The Photo-Taking Impairment Effect: They Deleted the Photos and It Happened Anyway

There is a sentence at the bottom of almost every article ever written about photographing your holiday, and it comes from an interview rather than a study.

“As soon as you hit click on that camera, it’s as if you’ve outsourced your memory.”

That is Linda Henkel, whose 2014 experiment started this field, speaking to NPR. It is a vivid explanation, it is the reason everyone remembers the finding, and her paper contains no test of it whatsoever. There was no deletion condition, no manipulation of whether the photo was retrievable, nothing that could distinguish outsourcing from any other mechanism.

When researchers did test it - twice, in two independent laboratories - it failed.

The original finding

Henkel (2014), in Psychological Science, took undergraduates through the Bellarmine Museum of Art.

Experiment 1: 27 participants. Thirty objects, fifteen photographed and fifteen merely observed, twenty seconds of viewing each.

Measure	Photographed	Observed	Statistic
Name recognition	.88	.91	F(1, 26) = 4.06, p = .05, η2 = .16
Visual details	.55	.64	F(1, 26) = 10.95, p = .003, η2 = .42
Free recall	M = .47, SE = .04	M = .48, SE = .03	F(1, 26) = 0.09, p = .77

Note where the strength is. The details effect is solid at p = .003. The recognition effect - the one that generated the headlines - sits exactly at p = .05 with a small effect size in a sample of 27. Free recall was flatly null.

Any sentence claiming people could not recognise what they had photographed is leaning on the weakest number in the paper.

Experiment 2: 46 participants, adding a condition in which participants zoomed in on one part of an object. And this is the result that should have shaped everything afterwards:

Verbatim: “When objects were photographed by zooming in on one part, memory for details was not harmed and was, in fact, comparable to memory for details when objects were observed.” And: “zooming in on one aspect of an object served to protect memory not only for the part that was zoomed in on but also for the part that was not zoomed in on.”

Zooming in did not create a trade-off. It abolished the impairment entirely, including for the part never zoomed in on. The camera was still there. The photo was still saved. The impairment disappeared as soon as the photography demanded attention.

The deletion experiments

Soares and Storm (2018), in the Journal of Applied Research in Memory and Cognition, went straight at the outsourcing hypothesis.

Experiment 1, N = 42. Three conditions: observe with the phone face-down; photograph and save to the album; or photograph and send via Snapchat, having been told the image “would not be saved on the phone (or anywhere else) for participants to access later.”

Main effect F(2, 82) = 15.35, p < .001, η_p² = .27. Observe versus Camera t(41) = 3.12, p = .003, d = 0.70. Observe versus Snapchat t(41) = 5.78, p < .001, d = 1.21. And Camera versus Snapchat t(41) = 2.23, p = .031, d = .48 - the unsaveable photo was worse than the saved one.

Experiment 2, N = 51, made it explicit: participants photographed each painting and immediately deleted the photo before continuing.

Main effect F(2, 100) = 4.26, p = .017, η_p² = .08. Observe versus Camera t(50) = 2.30, p = .026, d = 0.58. Observe versus Delete t(50) = 2.92, p = .005, d = 0.61. Camera versus Delete t(50) = .25, p = .80 - no difference at all.

The authors’ conclusion, verbatim: “These results are inconsistent with the offloading hypothesis. If taking photos causes people to forget because they think of the camera as a prosthetic memory device onto which they can offload memory, then making the camera less reliable (or entirely unreliable in the case of the Snapchat and Delete conditions) should have eliminated or significantly reduced the extent to which memory was impaired. Instead, memory was impaired by photo-taking regardless of whether participants thought the photos would be saved.”

A completely separate research group reached the same place independently. Barasch and colleagues (2017) included a photos-deleted condition in a study of 751 participants and found their effects fully intact in it.

Two labs, different designs, different outcome measures, same result: deleting the photo changes nothing. That is the strongest single fact in this literature, and it kills the explanation that nearly all popular coverage still runs on.

The result that looks like a contradiction and is not

Barasch, Diehl, Silverman and Zauberman (2017), also in Psychological Science, ran four studies - one field, three laboratory - and found the opposite of Henkel on visual memory.

Study	N	Visual recognition (camera vs none)	Auditory recognition
1 (museum, field)	297	66.51% vs 59.83%, F(1,292) = 12.61, p < .001	75.34% vs 77.93%, p = .181
2 (lab, gallery video)	298	89.33% vs 76.69%, F(1,296) = 26.93, p < .001	48.97% vs 56.66%, p = .002
3 (saved / deleted / none)	751	saved 84.22%, deleted 85.71%, none 77.51%	saved 50.30%, deleted 51.44%, none 56.56%
4 (camera / mental photo / none)	349	camera 90.27%, mental 88.03%, none 80.79%	camera 52.54%, mental 54.81%, none 57.77%

Their internal analysis across the four studies: photo-taking had “a reliable positive effect on visual memory (d = 0.54, 95% CI = [0.45, 0.63])” and “a smaller but reliable negative effect on auditory memory (d = 0.26, 95% CI = [0.16, 0.36]).”

