This qualitative study investigated participant experiences of AVATAR therapy based on semistructured interviews. Adopting a peer research approach, this study was nested within the AVATAR2 trial (ISRCTN55682735, registered on January 22, 2020), a multicenter RCT conducted across 4 trial centers in England and Scotland (Institute of Psychiatry, Psychology, and Neuroscience, King’s College London; University College London [UCL]; The University of Manchester; and University of Glasgow). The additional research ethics approval for this study was granted by the London-Camberwell St Giles Research Ethics Committee in December 2022 (reference 20/LO/0657).

In line with open science processes, a preregistration form (based on Haven et al [28] and Staniszewska et al [29]) detailing the study protocol was published on the Open Science Framework to ensure transparency and reliability [30].

A total of 16 PPI representatives (termed peer researchers in this study) were involved across all stages of this study, with representation from all 4 trial sites. Most individuals had lived experience of hearing distressing voices, with some having received AVATAR therapy as part of the AVATAR1 or AVATAR2 trial, as well as a small number of people who had cared for a loved one who had experienced this. This included (1) coproduction of participant-facing documentation, (2) collaborative development of the topic guide, (3) codelivery of interviews, and (4) contributions to data analysis. This peer research approach aligns with the National Institute for Health Research INVOLVE guidelines [31], which highlight 6 key values—respect, support, transparency, responsiveness, fairness of opportunity, and accountability. For example, all peer researchers were paired with a site research worker who provided flexible and personalized support in order to develop a collaborative working relationship [32]. Peer researchers were remunerated according to INVOLVE guidelines [31]. More details are provided in Multimedia Appendix 1, and PPI across the trial is further discussed in Owrid et al [32].

We recruited participants who had received AVATAR therapy as part of the AVATAR2 trial [33]. This included those randomized to the AV-BRF and AV-EXT arms. All participants met eligibility criteria for the trial. Inclusion criteria were (1) aged ≥18 years, (2) under the care of a mental health team, (3) hearing a distressing voice in English for at least 6 months, (4) adequate English to take part, and (5) schizophrenia spectrum disorder or an affective disorder with psychotic symptoms. Exclusion criteria were (1) lacking capacity to consent, (2) currently undertaking individual psychological therapy for voices, and (3) currently experiencing an acute mental health crisis. In addition, specific inclusion criteria for this study were as follows: (1) received AVATAR therapy in the AVATAR2 trial, (2) able and willing to provide informed consent to take part in the interview, and (3) willing to have the interview audio-recorded.

As IPA principles require homogeneity around a phenomenon of interest [19,34], this was implemented in terms of adults experiencing distressing voices who have received AVATAR therapy. A predetermined recruitment strategy was consistently followed by trial coordinators at each site. This involved initially using consecutive sampling methods to invite participants for interview who had recently completed the final phase of the AVATAR2 trial. To achieve representativeness across sample characteristics and therapy delivery, purposive sampling methods were then applied to ensure that (1) the sample represented all 4 trial sites, (2) the gender and ethnicity of the sample were representative of participants across the trial, (3) the sample represented both AV-BRF and AV-EXT therapy arms, (4) the interviews covered experiences of working with a range of therapists, and (5) those recruited included a proportionate number of those who did not complete the intervention.

The aim was to recruit 20 participants across the 4 trial sites, including 4 participants who dropped out of therapy.

Semistructured interviews were guided by a topic guide (Multimedia Appendix 2) that was developed iteratively in collaboration with peer researchers, trial coordinators, therapy leads, and site research workers. Key areas of questioning were based on the research questions and initially drafted by the study team in line with IPA principles, such as focusing on the individual’s emotional experiences and associated meanings. The guide was then revised across 4 consultations, involving a total of 5 peer researchers. For example, questions were reworded to ask about “the voices” instead of “your voices” to ensure destigmatizing and nonjudgmental language. Interview questions explored participants’ (1) experience of dialoguing with a representation of their distressing voice, (2) TA formed with the therapist, and, where applicable, (3) reasons for not completing therapy.

