difficult to follow up with help-seekers, as evidenced by the 6% response rate to a 2-week follow-up survey conducted by the National Domestic Violence Hotline [68]. Further, many of the help-seekers in this sample indicated that it is unsafe to speak aloud about their experiences, so qualitative data collection with this sample would likely have an even lower response rate. It would be more feasible to speak with counselors about their experiences, but their perspectives may be disconnected from those of the help-seekers. Despite this limitation, we incorporated the help-seekers’ perspectives through the postconversation survey, which is more than is usually possible in secondary data analysis.

Our work may not generalize to conversations unrelated to child maltreatment. As a child maltreatment–specific hotline, all conversations included elements of child maltreatment. Conversations about other topics may require other approaches. However, our results are consistent with prior work on building rapport in other forms of counseling [48,55,56], so it is reasonable to expect these findings would translate to written conversations about other topics.

To the best of our knowledge, there is no other work examining specific ways to build a therapeutic relationship in written mental health counseling or crisis counseling. However, the ways that crisis counselors attended to the dynamics of the conversations were generally like those found in in-person counseling [48,55,56].

Telehealth approaches to counseling may be particularly important for young people experiencing maltreatment. Other formal resources, such as law enforcement, schools, and child protection systems, often fail to respond adequately [53,69,70]. Further, internet-based approaches, particularly written approaches, are highly acceptable to young people experiencing maltreatment [69]. In our sample and past research, children shared that they could not call resources because an audible conversation would cause parents to know they were seeking help. In work conducted with Crisis Text Line, it was common for young people sharing child maltreatment to report that the abuse escalated when parents discovered their attempts to seek help. Written, anonymous communication that is available 24/7 may be a safer way for these young people to seek help. Thus, written communication may be particularly important for children in unsafe homes.

There is also limited evidence on how to respond when young people share maltreatment experiences. Regardless of the ability to impact or end the maltreatment, individuals who receive a child maltreatment disclosure need to receive an appropriate, supportive response [71-74]. Supportive responses encourage the young person experiencing maltreatment to reframe their experience, which substantially reduces the likelihood of poor outcomes otherwise associated with maltreatment [75]. Conversely, unsupportive experiences often have long-lasting consequences [74,76]. Receiving a hurtful or unsupportive response increases the likelihood that the young person will experience more significant physical and mental health issues [74,76,77]. Unfortunately, many young people receive unsupportive responses to their disclosures [53,70]. Often, they report that others, particularly adults, do not believe them and are unwilling to help [70,78]. These experiences reduce their willingness to seek help or share their experiences in the future [70]. Our work suggests that responding to these disclosures adequately in written conversations is possible.

Our work also contributes to a small body of literature on using text and chat hotlines to provide services to people experiencing violence more generally. Michigan State University added chat services to its existing sexual assault support and advocacy hotline. Their evaluation was consistent with many of the benefits and limitations of other forms of written counseling, including challenges with nuance, misunderstanding written language, and communicating empathy [3]. However, the format also gave help-seekers a greater sense of control [3]. Another study focused on agencies providing digital violence-related support and advocacy services [32]. This work also emphasized the importance of clear communication and building rapport, although help-seeker perceptions of these factors were not assessed [32].

Building therapeutic relationships and social presence are important components of digital interventions involving mental health professionals. Prior research suggests that they can be challenging to develop in written conversations. Our work demonstrates characteristics of conversations associated with greater satisfaction among help-seekers. These findings may be adopted by other organizations building mental health or support interventions that include written communication. However, additional research is needed to identify how to train providers to adopt these strategies while also tailoring their approach to the help-seeker. Further, our findings may inform future work with large language models, including how large language models could contribute to these interventions. However, future research is needed to understand how help-seekers would interface with these methods and to ensure that the models consistently convey appropriate, supportive information.