Accessibility is supposed to be unambiguously good — the more available knowledge is, the more knowledge people will have. This seems obvious. It is also, under certain conditions, wrong. There is a specific relationship between the effort required to find an answer and the depth at which that answer is retained and understood. When the effort disappears, something valuable disappears with it — and what disappears is not always visible in the moment of finding the answer, only later, when you discover it was never really yours.
There is a phenomenon I noticed in myself long before I found the research that described it, and it goes like this: I do not try to remember the things I know I can look up.
It is not a conscious decision. It is more like a background policy that operates below deliberate thought. When I encounter a name, a date, a formula, a technical term — anything that falls into the category of things-that-can-be-retrieved-later — my memory simply does not engage in the same way it would if retrieval later were not available. The information passes through without sticking, because the sticking feels unnecessary. Why work to retain something that will be waiting for me whenever I need it?
I became aware of this mechanism clearly during a period when my internet access was unreliable — travelling somewhere with intermittent connectivity — and I found, repeatedly, that things I was certain I knew how to look up turned out to be things I did not know at all. The information was not stored; it had been offloaded to the expectation of access. And the expectation of access, it turned out, was doing cognitive work that I had not noticed was being done — specifically, the work of deciding that retention was unnecessary.
The Google Effect
In 2011, a team led by Betsy Sparrow at Columbia University published a study in Science that gave a name to what I had been experiencing and documented it with experimental precision. The study, which Sparrow and her colleagues called an investigation of "the Google effect," found that people are less likely to remember information if they believe the information will be available to them via a computer later — and more likely to remember the location of the information than its content.
In one of the study's experiments, participants read a series of trivia statements and were told either that the statements would be saved on a computer or that they would be erased. Those who believed the statements would be saved showed substantially worse recall of the statements themselves. In another experiment, participants remembered the name of the folder in which information was stored better than they remembered the information itself. The cognitive system, it appeared, was treating the computer's memory as a reliable external store and adjusting its own encoding effort accordingly — remembering where to find things rather than what those things are.
This is a rational strategy in one sense: if retrieval from an external source is reliable and cheap, the cognitive cost of encoding information internally can be redirected elsewhere. We have always used external memory — books, notes, other people — and there is nothing inherently wrong with the practice. What is new is the reliability, the comprehensiveness, and above all the ease of the external retrieval. What was once a significant effort — finding a library, locating a reference, asking an expert — has become, for a very large and expanding category of information, a matter of seconds.
And the ease of retrieval, it turns out, is not separable from the depth of encoding.
The Effort-Encoding Relationship
The relationship between retrieval effort and learning depth is one of the most consistent findings in cognitive psychology, and it runs directly against the intuition that ease of access should be educationally beneficial.
The generation effect, documented by Slamecka and Graf in 1978 and replicated extensively since, shows that information generated by the learner is remembered better than information passively received. The testing effect — sometimes called retrieval practice — shows that attempting to retrieve information from memory, even unsuccessfully, produces better long-term retention than re-reading the same information. The spacing effect shows that distributed practice, which allows some forgetting to occur between study sessions, produces better retention than massed practice, precisely because the partially forgotten information requires more effortful retrieval.
The common thread across all of these findings is that the effort involved in engaging with information — generating it, retrieving it, relearning it after partial forgetting — is not an obstacle to learning. It is the mechanism by which learning occurs. The neural encoding that produces durable memory is strengthened by the cognitive operations associated with effortful processing and weakened by the passivity associated with frictionless access.
When the answer is one click away, the effort of finding it is eliminated. And with the effort goes the encoding. The answer is received; it is understood, in the sense that it makes sense when encountered; it is used for whatever purpose prompted the search. And then it is gone — not because memory is unreliable, but because memory is a system that encodes in proportion to the effort of engagement, and the engagement, in this case, required almost none.
Knowing and Knowing Where to Find
There is a distinction that the easy availability of answers has made practically important but philosophically murky: the distinction between knowing something and knowing where to find it.