Interviews were conducted either in person, online via Microsoft Teams, or via telephone, depending on participant preference. All interviews were conducted by a site research worker and peer researcher. In line with Harding et al [35], peer researchers took the lead in interviewing all participants while the accompanying site research worker provided support and input as required. Peer researchers disclosed their relevant lived experience when introducing themselves, but it was at their discretion how much detail they shared during the interview. A total of 14 peer researchers were involved in leading interviews, all of whom attended training (Multimedia Appendix 3), role-play practice, and supervision. The lead author, ERE, was available to peer researchers, site research workers, and trial coordinators for support with all aspects of protocol delivery to ensure a standardized approach.

Data were analyzed using a dual approach of IPA and template analysis, as described by Bond et al [24,36]. Nineteen is a relatively large sample for IPA’s rigorous and idiographic approach. Therefore, a small core sample of 5 (26.3%) transcripts was included in IPA processes, with analysis then extended across the remaining transcripts using template analysis. In line with IPA principles, the selection of transcripts was based on the richness of data and ensured diversity in trial sites, gender, and ethnicity. To ensure homogeneity, as required for IPA [19,34], all selected transcripts were from therapy completers. This selection process was led by ERE with guidance from experts in qualitative research methods and input from peer researchers who conducted interviews.

IPA followed analytic processes described by Smith et al [18] and Smith and Nizza [37]. The 4 peer researchers who conducted the 5 selected interviews first listened to their interview recordings to reflect on their emotional responses and perspectives of participants’ experiences. ERE then met individually with each peer researcher in 1-hour consultations to discuss interpretations. Using a case-by-case approach, line-by-line annotations were made for each transcript, incorporating the peer researcher’s perspectives and noting descriptive, linguistic, and conceptual comments. Identified meanings were therefore informed by peer researchers’ interpretations and then formulated into experiential statements, which were subsequently clustered and interpreted for each case. Convergences and divergences of themes were then considered across cases, generating group experiential themes and subthemes. Excerpts of the analytic process are presented in Multimedia Appendix 4.

Before applying template analysis, the resulting coding structure (ie, group experiential themes and subthemes) alongside illustrative quotes was reviewed by 6 peer researchers across 4 consultations to develop the provisional template (Multimedia Appendix 5). This input extended initial interpretations by drawing on lived experience, for example, emphasizing the complexity of the avatar-voice association, highlighting the profound nature of participants’ emotional experiences, and helping to identify language that felt authentic. This template was then used to guide template analysis of the remaining transcripts, following the processes outlined by King et al [21]. Transcripts were introduced in subgroups, focusing first on therapy completers and, second, on those who dropped out of therapy. This clearly identified where these 2 subgroups aligned and where new experiences arose. Where data diverged, new codes were assigned, and the template was changed to accommodate these perspectives. Once all transcripts had been coded and the template had been developed iteratively, all transcripts were reviewed again to ensure reliability.

Furthermore, 4 peer researchers reviewed the finalized template and contributed to refining codes, wording themes, and interpreting results. For example, the subtheme initially named “Difficulties generalizing learning” was reworded to “Difficulties translating changes to voices” to ensure accessible, nonclinical language. Additionally, the subthemes “Initial doubts and anxieties” and “Positive, open-minded attitudes” were merged into “Open-minded attitudes despite initial doubts” to reflect peer researchers’ interpretation that open-mindedness acted as an overriding force against early anxieties.

IPA and template analyses were conducted by ERE with input from peer researchers across the analytic process, ensuring that lived experience actively contributed to meaning-making. The outcome of this dual approach was a structure of key themes and subthemes (ie, the finalized template) representing all 19 interviews, including the 5 transcripts analyzed using IPA.

All members of the AVATAR2 team reflected on the perspectives they brought to the study design, conduct, and analysis through whole team meetings, peer group supervision with site research workers and peer researchers, and the lead author’s reflective research journal. The team approach and wider context of the qualitative research program were published on the Open Science Framework [30]. The lead author (ERE), a White British female trainee clinical psychologist, had no involvement in the AVATAR1 trial nor in the wider AVATAR2 trial, as her role was limited to this qualitative study, and had no prior contact with participants. While her clinical experience of working with people with psychosis, including delivering cognitive behavioral therapy, may have fostered positive attitudes toward psychological therapy, she had no clinical experience specifically with AVATAR therapy. Completing the Jacobson and Mustafa [38] positionality map supported reflection on social identity and potential sources of bias.

