News / August 2026

What the Google VR study actually found

Leveraging Virtual Reality to Enhance Diversity and Inclusion training at Google is a real academic paper published at CHI 2024, not a product page. Elie Bursztein, Karla Brown and Patrick Gage Kelley of Google wrote it together with Leonie Sanderson, whose affiliation on the paper is listed as Equal Reality, Australia. It was published in the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA '24), 11-16 May 2024, in Honolulu. PTR did not write it.

It is a genuinely encouraging paper. But the most useful thing in it is not the headline satisfaction number. It is a single line, from an open-ended survey response, about what happens once you take the group discussion away.

What was built, and by whom

The application is four short VR scenarios, built in the Unity engine, that put a trainee inside a workplace bias incident: a colleague's technical expertise wrongly assumed because of how they look, or a new team member left out because the group assumes everyone shares the same cultural references. They were developed between 2021 and 2023 by a group of 13 Google volunteers involved in Google DEI activities, working in partnership with a VR studio. The paper does not name the studio in that passage, though its acknowledgements thank "the entire team at Equal Reality who helped us design and build the virtual reality environments". Every scenario works the same way: the trainee experiences the situation from one perspective, then again from the other person's point of view.

What the numbers actually were

These are small numbers, and the paper is upfront about that. Development wrapped with seven QA testing sessions and 20 volunteer testers; 11 of them, 55 percent, filled out the anonymous follow-up survey. Of those 11, 10 (91 percent) said yes when asked whether the scenarios felt real. The other one, R9, said it felt no different from video training.

Google's first live cohort ran in September 2023. Invitations went to 32 members of the Google security team's senior staff; 26 accepted and took the training, 23 of them in a headset and three on desktop. Seven of the 26 returned an anonymous follow-up survey afterwards, the paper's own count, given as "7 (30%)". All seven called the training a satisfying or very satisfying experience, a 4.7 out of 5 average. On whether it helped them better understand DEI issues and how to resolve them, the paper's own figure is "(7/8)" at a 4.1 out of 5 average agreement, a fraction that does not square with the topline count of seven. The paper does not explain the gap, and neither will we invent one.

Why the authors kept the debrief

This is why this article exists. When the QA testers were asked how important the post-VR group discussion was, 9 of the 11 (82 percent) rated it useful or very useful. In the open-ended feedback, one tester put it more bluntly than any statistic could:

"Without the discussion component, [this training] doesn’t have any more value than past [styles of] DEI training"

The authors took that seriously, and their own gloss is careful: this feedback "supports the hypothesis that having a post session is critical to the success of VR DEI training". A hypothesis supported by opinion is not a demonstrated finding. The study did not test the debrief separately, and it did not measure what anyone learned. What it does show is that the people who went through it rated the discussion highly, and that the authors changed their own practice because of it: they allocated 25 minutes to the debrief in every training session that followed.

What the authors say still doesn't work

The paper's own discussion section names five concerns, in this order:

1. Hardware issues. Discomfort from a headset's weight or from glasses worn under it, dizziness reported elsewhere in the literature, and headsets that remain inaccessible to some users. The authors say alternative training formats will need to keep being offered.
2. Scaling costs. Hardware cost, scaling tech support, remote access to the technology, and training enough facilitators.
3. Fidelity, quality. Higher fidelity costs more to build, which the authors say "may restrict the trainings to only the highest-end, and also most expensive, headsets."
4. Scenario depth. Participants wanted more agency, more freedom of movement and engagement, and more complex storylines with more conversational turns. The paper raises AI-driven characters as one possible answer, then immediately qualifies it: using that technology for training "is unproven, and would require further research and strong guardrails to meet learning objectives and not lead to unexpected or offensive responses in sensitive conversations."
5. Novelty. The authors' own words: "a variable we couldn’t control for is how much this positive feedback is due to the novelty effect."

What this means if you're considering VR training

Read this paper as an honest first data point, not a verdict. It studied one team, at one company, with sample sizes in the tens rather than the hundreds, and it reports self-reported feedback, not measured behaviour change. What it does support is that participants rated the discussion afterwards highly, and that the authors treated it as essential enough to build 25 minutes of it into every session. PTR's own delivery builds that debrief into every session with a trained facilitator. Ask any VR training provider, us included, what happens in the room once the headset comes off.

The paper

Bursztein, E., Brown, K., Sanderson, L., and Kelley, P. G. (2024). Leveraging Virtual Reality to Enhance Diversity and Inclusion training at Google. In Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA '24), May 11-16, 2024, Honolulu, HI, USA. DOI: 10.1145/3613905.3637128.

Read our summary of the paper, the case study this training became, the paper itself as a PDF, or more on how PTR facilitates VR training.