Sometimes, and for reasons that have very little to do with headsets. Across the research, immersive training produces a moderate average benefit that shrinks close to zero once you look only at the best-designed studies, while the things that reliably predict a result are how well the simulation matches the real task and whether the design makes the learner do something. We sell an immersive training platform and we would rather you knew that than found out later.
Key takeaway: The defensible claim is not that VR beats classroom training. It is that a well-matched simulation of a task you cannot safely or affordably rehearse is worth building, and that the display device is close to irrelevant to the learning. Buy on fit, reach and cost, not on effect size.
The headline number, and what happens to it
The most rigorous recent work is a second-order meta-analysis: Layadi and colleagues pooled 26 meta-analyses covering 180 studies, 281 comparisons and 18,792 participants, then stratified by risk of bias. Overall effect: SMD 0.55. Studies at high risk of bias: 0.38. Studies with some concerns: 0.43. Studies at low risk of bias: 0.35, and not statistically significant. They found modest publication bias, substantial unexplained heterogeneity, and concluded the evidence is “insufficient to support firm, general conclusions” (Educational Psychology Review, 2026).
That is the shape of this entire literature. Positive on average, and the better the study, the smaller the effect.
| Review | Scope | Result |
|---|---|---|
| Layadi 2026 | 26 meta-analyses, 18,792 participants | SMD 0.55 overall, 0.35 and non-significant in low-bias studies |
| Scorgie 2024 | VR safety training, 52 articles | Knowledge g=0.640, retention g=0.838, but prediction intervals cross zero |
| Kaplan 2021 | XR training meta-analysis | XR “equally effective” as non-simulated control training |
| Howard 2021 | VR training programmes | About 0.5 SD benefit; hardware not a moderator, task fit “a very strong moderator” |
| Coban 2022 | 48 studies, 3,179 students | g=0.38; mixed reality showed negative effects |
| Cromley 2023 | STEM learning, 18 studies | g=0.33, and desktop effects exceeded head-mounted display effects |
Read the Scorgie paper properly if you read one. The knowledge-acquisition prediction interval runs from -0.86 to 2.13 and the retention interval from -1.14 to 2.82, which means roughly a one-in-five chance that traditional training outperforms VR in any new study. Its retention finding also carries significant publication bias, Egger p=0.001. Retention is the most marketed claim in this industry and the least well supported.
The finding that matters most for buyers
Howard and colleagues found that display hardware, input hardware and game elements were not significant moderators of VR training effectiveness, while task-technology fit was (Computers in Human Behavior, 2021).
Parong and Mayer showed the mechanism from the other direction. In their first experiment a slideshow beat immersive VR on the post-test, d=0.92, while VR won on motivation and interest. In the second, VR plus a written summarising activity beat continuous VR, d=1.12 (Journal of Educational Psychology, 2018). Makransky’s team found the same pattern with EEG-measured cognitive load: the desktop group learned significantly more, d=0.80, while the VR group reported far more presence (Learning and Instruction, 2019).
Immersion buys attention. Attention is not learning. What converts one into the other is making the learner decide, retrieve or explain something, which is a design choice available on any device.
What about browser and 360° delivery
This matters to us commercially, so here is the unflattering version first. Schroeder, Siegle and Craig meta-analysed 26 studies, 82 comparisons and 2,142 participants on learning from 360° video and found “no significant benefit nor detriment” compared with other media (Computers & Education, 2023).
What the same body of work does support is that the headset is not where the effect lives:
- A randomised study of 136 university students comparing a Quest 3 with a laptop or tablet on identical content found no significant learning difference, F(1,133)=2.997, p=0.086, and significantly higher cognitive load in the immersive condition (Applied Sciences, 2026).
- A randomised pilot with medical students watching the same 360° physical examination video on a headset or a monitor found no significant difference in performance, p=0.267 (Journal of Medical Extended Reality, 2026). Small, n=19, so treat it as directional.
- Cromley’s STEM meta-analysis found desktop effects larger than head-mounted display effects.
Put together: 360° scenarios do not have a learning advantage over other good media, and they do not have a disadvantage either. The case for them is cost, reach, speed of production and the fact that a whole cohort can do them this week on the devices they already own. That is a real case. It is just not an effect-size case, and anyone selling it as one is ahead of the evidence.
Clinical training, where the evidence is strongest
Healthcare has the best studies, largely because it has the best outcome measures.