This study was approved by the London-Camberwell St Giles Research Ethics Committee in December 2022 (reference 20/LO/0657) and was conducted in accordance with the principles of the Declaration of Helsinki.

The participants in this qualitative study signed informed consent before participating in the interview that included consent for publication. All potentially identifiable information has been removed from published material included in this study. All interviews were audio-recorded, transcribed, and anonymized. Participants were paid £20 (US $26.65) for taking part in the interview.

A total of 19 participants took part in qualitative interviews, which were conducted between July and September 2023 and took place between 11.4 and 96.4 (mean 53, SD 29.1) weeks after their final AVATAR therapy session. Overall, 6 (31.6%) participants were recruited from King’s College London, 4 (21.1%) from UCL, 4 (21.1%) from Manchester, and 5 (26.3%) from Glasgow. Demographic and treatment information of participants are presented in Table 1 alongside that for the whole study sample of those who received AVATAR therapy (AV-BRF and AV-EXT). Participants in the interview sample were broadly representative of the wider sample; however, differences were not statistically tested.

A total of 9.9% (16/161) of therapy completers and 4.3% (3/69) of those who discontinued therapy are represented in the sample. Of those who dropped out of therapy, 2 participants decided not to continue after 1 session prior to any avatar dialogue, and 1 participant dropped out after 8 sessions (including 5 avatar dialogues). The 3 participants who dropped out, Grace, James, and Alexander (pseudonyms), were all allocated to AV-EXT.

Integrating IPA and template analysis processes, 4 overarching themes were identified, from which a total of 14 subthemes emerged. Details of themes alongside the number of participants are represented in Table 2 and subsequently presented in more depth. This is summarized in Multimedia Appendix 6. Illustrative quotes are presented within the text and in Multimedia Appendix 7. Pseudonyms are used throughout.

This theme explores how participants’ experiences of working dialogically with verbatim voice content evolved across therapy. The subthemes “Initial challenges adjusting to avatar” and “Collaborative efforts facilitated meaningful connection to avatar” focus on initial adjustment processes. Subsequently, the subtheme “With therapist support, participants felt empowered to stand up to avatar” explores communication in dialogues, and the final subtheme, “Positive shift with voices,” highlights improvements in coping.

The role of the therapist in AVATAR therapy was emphasized. The subthemes “Felt safe, supported, and understood” and “Person-centered flexibility” highlight valued aspects of the therapist’s approach and qualities, and the subtheme “Significant impact of TA” explores the effect of the therapeutic relationship on engagement and outcomes.

This theme emphasizes the impact of individual participants’ approaches and experiences. The subthemes “Open-minded attitudes despite initial doubts” and “Determination facilitated engagement and outcomes” highlight personal strengths. The subtheme “Profound emotional experience” focuses on the depth of emotional experience. The subtheme “Offered novel approach to tackle voices” considers the novelty of the therapy approach to the individual.

The subthemes “Emotional challenges with avatar” and “Not the right approach for the individual at that time” focus on barriers and challenges faced by participants who dropped out of therapy, also relevant to some therapy completers. The subtheme “Difficulties translating changes to voices” relates to participants who completed therapy but struggled to access positive outcomes.

This study used peer research methods and a dual approach of IPA and template analysis to explore the experience of AVATAR therapy, with a focus on direct work with voice content and the role of TA. Findings align with many of the themes identified in AVATAR1 qualitative work [11], particularly in highlighting the value of communicating with a digital representation, learning strategies, and the role of therapist support. Those who discontinued therapy described challenges with the realism of working dialogically with an avatar, reflecting that this approach may not be effective for everyone.

