한국어

The Google Effect on Memory: Do We Forget What We Look Up?

At a glance
  • None of the studies reviewed here show that search or devices damage memory itself.
  • People who used Google expected to score higher on their own later, but did not actually know more.
  • Handing work to a device cost memory, but knowing a memory test was coming won most of it back.
  • The first experiment of the famous 2011 "Google effect" paper did not hold up in a large replication.
  • Before you search or ask AI, write your own answer first, then recall it the next day with the page closed.

None of the studies reviewed here show that searching or using devices damages memory itself. What they show is narrower. People easily mistake what they found online for what they know, and they keep less of what they hand to a device when they have no plan to remember it. When people tried to answer first, or knew they would need the information later, most of that cost went away.

What is the Google effect on memory?

The Google effect is the finding that people remember less of what they expect to be able to look up again. Researchers also call the broader habit cognitive offloading: handing a mental job to a tool or to the world around you.

The name comes from a 2011 paper in Science by Sparrow, Liu and Wegner. Across four studies, people who expected to have future access to information recalled less of the information itself and recalled better where to find it. The authors described the internet as a kind of external memory that we share. "Digital amnesia" is the popular name for the same worry, though the studies below measure recall in experiments, not any condition or diagnosis.

Did the Google effect replicate?

Only partly. The first experiment of the 2011 paper did not show up again in a much larger replication.

Camerer and colleagues (2018) repeated 21 social science experiments published in Nature and Science between 2010 and 2015, with samples about five times larger than the originals on average. One was Sparrow's first experiment: after hard trivia questions, people were slower to name the color of computer words such as "Google", as if those words were already on their minds. That result did not replicate. Sparrow replied that the word list and procedure differed. In 2020 Guido Hesselmann ran it again with her suggestions built in, and still found no effect.

13 of 21 Across the whole project, 13 studies (62%) showed a significant effect in the same direction, and the effects were on average about half the size of the originals (Camerer et al., 2018).

It matters which part failed. The replication was of the word-timing experiment, not the memory experiments about expecting to look things up later. So neither "the Google effect is fake" nor "search erases memory" is a fair summary. More recent experiments answer the question more directly.

Do people mistake the internet's knowledge for their own?

Yes. People who found answers with Google expected to do better on their own later, even though they did not actually know more.

Adrian Ward (2021) showed this across eight experiments with 1,917 people. In Experiment 3, participants took a trivia quiz either with or without Google. Everyone then saw the correct answers, predicted how many of 10 questions they would get right on a second quiz with no internet, and took that quiz.

The Google group expected to score higher, but their actual scores did not differ

  • First quiz with Google
  • First quiz alone
0246810First quiz with Google · Predicted score · 5.955.95First quiz alone · Predicted score · 4.584.58Predicted scoreFirst quiz with Google · Actual score · 3.733.73First quiz alone · Actual score · 3.173.17Actual scoreNo reliabledifference
Mean number of 10 trivia questions people predicted they would answer with no internet, and how many they actually answered. Only the first quiz was taken with or without Google.Source: Ward (2021), Experiment 3 results
See the numbers
First quiz with GoogleFirst quiz alone
Predicted score5.954.58
Actual score3.733.17

The Google group predicted 5.95 correct; the group that answered alone predicted 4.58. Their actual scores were 3.73 and 3.17, a difference that was not statistically significant. The people who had used Google misjudged themselves by more. In Experiment 1, Google users also rated themselves higher on statements like "I have a better memory than most people."

One simple step removed the illusion. In Experiment 5, people who wrote down their own answer before searching predicted 4.47 correct instead of 5.80. Ward's explanation is that search returns answers faster than memory does, so there is never a moment where you notice you did not know.

We did not find a study that ran this experiment with ChatGPT. If Ward's explanation is right, our reading is that an AI that hands back a finished answer to every question could make the illusion easier, not harder.

Does offloading to a device make you remember less?

It does when you offload with no goal to remember. When people knew they would need the information later, they won most of it back.

