Does Google Make You Forget Things?

You cannot remember your best friend's phone number. You have looked up the same fact four times this month. Somewhere you picked up the idea that this is your phone hollowing out your memory, and that the science has proved it.

The science has done something more interesting than that, and two of the most repeated pieces of evidence for the scary version do not survive contact with the replication record.

The study everyone cites

In 2011, Betsy Sparrow, Jenny Liu and Daniel Wegner published a paper in Science called "Google Effects on Memory". It is the origin of essentially every article on this subject, including most of the ones ranking above this page.

It reported four experiments on roughly 189 American university students in total. Two findings travelled:

• After being asked hard trivia questions, people were slower to name the colour of computer-related words, which the authors read as evidence that a difficult question makes you think about computers.
• People who were told a fact had been saved to the computer remembered it less well than people told it had been erased.

Two things about the paper are worth knowing before anything else. The cells were small, around fifteen people each in the save-versus-erase experiment. And the paper reports no effect sizes at all: no confidence intervals, no measure of how big any of this was. That was normal in 2011 and it is a problem now.

The authors' own conclusion was also not the alarming one. They described an adaptive shift in what people store, not damage.

What happened when people tried to repeat it

In 2018, a large collaboration attempted to replicate 21 social-science studies published in Nature and Science, using samples about five times larger than the originals. Sparrow's first experiment was one of them.

It failed on every measure the project used. The replication effect was not distinguishable from zero, it fell below the predicted range, it was significantly smaller than the smallest effect the original had power to detect, and the Bayesian analysis pointed towards the null. Across the eight studies in that project that failed, the average effect came out at essentially nothing.

An independent pre-registered replication published in PeerJ in 2020 tried the same experiment again with 89 analysed participants and found nothing, with the difference running numerically in the opposite direction. Its Bayes factor gave moderate evidence in favour of there being no effect at all.

One point of fairness, because this matters. The replication project reported that Sparrow's was the one study in its set designed without the original authors' collaboration, because they did not respond to queries, and the replicators said this made the design harder to match.

Still: the idea that a hard question automatically primes the word "computer" in your head is not supported.

The half that survived, and what it actually shows

The save-versus-erase finding has held up better, and the work that followed it is more precise about why.

A pair of pre-registered experiments published in Cognition in 2022, with 90 and 110 participants, had people type word lists into a computer. For the first three rounds the saved list really was available at the test, which taught people the store could be trusted. On the fourth round it was not, but only half the participants were told so.

The people who still expected the file to be there recalled far less: 24 per cent against 47 per cent in the first experiment, and 26 against 43 in the second. These are large effects by the standards of memory research.

The mechanism is the useful part. People expecting the store spent less time studying and were far less likely to report using any memory strategy at all. Change in study time accounted for around a quarter of the effect and strategy use for around 40 per cent, and a direct effect remained after both, so the authors conclude it is multiply determined.

Read that carefully, because it is not "the machine drains your memory". It is that when you know something is saved, you stop trying to learn it. The trust has to be earned first, too. The effect depends on the store having proved reliable.

Saving something can improve your memory for the next thing

Here is the finding that almost never makes it into the coverage, and it comes from the same laboratory tradition.

In 2015, Storm and Stone showed people a list of words, had them either save the file or delete it, then gave them a second list. People remembered the second list better when they had saved the first one, at effect sizes between roughly 0.7 and 0.9. Saving the first list freed up capacity for the next.

That result is not a quirk of computers. A direct conceptual replication with physical paper, in which fifty participants either filed a sheet or fed it into a shredder, found the same thing: about 18 per cent more words recalled from the second list after filing than after shredding.

So offloading reallocates memory rather than subtracting it. Both halves of that sentence have experimental support, and only one half gets reported.

"Digital amnesia" was coined by an antivirus company

The term itself is not a clinical finding. It comes from a 2015 report published by Kaspersky Lab, which commissioned a market research firm to run an online survey of 1,000 US consumers in May 2015.

That is the entire evidence base for the phrase. An opt-in online panel, no published question wording, no margin of error, no raw data, and a methodology section four sentences long. The report's own conclusion is a sales pitch: that people underestimate how exposed their externally stored memories are and should protect them with antivirus software. Its stated sample is 1,000 people, while one of its own charts gives a base of 1,054, and the discrepancy is not explained.

The most quoted number from it, that 91 per cent of people have digital amnesia, is a misreading of a different question. What 91.2 per cent of respondents agreed to was that they use the internet as an online extension of their brain. That is a statement about behaviour, not a measurement of memory.

The best detail is that the report quotes two academics who partly disagree with its own framing. One notes that offloading could make it easier to recall what you have retained. The other says forgetting is in no way a bad thing, quite the contrary.

