This study followed the CONSORT (Consolidated Standards of Reporting Trials) guidelines (Multimedia Appendix 1) and used a randomized controlled experimental design with 3 groups: a control group, a comparison group, and an intervention group. After eligibility screening, we recruited 156 participants from a large public state university in Texas between June 2023 and August 2023. The recruitment ended when the required sample size was attained. The sample size was determined based on a power analysis wtih with α=.05, power=0.90, effect size=0.25, and correlation among repeated measures=0.50. The participants were randomly allocated to 1 of the 3 experimental conditions in a 1:1:1 allocation ratio (ie, 52 in each group). Random assignment was conducted by one of the authors (KP) using Microsoft Excel using the RAND() function. Participant codes, initially listed by group (eg, control group from S001 to S052, comparison group from S053 to S104, and intervention group from S105 to S156),) were sorted in ascending order of the generated random numbers. After excluding participants who did not show up or acted as pilot-testers of the research protocol, the remaining participants received no intervention (control: n=33), alternative intervention (comparison: n=37), or target intervention (intervention: n=37). Figure 1 shows the CONSORT diagram of participant flow in this study. No restrictions such as blocking or stratification were applied during randomization. Allocation was not concealed from the researchers assigning participants to the experimental conditions due to the nature of the interventions; however, participants were not informed of alternative conditions or study hypotheses at the time of assignment. The university enrolls a diverse student body of more than 44,000 undergraduate and graduate students, providing access to a broad and representative sample of young adults. Recruitment was conducted using TechAnnounce, the institution’s official campus-wide electronic announcement platform, which distributes daily bulletins to all enrolled students, faculty, and staff via email.

To ensure consistency and broad visibility, the recruitment message was posted once per week during the recruitment window. Each announcement contained a standardized description of the study purpose, eligibility criteria, and a hyperlink to an online sign-up sheet. Interested students who clicked the link were directed to a screening form that assessed the inclusion and exclusion criteria.

The eligibility criteria required participants to be aged at least 18 years and currently enrolled at the university. To control for previous exposure to atypical emergency experiences and the baseline emotional state, individuals who currently worked or had previously worked as first responders were excluded, as were those who self-reported a history of depression, PTSD, or other serious mental health concerns. Additionally, students who indicated a fear of water or a tendency toward motion sickness with VR applications were excluded to minimize the risk of adverse reactions during the experiment.

The mean age of the participants was 25 (SD 6.00; range 18-50) years. There was no significant age difference among participants across the three groups (F_2,104=0.214; P=.81).

Participants followed a 3-stage experimental protocol (Figure 2). In stage 1, participants in all groups watched a 3-minute video of a real house fire that occurred in a residential area in the United States. A large TV monitor (65-inch type) was used to display the video, enhancing the realism of indirectly witnessing the fire. The video was recorded by a local fire department in 2019 and posted on YouTube (Alphabet Inc) shortly after. Figure 3 shows different snapshots of the house fire video. The fire ruined most of the house and spread to a neighboring house.

In stage 2, each group was exposed to different treatments. Control group participants did not do any task while staying seated for 3 minutes. Participants in the comparison group read a paragraph describing the Red Sea. The length of the text was adjusted so that it would take approximately 3 minutes to read. Excerpts from the paragraph are as follows:

Participants in the intervention group watched a 3-minute 360° VR video of the Red Sea that shows similar natural scenes to those described in the paragraph. Figure 4 shows snapshots of the VR video (source: AirPano). This VR video provides a live recording of underwater nature that contains sea animals and coral reefs. The choice of the Red Sea as semantically irrelevant VR content was based on the opposite attributes of the underwater nature to the house fire in several representative aspects: red versus blue (color), fire versus water (element), hot versus cold (temperature), and artificial structure versus underwater nature (environment).

In stage 3, a forced recognition test was conducted to assess the memory accuracy of the house fire that participants in all three groups watched in stage 1. A forced recognition memory test was used, in which participants are required to choose 1 image from multiple options that they believe they have seen previously. The forced recognition test is widely used in memory testing studies [28,29].

To make fake images of the house fire video, we used Adobe Firefly (Adobe Inc) [30], an image-generating artificial intelligence (AI) program available to the public. This generative AI program has multiple features, such as generative fill. An example of the fake images used in the forced recognition test is shown in Figure 5.

A total of 15 pairs of a true image and a fake image were used in the forced recognition test. Memory accuracy was calculated using the following equation:

This study was conducted in accordance with a research protocol approved by the Texas Tech University’s institutional review board (institutional review board number 2022-966). Written informed consent was obtained from all participants before their participation, covering both their voluntary participation in this study and the publication of data collected from the research.

Findings from this study showed that a semantically irrelevant VR experience (diving in the Red Sea) boosted positive emotion and lessened negative mood after indirectly witnessing an adverse event (ie, watching a residential house fire). Specifically, participants’ positive affect improved only with the semantically irrelevant VR experience, whereas negative emotions were mitigated by watching a wall (control group), reading a text (comparison group), and watching a VR video (intervention group). Nevertheless, there was no difference in memory accuracy between the groups. These results indicate that the semantically irrelevant VR experience of this study (ie, watching an immersive VR video) did not affect the memory of the house fire. However, regardless of how accurately participants remembered the details of the previous traumatic event (ie, the house fire), providing the semantically irrelevant VR experiences such as doing virtual scuba diving and reading a paragraph about a tropical sea helped participants adjust their emotional states.