Grinschgl and colleagues (2021) gave people a tablet task: copy a layout of 12 object pictures from a model window into a workspace. Only one window could be open at a time, so people could either memorize several items per look or keep reopening the model and lean on the screen. Afterwards, without warning, they rebuilt the layouts from memory.

In Experiment 1, people whose model window opened instantly checked it more often than people who had to wait 2 seconds. They finished faster and remembered less.

Experiment 1Opens instantly2-second wait
Model window openings per layout3.612.31
Seconds per layout49.3355.06
Objects placed correctly (of 12)5.917.07

Experiment 3 went further. One group was forced to offload as much as possible, copying a single object each time they opened the model. The other group chose for themselves. Half of each group was told in advance about the memory test.

Even when forced to offload, people who knew about the test remembered almost as much

  • Did not know about the test
  • Knew about the test
036912Did not know about the test · Chose how much to offload · 4.694.69Knew about the test · Chose how much to offload · 5.185.18Chose how much to offloadDid not know about the test · Forced to offload (one object per look) · 1.841.84Knew about the test · Forced to offload (one object per look) · 4.184.18Forced to offload(one object per look)1.84 upto 4.18
Mean number of objects (of 12) put back in the right place when rebuilding each layout from memory. 172 people took part.Source: Grinschgl, Papenmeier & Meyerhoff (2021), Experiment 3, Table 8
See the numbers
Did not know about the testKnew about the test
Chose how much to offload4.695.18
Forced to offload (one object per look)1.844.18

Forced offloading without warning left people with 1.84 objects in the right place. The same forced offloading with a goal to remember gave 4.18, not reliably different from the groups who chose for themselves (4.69 and 5.18). What cost memory was not leaning on the screen as such, but leaning on it without intending to remember.

Is saving things to a device bad for memory?

No. Saving something you trust the device to keep can help you learn the next thing.

Storm and Stone (2015) had people study two files of words in turn. In three experiments, saving the first file before studying the second improved memory for the second file. The benefit disappeared when people thought the save was unreliable, and when the first file was too small to get in the way.

So you do not need to memorize everything. Put what you do not need right now somewhere you trust, and spend your attention on what you are learning. The cost shows up when you confuse what you stored with what you know, and when you let the things you meant to learn slip by with no plan to remember them.

How do you keep what you look up?

Write your own answer before you search or ask AI, pick what you want to keep, and recall it the next day with the page closed. A sheet of paper is enough.

  1. Answer for ten seconds before you ask. Write one line, even a guess. In Ward's Experiment 5 this one step removed the "I knew that" feeling. Why a wrong first attempt still helps is covered in Is Using ChatGPT to Study Bad?
  2. Split what to remember from what to store. Phone numbers, syntax details and settings can live on the device. For the things you want to own, decide "I will remember this." In Grinschgl's study that goal made the difference.
  3. Store in one place you trust. In Storm and Stone's study, saving only helped when people trusted the save.
  4. Recall it the next day with the page closed. Say the answer out loud before you look. When to come back after that is in Spaced repetition schedule.

In Aloom

In Aloom you do not save every answer you get from AI. You pick the ones to keep, and that choice is the moment you decide "I will remember this." Kept answers join your knowledge graph as nodes, and each day you recall the number you chose, 1, 3 or 5, as a question before you see the answer. Because the answer does not arrive first the way a search result does, you find out whether you really know it. The more you recall a node, the darker its ink. Take a look at Aloom

FAQ

Is digital amnesia real?

Partly. People do remember less of what they hand to a device without meaning to remember it (Grinschgl et al., 2021), and they overrate how much they know after searching (Ward, 2021). The studies reviewed here did not find lasting damage to memory, and the best-known experiment behind the idea did not replicate.

Does AI affect memory the same way as Google?

We did not find an experiment that tested this with an AI assistant directly. The mechanisms in these studies, answers arriving faster than memory and offloading without a goal to remember, apply to AI answers too, so the same habits help: answer first, choose what to keep, and recall it later.

What is cognitive offloading?

It is using a tool or your body to take load off your mind, such as setting a reminder, using a calculator or searching for an answer. It helps you finish tasks faster, but without a goal to remember, less of what you offloaded stays with you (Grinschgl et al., 2021).

References