Kyaw and colleagues, in a UK-led review for the Digital Health Education Collaboration, found VR improved knowledge against traditional learning, SMD 0.44, 95% CI 0.18 to 0.69, and cognitive skills at SMD 1.12, 95% CI 0.81 to 1.43. They also noted that no included study reported patient outcomes, behaviour change or adverse effects (JMIR, 2019).
Ropponen’s 2025 meta-analysis of 19 randomised trials with 769 nursing students found significant effects on knowledge, d=0.53, and skills, d=0.65, but non-significant results for confidence and satisfaction, with very high heterogeneity (Nurse Education Today). Both significant confidence intervals nearly touch zero.
For surgical simulation, which is the oldest and best-evidenced corner of all this, a UCL-led meta-analysis found VR training improved OSATS scores against no additional training, mean difference 6.22, and cut operating time, but found no significant difference between VR and other forms of simulation training (BJS Open, 2022). The Cochrane review reached a similar place: shorter operating times, all trials at high risk of bias, no trial reporting any patient outcome (Cochrane, 2013).
Does it last, and does it transfer
Rarely measured, is the honest answer. Stefan and colleagues mapped 136 safety-training studies against the Kirkpatrick model: 72% measured learning and 66% measured reaction, while “very few” measured behaviour or organisational results (Virtual Reality, 2023).
The best long-follow-up trial randomised 117 medical students and followed 108 of them for a year. VR came out ahead on specific OSCE stations and on self-reported career determination, but there was no significant difference in overall OSCE score at one year (BMC Medical Education, 2023).
And in the only real organisational field study we could find, Bertram and colleagues trained actual police officers on ground-to-helicopter coordination and found VR transfer to real complex scenarios comparable to standard training, not better, with VR scoring lower on motivation and perceived value (Computers in Human Behavior, 2015).
Statistics you should stop repeating
If you have been sent a VR business case recently, it probably contains at least two of these.
| Claim | Where it actually comes from |
|---|---|
| ”4x faster to train, 275% more confident, 3.75x more emotionally connected” | PwC’s own 2020 study, not peer reviewed, self-reported outcomes, with the VR content built by vendor Talespin |
| ”Walmart associates scored 70% higher” | Strivr customer story and Walmart newsroom. No methodology, no independent study |
| ”Learners retain 75% of what they learn by doing” | Edgar Dale’s Cone of Experience with percentages Dale never published. No empirical source exists |
| ”VR improves recall by 12%“ | Krokos 2019, a real study in which people recalled the positions of 21 faces in a memory palace. Not training research |
| ”30% increase in safety awareness” | Qawqzeh 2025, non-randomised, purposive sampling, entirely self-reported, three identical round 30% figures |
| ”A US government study found VR training effective” | The DHS SAVER TechNote contains no effectiveness data at all |
The same standard applies to us. The figures on this site, 94% first-time pass at Cardiff and Vale, 87% inspection completion at Zurich, 45% better hazard identification at ACT Training, are customer-reported results from single deployments with no control group. They are real numbers from real projects and they are not evidence that immersive training works in general. Ask every vendor, including this one, which of those two things a number is.
So how should you decide
Skip the effect sizes and ask four questions.
- Can you practise this task any other way, safely and affordably? If yes, the research says you will probably get a similar result doing it that way. If no, simulation is the only option and its cost comparison is against not training at all.
- Does the simulation match the real task closely? Task-technology fit is the strongest moderator in the literature. A scenario shot in your own ward or workshop scores better on this than a generic environment, whatever the render quality.
- Does the learner have to do something? Decisions, retrieval, explaining. A 360° video with no decision point in it is a video.
- Can everyone actually access it? A method that reaches a whole cohort at 0.35 effect beats a method that reaches nine people at 0.55. This is where browser delivery earns its place, and it is a logistics argument rather than a learning one.
The short answer
Is VR training effective? On average yes, modestly, and the effect shrinks in the best-designed studies. It is not reliably better than good conventional training, the headset is not where the benefit comes from, and the retention claims are the weakest-supported part of the whole field. Build immersive training when the task cannot be rehearsed any other way, match the simulation to the real job, make the learner decide something, and measure your own results rather than quoting anyone’s.
We keep this post updated as new reviews land. If you want the practical side, do you need headsets for VR training covers deployment, and off-the-shelf library vs building your own covers the build decision.