These findings show the promise of VR technology as an alternative tool to address chronic mental health problems among first responders. Given that existing mental health interventions, such as postincident debriefing, mindfulness meditation, and CBT, require professional support and long-term commitment, semantically irrelevant VR experiences represent a more accessible solution.

This study’s findings corroborate the remedial effects of a virtual trip, or “a virtual getaway effect,” derived from immersive VR environments [24,25]. Interestingly, reading a paragraph describing the Red Sea also helped reduce negative emotions, whereas participants in the control group did not report any substantial changes in positive emotions. The latter signifies the needs for mental health support in any form, ranging from simple guided imagery training [41,42] to more interactive solutions such as VR experiences [43]. Thus, semantically irrelevant VR experiences can act as an alternative and complementary approach, compatible with the unique cultural traits of first responders, due to their high accessibility. As highlighted in a review of VR solutions for mental health [44], VR-based interventions can be useful in treating a variety of mental health issues, such as eating disorders and depression, especially if VR is integrated with other advanced technologies such as large language models and gamification.

It is also noteworthy that even a brief virtual getaway (3 min) produced substantial changes in participants’ emotional states. In contrast, taking an actual vacation requires considerable time and expense. Moreover, the temporal gap between a traumatic event and a real trip may be long, potentially reducing the remedial effects of the vacation. Thus, a virtual getaway using the semantically irrelevant VR experience offers a quick and affordable way to alleviate negative emotions following exposure to traumatic events. Additionally, our findings corroborate the stress-reducing effects of immersing oneself in natural environments (eg, forests and mountains) [45,46]. In future studies, diverse VR apps such as virtual concerts and sports events [47] can be considered as options to personalize the VR experience as a psychological aid.

One methodological implication of this study is the use of an image-generating AI program (Adobe Firefly). Previous studies used simple drawings, photos, and verbal narratives to examine the malleability of human memory [48,49]. Owing to advanced AI technology, it now takes only a few mouse clicks to create a fake image that looks genuine. To our knowledge, this is the first study to use a generative AI tool to create fake images for a memory test. Considering the rapid development of generative AI applications, the methodology established in this study will inform future experiments that necessitate AI-generated artifacts. Nevertheless, the AI-generated images were not validated for their realism against a separate and independent sample. Thus, future studies need to assess the perceived level of fakeness of AI-generated images when they are used in recognition-based memory tests.

Despite the novel benefits of the semantically irrelevant VR experience on emotions, our study could not reveal the memory suppression effect, given the nonsignificant difference in memory accuracy. Therefore, the relationship between the adverse event memory and its associated emotions remains inconclusive. We suspect several reasons behind this lack of evidence.

First, the forced recognition test may not be the best way to assess the memory accuracy of an initial event (ie, a house fire). As shown in Figure 5, only a limited number of items were modified (eg, a lawn under fire to a small car). In the house fire video, there are numerous elements that a participant could potentially catch or miss (eg, flames, adjacent trees, firefighters, and houses). Thus, there was no guarantee that the participants observed the items that were modified in the fake images. If the participants did not observe such items, the recognition-based test becomes a binary choice with random chance. To address this issue, alternative methods for assessing an individual’s memory should be considered, such as a recall-based memory test, which has been used in previous memory research [29].

Second, it is unclear whether watching a house fire created sufficiently negative emotions. Experiencing a house fire directly can cause immense stress and suffering. However, watching a house fire on a computer screen without any severe injuries and damages may not feel as traumatic, which may have weakened the memory suppression effect of the subsequent semantically irrelevant VR experience. In future studies, a stronger dosage of a traumatic event needs to be considered to create more distressing emotions. Alternatively, future studies may consider recruiting those with traumatic memories associated with a fire to see how the same semantically irrelevant VR experience influences their emotions. Relatedly, we did not assess the pre-experiment emotional state, leading to a limited manipulation check to see whether the house fire video made sufficient changes in the viewers’ emotions. Thus, future studies require a prestudy assessment of the participants’ emotional state. In addition to self-reported surveys such as the PANAS survey [31], we recommend that physiological measures, such as heart rate, skin conductance, or eye-tracking metrics, be collected to find associations with self-reported survey results.

Third, our study used a passive VR experience that only required participants to watch. Therefore, watching the VR video (Red Sea) may not have consumed much cognitive resources for attention and perception. It is speculated that if a participant uses more cognitive resources in a VR environment, the memory suppression effect and the psychologically remedial effect would be stronger, making it more difficult to retrieve the memory of a previous traumatic event. Indeed, previous studies report that active VR requiring psychomotor activities consumed more cognitive resources [50,51]. Moreover, this study did not reveal whether the changes in emotions occurred due to either the semantically irrelevant experience or the immersive environment afforded by the VR. Thus, future research needs to consider using active VR that consumes more cognitive and physical resources. Furthermore, it is recommended to vary the VR modality (2D experience vs 3D experience) while keeping the semantically irrelevant content constant.

Additionally, longitudinal studies with multiple follow-up assessments are recommended to examine the long-term effects of semantically irrelevant VR on memory and emotions, as this study examined only immediate effects.

Our work showed the promise of immersive semantically irrelevant VR as a novel solution to addressing mental health issues faced by first responders. Although semantically irrelevant VR did not result in a memory suppression effect, this study found that a virtual trip had substantial effects on both positive and negative emotions after individuals indirectly witnessed an adverse event. Considering the prevalence of severe mental health issues among first responders and the paucity of mental health support for them, findings from this study can inform quicker and more accessible VR solutions for psychological issues, in combination with existing mental health interventions.