Google Effect: Why People Forget What They Can Look Up

The Google effect is a trade, not a loss: effort that would have gone into holding a fact goes into holding a pointer to it. Daniel Wegner had described the same arrangement between people long before search engines, under the name transactive memory. In a couple who have been together for decades, one of them keeps the birthdays and the other keeps the car history, and each knows who to ask. A search engine is that partner with no limits and no bad moods.

Two results in the 2011 paper make the trade visible. When people expected the information to be available later, they recalled less of it. And memory for what had become of a statement ran the other way: participants were more accurate about which statements had been saved than about which had been erased, while their memory for the statements themselves was best for the erased ones. Memory followed what would be useful: the content when no other copy exists, the fact of the copy when one does.

Treated as a strategy rather than a defect, this is what Evan Risko and Sam Gilbert called cognitive offloading in 2016: using the body and the world to reduce what the head has to hold, from writing a list to tilting your head to read a rotated picture. Offloading is usually adaptive. It becomes costly in the narrow cases where the external copy fails or cannot be consulted: an exam, an outage, a conversation, an emergency, or any skill that has to be fluent rather than looked up.

The caveats matter on this page more than on most. Sparrow and colleagues' first experiment, in which hard trivia questions slowed people down when naming the color of computer-related words, failed to replicate twice in the large replication project reported by Camerer and colleagues in 2018, and again in a closer replication by Guido Hesselmann in 2020. The saved-and-erased result in the second experiment has not been through a comparable large replication attempt, and it rested on 60 participants in four groups. There is also a related finding that cuts the other way: Storm, Stone and Benjamin reported in 2017 that people who had used the internet to answer one set of questions were more likely to reach for it again on the next set, even for questions they could have answered from memory. The popular version of all this, that search engines are eroding memory itself, is not what these experiments measured.

Betsy Sparrow, Jenny Liu and Daniel Wegner published "Google effects on memory: Cognitive consequences of having information at our fingertips" in Science in 2011. Four experiments were reported; the second and third are the ones that gave the effect its name in the popular press.

In the second experiment, participants read 40 memorable trivia statements of the kind people look up, such as "an ostrich's eye is bigger than its brain," and typed each one into a computer. Half were told the computer would save what they typed and half that each entry would be erased. Within each of those groups, half were also asked explicitly to try to remember the statements. Afterward everyone wrote down as many statements as they could.

Only the saved-or-erased manipulation mattered. Participants who believed their typing was being erased recalled 31 percent of the statements without a memory instruction and 29 percent with one; participants who believed the computer was keeping a copy recalled 22 percent and 19 percent. The instruction to remember had no significant effect, which is a familiar result in the learning literature.

The third experiment asked where the information had gone. Thirty statements were typed, a third marked as saved generically, a third saved into one of six blandly named folders, and a third erased. On a later recognition test, accuracy for the statements themselves was highest for the erased third, at 93 percent, against 88 percent for the generically saved and 85 percent for the folder-saved. The pattern reversed for remembering what had happened to a statement: people were more accurate about what had been saved, at 61 and 66 percent, than about what had been erased, at 51 percent. Believing a copy exists shifts memory from the thing to its address.

The trade shows up wherever a reliable external copy exists. The rows below are illustrations, not findings from the studies cited on this page.

The aim is not to stop offloading, which is usually sensible, but to choose what gets offloaded.

• Decide in advance which things have to live in your head: the material for an exam, the first steps of first aid, the shape of a talk, the numbers you would need in an emergency. Everything else can have an address instead.
• Test yourself rather than reread. Closing the source and retrieving the answer is what the research on the testing effect supports, and it feels less productive than rereading while working better.
• Write notes in your own words. Copying moves text; rewriting forces the content through your own understanding, and leaves more behind.
• Make the address good. If the external copy is the plan, then naming, tagging and filing are part of the memory system, and the third experiment found memory shifting toward the copy's existence whenever one was expected.
• Keep one no-device rehearsal of anything that has to be fluent under pressure. Looking something up takes time you may not have.
• Notice the pull. Storm and colleagues found that using the internet to answer one question made people more likely to use it for the next one, so the habit builds on itself; a short pause to try recall first is the counterweight.

Several memory findings sit near this one and are often blended with it.

• It is not absent-mindedness, the failure to encode something because attention was somewhere else. In the Google effect the material was read and typed; what changed was the expectation of future access.
• It is not source confusion, remembering a fact but misremembering where it came from. The third experiment ran the other way: participants remembered whether a statement had been saved more accurately than whether one had been erased, while the erased statements themselves were recognized best.
• It is not the generation effect, the better memory for material you produce yourself rather than read. Generation is about how the material enters; the Google effect is about what you expect to be available afterward.
• It is not a diagnosis. "Digital amnesia" is a marketing phrase, not a condition, and nothing in these experiments speaks to memory disorders.