Despite initial challenges with early exposure and in vivo communication, working dialogically with voice content led most to experience an increased sense of power and control over voices, consistent with other cognitive approaches for distressing voices [39]. Participant accounts aligned with core therapeutic aims and processes, particularly around empowerment and self-esteem [8,40]. Consistent with inhibitory learning theory [41], participants described challenging distressing beliefs and targeting safety behaviors (including submission) within avatar dialogues, resulting in reduced anxiety over time. Positive outcomes were varied and individualized, including reduced voice frequency and omnipotence, enhanced understanding and acceptance, and improved interpersonal functioning. This aligns with relational approaches for distressing voices [3,7,42] and reflects the broader secondary outcomes observed in the main trial, such as improved voice understanding and empowerment [9].

Although use of technology initially caused some anxiety, its role in externally representing voices was well accepted by most, mirroring AVATAR1 qualitative findings [11]. Participant accounts regarding their emotional connection to, or disconnection from, the avatar relate to the concept of sense of voice presence [12,13]. This is a key mechanism in fear activation [43], promoting emotional processing and generalization of learning [44,45]. In line with earlier work [13], most participants reported at least some degree of voice presence during avatar dialogues. However, early challenges involved either high levels of perceived realism leading to fear activation (particularly in the context of safety behaviors, especially avoidance, and reactivation of trauma memories) or difficulties accessing voice presence inhibiting emotional processes. This reflects the bidirectional relationship between presence and emotion [46]. Participant accounts support the assertion that engagement and positive outcomes are contingent on voices being brought online within dialogues to activate emotion (ie, anxiety) within a safe therapeutic space [13].

Some participants indicated that, in the context of a high sense of voice presence, they experienced temporary symptom exacerbation. However, consistent with previous findings [13], the emotional intensity of this experience appeared well-tolerated by most and promptly eased across therapy. In fact, many specifically valued coming through this experience because it meant they had directly confronted and overcome core issues. This is consistent with evidence of temporary symptom exacerbation when working through challenges in trauma-focused imaginal exposure [47] and other exposure therapies [48,49].

Participants identified how open-minded attitudes, resilience, and determination helped to overcome the challenge of direct work with voice content, particularly when navigating anxiety-provoking early dialogues. While often conflated with attendance [50], many emphasized that engagement went beyond simply turning up each week and involved “leaning into” therapy to access voice presence, emotional processes, and meaningful outcomes. The emotional experience of therapy ending was also highly individualized across both AV-BRF and AV-EXT, with some experiencing a profound sense of closure and others struggling with loss. This variability reflects individual differences and the diverse meanings and relationships involved in voice-hearing, highlighting that the experience of therapy ending can be shaped by personal meaning and relational factors [51].

Generalizing learning from avatar dialogues to voices in daily life could be challenging, reflecting broader literature and emphasizing the need for continued use of skills [52]. The 3 therapy completers did not experience notable positive outcomes, possibly reflecting reported challenges in accessing voice presence, as this has been linked to the transfer of knowledge and skills [13].

It is important to note that participants who discontinued therapy were underanalyzed in relation to voice presence, with only 1 included in the theme “Collaborative efforts facilitated meaningful connection to avatar.” The 2 other participants disengaged prior to any dialogues in the context of challenges with the realism of the software, meaning they did not reach the point of establishing presence. These issues are further considered in relation to dropout experiences below.

Participant accounts align with the concept of TA [53], emphasizing therapist qualities like warmth and empathy. Strong TA was established for all, including those who dropped out, underscoring its central role in therapy experience, engagement, and outcomes, consistent with other psychosis interventions [54-56]. Developing trust was perceived to be vital given sensitive therapy processes, such as sharing verbatim voice content, exposure to feared stimuli, and working with trauma. In particular, participants valued transparent, respectful conversations to ensure an appropriate balance of voice presence and safety, particularly when reenacting abuse via the avatar [8]. Therefore, TA could mitigate the impact of the therapist operating the avatar and facilitate continued engagement despite anxiety. Therapists’ emotional attunement within avatar dialogues enabled person-centered therapeutic input, which progressively reduced across therapy [40].