This is not a contradiction of Henkel. It is a different experiment.

Assignment versus volition. Henkel told participants which objects to photograph. Barasch let them choose. Within the camera condition, visual memory was better for aspects participants chose to photograph than for aspects they did not: d = 0.33, 95% CI [0.27, 0.40]. Choosing a shot is itself an act of encoding.

Modality. Barasch does not deny a cost; it locates one. Visual memory up, auditory memory down. Henkel never tested auditory memory.

And Study 4 settles the mechanism question from the other side. Participants asked to take a mental photograph - no device at all - performed indistinguishably from those using a camera (camera vs mental, F(1,346) = 0.89, p = .35), and both beat no-camera on visual memory. Whatever is happening does not require a device.

What is actually going on: unresolved

With offloading excluded, the field’s default explanation became attentional disengagement - the idea that operating the camera takes attention away from the object. That account is now failing its own tests.

Lurie, Fabrizio and Westerman (2025), in Memory & Cognition, report that “the magnitude of the photo impairment effect did not increase with task complexity, undermining the idea that divided attention is the primary cause.” Their sharpest result runs the wrong way for the theory: screenshotting - the least effortful capture method available - produced the largest deficit, 54.02% against 76.07% for viewing, F(2,38) = 25.22, p < .001, η_p² = .40, d = 1.00.

Croes and colleagues, in a preprint reporting two preregistered experiments, found that “increasing attentional demands did not produce greater memory impairment, contrary to the attentional disengagement hypothesis,” and - the more damaging finding - that “when participants were given additional viewing time after photographing the paintings, the photo-taking impairment effect disappeared.”

If extra viewing time abolishes it, part of what has been called a memory effect may be an exposure-time confound.

And a preregistered null

Braun, Croes, Vaitkunaite and Verheyen (2024), in the Proceedings of the Annual Meeting of the Cognitive Science Society, ran a preregistered within-subjects study with 107 participants across Observe, Camera and Delete conditions.

The preregistered analysis: F(2, 212) = .112, p = .894, η_p² = .002. Means of 5.70, 5.79 and 5.79. All pairwise comparisons at p = 1.000. An exploratory analysis on the full sample of 161 gave F(2, 320) = .704, p = .495.

No photo-taking impairment effect of any kind. The authors’ reading: “It thus seems that photo taking has a detrimental effect on memory for the photographed object only if the process of taking a photograph is distracting.”

Because no baseline effect emerged, their delete condition says nothing about deletion. It is a boundary-condition null, and it should be read as one.

The effect is also context-dependent in ways nobody predicted. Work from Storm’s own laboratory found that photographing lecture slides helped memory for the slide content, and in one experiment helped memory for spoken-only content too. A 2025 study found impairment in two in-person experiments and a photo-taking benefit in an online one.

What this leaves you with

The camera does not take the memory away. The moment of taking the picture borrows the attention. That is the most defensible one-line summary of the evidence.

Deleting the photo does not help. Two independent laboratories tested this. Neither found the offloading account survived it.

Choosing what to shoot appears to help rather than hurt visual memory, at some cost to what you hear.

Zooming in abolished the impairment in the original study - including for the part never zoomed in on.

Whatever the mechanism is, it is not settled. Offloading is largely excluded, divided attention now fails two direct tests, one preregistered study found no effect at all, and screenshots - the laziest capture there is - cost the most.

For what happens when you deliberately hand memory to a device, see digital amnesia. For a memory intervention that did not survive replication, see Sans Forgetica. For the wider mechanism, see the cognitive offloading guide.

Limitations of this article

There is no meta-analysis of this literature. The only pooled estimates are Barasch and colleagues’ internal analysis of their own four studies, which is not independent evidence and must never be described as “a meta-analysis found.”

Henkel prints classical η², not partial η². We reproduce the values as printed. They should not be recomputed from F and degrees of freedom, which would produce different numbers.

Widely reported coverage of the original study gives its first sample as 28 participants. The paper says 27. We print 27.

We could not verify the per-experiment sample sizes of Soares and Storm’s 2022 follow-up, and one supplementary row in the Barasch paper returned an internally inconsistent value on extraction. Neither is printed. See the withheld-figures record.

The Croes preregistered experiments are a preprint, not yet peer-reviewed, and we label them as such.

Funding note. The Barasch paper was supported by a Marketing Science Institute research grant, an industry-funded marketing research consortium, and all four authors are based in business schools. It is also the most transparent paper in the set, carrying an Open Data badge with materials deposited publicly. Both facts belong in the record.