For most of human history, the distinction was relatively clear, because access to information was effortful enough that the practical difference between the two was large. If you knew a fact, you could deploy it immediately, in context, in combination with other things you knew, to address a problem or answer a question in real time. If you only knew where to find it, you faced a retrieval process that took time, required access to the source, and was not always available at the moment of need. The person who knew was more capable than the person who knew where to find; the gap between the two was practically significant.
As retrieval has become faster and more reliable, this gap has narrowed to the point where many people treat the two as equivalent. I know how to calculate X, in the practical sense that I can produce the calculation reliably whenever I need it — because I can look it up in thirty seconds. I know who said Y, in the practical sense that I can provide the attribution whenever asked — because I can search for it in five. The equivalence seems reasonable: if I can reliably access information, the fact that it is not stored internally feels like a mere technical detail.
But the equivalence breaks down under conditions that are more common than they might initially appear: when the answer is needed in the middle of a thought process that cannot be interrupted, when the connection between the retrieved information and other information in memory needs to happen fluidly rather than by switching contexts, when the question is not what the information is but what it implies, connects to, or changes about related things you know.
Understanding — as distinct from information access — is constituted by exactly these kinds of fluid connection. Understanding a concept is not merely having access to its definition; it is having the definition in a state where it can interact spontaneously with related concepts, generate applications, notice inconsistencies, and produce the associative movements that constitute thinking. Information that is stored externally and retrieved only when explicitly needed cannot do this. It is available but not active. It is accessible but not integrated.
Transactive Memory and Its Limits
The psychologist Daniel Wegner introduced the concept of transactive memory to describe how couples, teams, and other close-knit groups distribute memory across multiple individuals — each member of the group storing different information, knowing that other members store complementary information, and collectively possessing a memory system more capable than any individual's. The concept is elegant and describes something real: human memory has always been partly social, extending into the knowledge of people around us.
Sparrow and her colleagues extended this concept to suggest that we now treat computers and the internet as components of our transactive memory system — that we offload information to digital storage in the same way we might rely on a partner who knows things we do not. On this account, the Google effect is not a pathology but an adaptation: a rational extension of a fundamentally social cognitive practice into the digital domain.
The extension is plausible, but there is a disanalogy worth noting. Transactive memory in social groups works because each member of the group is a cognitive agent — someone who not only stores information but processes it, draws inferences from it, combines it with other information, and can be asked to do something with it rather than simply to retrieve it. When I rely on my colleague who knows the regulatory details, I am relying on her to tell me not just what the rules say but what they mean for the specific problem we face — to interpret, apply, and reason from the stored information.
When I rely on a search engine, I am relying on retrieval, not processing. The retrieved information arrives as text that I must then process — must read, evaluate, integrate, and apply. The cognitive work that my colleague would do in the act of remembering and communicating the information falls back to me when the source is external storage. Which means that my processing capacity, relative to the stored information, becomes more important, not less — and that the depletion of internal knowledge that accompanies heavy reliance on external storage degrades precisely the processing capacity that the external storage requires me to supply.
The Depth Illusion
There is a particular cognitive experience that instant access enables that is worth describing precisely, because it masquerades as something more valuable than it is.
When you search for information you do not know and find it quickly, there is a moment of recognition — the moment when the information arrives and makes sense, when the question is resolved and the answer seems clear. This moment feels like learning. It has some of the phenomenology of learning: the resolution of uncertainty, the sense of the pieces coming together, the brief satisfaction of comprehension.
What it typically lacks is the depth that marks genuine learning as distinct from information receipt. Depth, in this context, is not a matter of how much you know about a topic but of how the knowledge is structured — how many connections it has to other things you know, how many contexts it can be applied in, how available it is for spontaneous deployment in situations where its relevance is not immediately obvious. Shallow knowledge answers the specific question it was retrieved to answer and does little else. Deep knowledge operates in the background of cognition, connecting to new information automatically, suggesting analogies, generating transfer to unfamiliar problems.