Despite initial concerns about the impact of technology on TA, most participants felt they had adequate access to human therapist support. In fact, the digitally mediated environment was perceived to provide therapists with unique insights into participants’ live experiences and behaviors, creating valuable opportunities for validation. Participants also valued therapists’ respectful curiosity about their experiences, values, beliefs, strengths, and cultural background. In this way, many described feeling empowered to take an active role in their therapy, emphasizing collaboration and choice [57,58].

Choosing to discontinue therapy could be an empowering choice, providing insights for recovery, but could also be experienced negatively, reflecting the contrasting experiences of therapy dropout [59]. This also emphasizes that attendance is only one component of engagement, and it is possible to meaningfully access therapeutic processes without reaching a treatment dose [50].

While challenges of exposure to the avatar eased promptly for most, the emotional toll of dialogues across therapy led 1 participant to drop out after 5 dialogues (8 sessions). Additionally, 2 other participants chose to discontinue prior to any dialogues due to difficulties with the realism of the avatar software. This suggests that AVATAR therapy may be overwhelming if voice-hearing experiences elicit high levels of anxiety [60], and, conversely, insufficient realism may disrupt voice presence and engagement. In this way, both too much and too little realism can create difficulties in establishing voice presence and activating target emotions in a safe setting, posing a risk for disengagement. However, many therapy completers who benefited from AVATAR therapy reported similar challenges in early sessions and initially questioned whether to continue, highlighting the role of personal choice. Factors such as life events outside of therapy, access to resources, and social functioning may influence an individual’s decision to continue with therapy despite challenges [61-63]. This was endorsed by therapy completers who acknowledged that AVATAR therapy may not be effective for all, depending on factors such as stage in recovery, stressful life events, and physical health, consistent with wider literature [47,64-66]. Taken together, this indicates that dropout may not simply reflect limitations in software realism but also readiness for exposure and suitability of the intervention for the individual at that time.

Similarly to AVATAR1 qualitative findings [11], 1 autistic participant struggled to engage with the idea of digitally embodying the voice, possibly suggesting additional challenges for those with this comorbidity. This highlights the need for further research specifically focused on autistic voice-hearers.

The study demonstrated methodological rigor with adherence to peer research methods and IPA principles, such as using a phenomenological approach, ensuring reflexivity, and using an iterative process of analysis [18]. Steps were taken to standardize interview delivery through training, role-play, co-interviewers, and supervision. The peer researcher status of interviewers likely supported participants to feel comfortable, understood, and willing to open up [67]. Additionally, efforts were made to include participants who discontinued AVATAR therapy, enhancing learning around potential reasons for making this decision.

While integrating 2 qualitative analysis methods enabled both depth and breadth [36], this introduced epistemological tension as IPA emphasizes idiographic depth while template analysis prioritizes cross-case patterns. The careful layering of these methods was guided by advice from experts in qualitative research methods and supported by reflexive practices and peer research methods. However, some complexity may have been reduced when moving to broader themes, with the risk of diluting the nuance of individual voices. This highlights the trade-off between depth and generalizability [21].

Selecting transcripts for IPA based on richness of data (often defined in terms of depth, detail, and context provided within narratives) has been questioned due to the associated risk of bias in privileging some accounts over others [22]. In line with IPA principles, transcripts of participants who discontinued therapy were not included in IPA to ensure homogeneity [19,34]. While the integration of template analysis enabled clear comparison of completers and noncompleters and ensured that all participants’ perspectives were systematically incorporated [23], this may have privileged voices that were more positive and engaged. This could limit the transferability of findings, as the therapeutic processes identified in detail may not fully reflect the experiences of those who disengaged or found the intervention unsuitable.

Participants were invited for an interview after completing the 28-week follow-up to avoid any influence on trial outcomes. Therefore, interviews were conducted several months after completing AVATAR therapy, enabling exploration of long-term perceived impacts. However, this delay may have also introduced recall bias or retrospective reinterpretation of participants’ experiences during therapy. Social desirability bias and self-selection bias may have impacted study recruitment. While we recruited a broadly representative sample with increased numbers of participants who dropped out of therapy compared to AVATAR1, only 4.3% (3/69) of those who dropped out were represented compared to 9.9% (16/161) of therapy completers. As it was not possible to recruit the intended 4 participants who dropped out, and the interview sample included 2 participants who dropped out prior to any dialogues, we have likely missed valuable experiences and perspectives from those who were not willing to participate. Additionally, as this study focused on TA between the voice-hearer and therapist, we may have missed important insights into the “triangle of alliance” between the participant, therapist, and digital platform (ie, avatar) [68].