The camera effect is real, and not for the reason you think

Photographing something does seem to impair your memory of it. The original 2014 museum study walked participants round an art gallery, had them observe some objects and photograph others, and found worse memory for the photographed ones. Its abstract also contains a detail that rarely gets repeated: when participants zoomed in on a specific part of an object, their memory was not impaired.

The explanation everyone reaches for is offloading. You took the photo, so your brain filed the job as done. A 2018 study tested that directly by adding a Snapchat condition, where the photograph is not retained and there is nothing to offload to.

Memory was impaired more in the Snapchat condition than with an ordinary camera. A second experiment had participants take a photo and immediately delete it, and the impairment appeared again. Whatever is happening, it is not that your brain outsources the memory to the file, because the effect is largest when there is no file.

The authors' alternative is more mundane and more useful: the act of taking the photo pulls your attention away at the moment you would otherwise be encoding. Consistent with that, other researchers have found the effect disappears when a wearable camera takes the pictures automatically, and separate work suggests that choosing to photograph an experience can actually improve memory for what you saw while impairing memory for what you heard.

Taking a picture is not deleting the memory. It is spending the moment differently.

Where the worry looks most justified

If there is a version of this claim with real-world teeth, it is navigation.

A study of fifty regular drivers in Montreal found that greater lifetime GPS use was associated with worse performance on spatial memory tasks, with correlations in the region of minus 0.20 to minus 0.33. Heavier GPS users also noticed and used fewer landmarks.

The clever part is the check the authors ran on the obvious objection. If people with a poor sense of direction simply use GPS more, then GPS use should track a poor subjective sense of direction. It did not. That is indirect evidence that the causal arrow points from GPS towards spatial memory rather than the other way round.

The unavoidable part is the size of the follow-up. Thirteen of the original fifty were retested about three years later, and while the correlations there were larger, thirteen people with confidence intervals running most of the width of the scale is not a strong foundation. The study is correlational and nobody was assigned to use GPS.

The stronger design in this area runs in the other direction. London taxi drivers studying the city's layout were scanned before and after, and compared both to trainees who failed the exam and to non-drivers. Grey matter in the back of the hippocampus grew only in those who qualified. And there was a price: the same qualified drivers performed worse than controls on a different memory task.

Building an internal map changes the brain and costs something else. That is a more honest summary of this whole subject than anything about amnesia.

What the evidence actually says to do

Three things are worth acting on, and one popular piece of advice should be dropped.

1. Test yourself instead of rereading. This is the best-supported finding in the entire field of learning. A meta-analysis of 159 effect sizes found retrieval practice beats restudying at a pooled effect of about 0.50, rising to 0.69 when the test is at least a day later and 0.73 when feedback is given. The important caveat is at the bottom end: if you cannot retrieve the answer and nobody corrects you, the benefit is essentially zero.
2. Space it out, and scale the gap to how long you need it. A synthesis of 839 assessments across 317 experiments found that the interval producing the best retention grows as the retention interval grows. Cramming works for tomorrow and fails for next month.
3. Decide deliberately what you want in your head. The Cognition experiments above are clear that expecting a store suppresses effort. If something genuinely matters, the useful move is to treat it as unsaved: look away, try to recall it, and only then check.

The advice to drop: writing notes by hand rather than typing them. The famous 2014 study is misreported almost everywhere. In its first experiment the difference on factual recall between longhand and laptop came out at p equals 0.91, which is no difference whatsoever. The large effect people quote comes from a later experiment and is not longhand versus laptop at all: it is longhand plus ten minutes of study time against everything else.

Two larger replications have since failed to find the learning advantage, including one with 142 participants published in Psychological Science in 2021, whose mini meta-analysis across similar studies found the same non-significant pattern. What does replicate reliably is that typists transcribe more words and more verbatim. What does not replicate is that this costs them.

The honest version

Your phone changes what you encode. You keep the pointer rather than the fact, and when you expect something to be saved you put less effort into learning it. That effect is real, it is sizeable in the lab, and it is worth knowing about.

It is a reallocation rather than an erosion. The same experiments that show saving a file costs you memory for its contents show it buys you memory for what comes next. The single most cited demonstration of the Google effect failed replication twice. The scary brand name came from an antivirus company's marketing survey. And in the one place the effect looks genuinely worrying outside the lab, navigation, the evidence is correlational with a follow-up of thirteen people.

If your memory has changed in a way that worries you, particularly if it has changed noticeably over months rather than years, that is a conversation for a doctor rather than a reason to delete an app. Nothing here is medical advice. If what you actually want is to hold more of what you read, getting your attention span back is the more useful place to start than blaming the search box.