Out-of-Body Perception Test Comes Up Empty, With One Exception

The laptop screen faced the wall. The door was shut. Seven meters away, a volunteer lay still, trying to leave her body.

Across thirteen trials analyzed in Brazil, volunteers described that hidden, computer-selected image no more accurately than random guessing would predict. The research team argues the likely explanation isn’t that anyone’s awareness actually left their body, it’s that a random image on a screen is a bad test for whatever a wandering mind actually picks up on.

That’s the clean, explainable part of the story. The messier part came later, in the debrief. Two volunteers who couldn’t describe a photograph correctly turned around and described — with no apparent way of knowing — exactly what the researchers on the other side of that door had been doing.

Claims like these, that consciousness can wander free of the body and pick up information no eye or ear could reach, have circulated in case reports and small studies since the middle of the twentieth century, most famously in a session where a woman known only as Miss Z correctly read a random five digit number from across a room, a one in 100,000 shot. None of the work that followed managed to repeat that kind of result reliably, which is what brought a team from the University of Virginia’s (UVA) Division of Perceptual Studies back to the question with a tool the earlier researchers never had: language models that can score how closely two descriptions match.

Forty practitioners at a Brazilian training institute for out-of-body techniques, the International Institute of Projectiology and Conscientiology, were asked how likely they thought they were to leave their bodies on command; the 21 who rated their odds above 40 percent got a session. Each lay in one room while, next door, a researcher pressed a key that pulled a single image at random from a set of 100 ordinary objects, photographed against a blank background, then turned the screen to the wall and left, closing the door behind. For up to an hour, sometimes ninety minutes, the volunteer tried to perceive it; afterward, in an interview, they described whatever they believed they had seen. Ten said they had left their bodies during the attempt; thirteen, including some who described only an inner mental screen rather than a full separation, gave a description detailed enough to score.

Scoring meant feeding each description, along with captions for all 100 possible targets, into a text embedding model and asking where the real target ranked among the field, from first place, a perfect match, to hundredth, a total miss. By that measure, the real target came in anywhere from 16th to 92nd across the 13 trials, a spread that put the group’s combined performance no closer to the top than pure chance would predict.

A Coin Flip, Repeated Thirteen Times

None of the 13 individual results reached statistical significance on its own, and a combined test across all of them landed at a middling average rank of 61.5 out of 100, against a chance expectation of 50.5, a gap too small to trust. A second, more sensitive pass on the same numbers even nudged toward the opposite of what a believer would hope for, a marginal dip below chance that the authors flag themselves as too fragile, in a sample of just 13, to mean anything. Split by whether someone reported a full out-of-body sensation or only an inner impression, neither group did better than the other, and neither beat the odds.

“I hope this study encourages researchers to think creatively about how we test these experiences,” says Marina Weiler, the neuroscientist at UVA’s Division of Perceptual Studies who led the work. “The unexpected findings give us a potentially useful new direction, but they also need to be tested prospectively and under rigorous conditions before we can know what they mean.”

The team had a way to check whether their scoring pipeline actually worked: they fed it artificial intelligence generated captions that described each target image accurately, on purpose, and asked it to find the match. Given a literal, accurate caption, the system placed the correct image first, out of 100, in all 13 trials without exception; given a looser, more conceptual caption, it still got there in 10 of 13. That is a strong case that a genuine match, had the volunteers produced one, would have shown up. But those control captions came from the same computer model that captioned the targets in the first place, speaking a kind of shared machine dialect, and the authors concede they cannot rule out that ordinary human phenomenological language, looser and stranger than an AI caption, would register as a weaker match even when it was accurate. There is also a plainer statistical problem: with only 13 usable trials, the study had roughly a coin flip’s chance, about 55 percent, of detecting even a moderate real effect, well short of the 80 percent researchers usually want before trusting a negative result.

None of that rescues the case for out-of-body perception; a null result stays a null result. It does mean a clean failure to find something is not quite the same as proving nothing is there, and the paper says as much itself.

What the Closed Door Let Through

What is harder to wave away happened outside the trial’s own scorecard. Two volunteers, unprompted, described the exact positions and activities of the two researchers behind that closed door: one reading in a corner while the other watched a laptop screen, in one session; one taking notes on the right while the other sat at the computer, in the other. Session logs confirmed both accounts were accurate at the time, and the researchers say there was no ordinary way for either volunteer to have seen inside.

Weiler and her colleagues do not treat this as evidence of anything, since it was not what the trial was designed or statistically powered to test, and they say so plainly in the paper. But they read it alongside a small, telling detail from the same data set: one of the two excluded participants, rather than attending to the assigned target, described an out-of-body experience in which she was in a war zone, helping children, a vivid, personal scene with nothing in common with a photographed object on a blank background. That detail lines up with a broader argument the same lab has made elsewhere, in a companion paper proposing five things that shape whether a laboratory test of this kind succeeds: a participant’s emotional state, their relationship with the researchers, the researchers’ own intentions, how practiced the participant is, and whether the target itself carries any personal weight. Their proposed fix for the next version of this trial follows that same logic and is almost playful: swap the anonymous object for something a researcher is wearing, an unusual or brightly colored item nobody could have anticipated in advance, and see whether that gets picked up instead.

If a future volunteer describes that hidden detail, in a properly blinded, prearranged test, it would count for more than anything a closed door aside can currently claim, and the number that stands for now is 13 trials, no signal, and a power calculation that leaves the door open rather than closing it. Weiler puts it this way: “This research is not only asking whether out-of-body experiences are real. It is asking what we mean by real in the first place, and whether our current understanding of reality is broad enough to account for everything human consciousness can experience.”

FAQ