A total of 10.5% (2/19) of participants were from Black backgrounds compared to 17.8% (41/230) in the wider AVATAR2 sample. The experiences of Black participants are the focus of a stand-alone qualitative study within the AVATAR2 Programme of Qualitative Research, which is currently being prepared for publication [69]. Concomitant recruitment for both qualitative studies is a likely factor in the lower representation of Black participants within this study.

The AVATAR2 trial aimed to evaluate the efficacy of 2 forms of AVATAR therapy, determine optimal therapy delivery, and consider implementation in National Health Service (NHS) settings. Supported by a National Institute for Health and Care Excellence Early Value Assessment [70], which has recently recommended further testing in routine NHS settings, these qualitative findings will support the optimization and wider implementation of this approach.

Establishing voice presence and activating target emotions within a safe setting has been emphasized as central to engagement and outcomes. Software developments may improve the realism of digital representation, enhancing sense of voice presence. Findings emphasize the need for improvements to the tailoring of avatars across different ethnicities and genders to ensure equal opportunities to establish voice presence. However, as technology advances, increased realism might activate strong emotional responses and increase fear within the digital environment, particularly pertinent given the ongoing trial of AVATAR-VR [71]. Therefore, flexible therapist support will be crucial to ensure participants feel safe, balancing exposure with tolerability [8,47]. Meanwhile, dropout experiences suggest that difficulties establishing presence may not only reflect technological limitations in software realism but also individual readiness and suitability, emphasizing the importance of personalizing AVATAR therapy to the person’s stage of recovery and capacity to tolerate exposure.

Limitations suggest that our findings are most transferable for understanding therapeutic processes when engagement is maintained, providing less insight into disengagement and nonresponse. Therefore, future work should further explore the perspectives of those who discontinue or gain limited benefit. Such insights could inform early screening and tailoring, for example, around readiness for exposure and tolerance of voice presence.

In this context, integrating emotion regulation and anxiety management techniques may enhance coping and sense of safety [47,72], as demonstrated by Paulik et al [73] in supporting voice-hearers to engage with exposure therapies. Therefore, future research could consider if an emotion regulation phase improves the tolerance of this experience, ensuring trauma-informed care and integrating other therapeutic approaches [47]. Additionally, tools to support normalizing discussions around common voice content and transparency over potential temporary voice exacerbation could facilitate collaborative engagement and reduce distress.

Findings highlight the importance of TA, fully informed consent, and patient choice when proceeding with exposure and dialogues. Peer support spaces could provide additional support, increase motivation, and promote engagement and outcomes [74,75]. Further research could explore the triadic relationship between the participant, therapist, and avatar. In particular, it would be valuable to examine how the concept of TA applies to the avatar, given its role in representing the distressing voice yet also the individualized relational context, meaning that there may be a therapeutic focus on fostering compassion toward the voice [8,76].

The range of emotional experiences across AV-BRF and AV-EXT highlights the complexities and relevance of individual differences. AVATAR2 RCT findings suggest that AV-EXT produces stronger results, but this was in the context of higher dropout rates. As AVATAR therapy moves from clinical trials and into routine care, there are increased opportunities for tailoring the therapy to context, including flexibility in duration, use of booster sessions, and potential integration into long-term therapy.

This study provides valuable insights into the participant experience of AVATAR therapy, enriched by peer research methods. Overall, experiences were positive, with all reporting strong TA and most gaining power over the avatar and, consequently, voices. There were challenges with in vivo exposure and dialoguing with voice content, but this was often mitigated by TA. Engagement and positive outcomes were contingent on establishing voice presence and activating target emotions while retaining a sense of safety. Potential barriers to engagement and outcome have been suggested, emphasizing the need for software developments, person-centered flexibility, and informed choice.