The depth illusion is the experience of having retrieved information and processed it sufficiently to feel comprehension, while actually having acquired only shallow familiarity that will not survive the specific context in which the retrieval occurred. The information was found; it was understood in the moment of finding; it will not be available an hour later without another search. It produced the feeling of learning without the durability that learning is supposed to produce.
The illusion is maintained by the availability of instant re-retrieval. Because the information is always findable, the shallow encoding that the easy finding produces is never directly confronted. The learner never reaches for the information and finds it absent — which is the diagnostic experience that would reveal the shallowness. They reach for it, find it immediately, and interpret the success of retrieval as confirmation of knowledge. What is confirmed is only that the retrieval system is working. The question of what is in the learner's head, independent of the retrieval system, is never tested.
What Friction Actually Does
The friction in finding information — the effort of retrieval, the time spent searching, the uncertainty of whether the right source will be found — is almost universally described as a problem to be eliminated. Faster search, better organisation, smarter retrieval: these are the improvements that information technology has consistently pursued, and the improvements are real. Less friction in access to information is, in most practical contexts, strictly better.
In the specific context of learning, this is not true. Friction in retrieval is a condition of the effortful engagement that produces encoding. Eliminating it is eliminating the mechanism. This does not mean that learning requires inaccessible information; it means that learning requires, at some stage, the deliberate introduction of retrieval effort even when retrieval without effort is available.
This is what spaced repetition systems are doing when they schedule review at the point of near-forgetting: they are deliberately reintroducing friction into a retrieval process that has become too smooth. The information is stored, but it has been stored in a way that makes retrieval effortful — and the effort of retrieval is the mechanism by which storage is strengthened for subsequent retrieval.
The person who never lets themselves forget because the answer is always one click away is the person who has eliminated the effortful retrieval that would have made the knowledge genuinely theirs. They have optimised for availability and sacrificed depth. The information is accessible, always. It is also, in a meaningful sense, never learned.
A Practice Worth Considering
There is a practice that cognitive scientists call the pre-question — the deliberate articulation of what you think you know about a topic before consulting any source. Before searching, you write down: what do I think the answer is? What do I know that is relevant to this question? What would I predict, based on what I already understand?
This practice is not about being right before consulting the source. It is about creating the conditions for effortful processing before the effortless retrieval. The pre-question activates prior knowledge, identifies gaps, generates predictions that can be confirmed or disconfirmed — all cognitive operations that the direct search eliminates. When the information arrives, it arrives into an active cognitive context rather than a passive one, and the encoding that results is correspondingly deeper.
I have used this practice unevenly and imperfectly, and I can report that it is consistently more effortful and consistently more educational than the alternative. It is also consistently resistant to habit, because the habit of instant search is established and the practice of pre-questioning adds a step that produces no immediate reward. The immediate reward — the answer — is available without the pre-question. The delayed reward — the retention and integration of the answer — is not available without it, or not as available. And delayed rewards are, as every behavioural researcher knows, extremely difficult to choose over immediate ones, even when the delayed reward is larger.
What This Means Now
The environment in which we learn has changed faster than the cognitive systems that do the learning. Those systems were shaped over a very long period in which information was scarce and its retrieval was effortful — in which the effort to find an answer was, by default, also the effort that encoded it. The adaptation was calibrated to that environment. In the current environment, where answers are instant and abundant, the calibration is wrong.
This is not a counsel for artificial scarcity or deliberate ignorance. It is a counsel for recognising that the availability of an answer does not substitute for the work of learning it — and that choosing to do the work, in a context where it is no longer required by the environment, is now a deliberate decision rather than an inevitable consequence of the search.
Part of what makes this difficult is that the decision to engage effortfully rather than retrieving instantly produces no visible short-term reward. The person who works to retain something that they could have looked up is slower, less efficient in the moment, and produces no output that distinguishes them from the person who simply looked it up. The difference appears only later, when the person who struggled to retain the information can use it spontaneously — in conversation, in an unfamiliar problem, in a context where retrieval from a device is not available or appropriate — and the person who relied on instant retrieval cannot.
Later is hard. Immediate is easy. And the design of every information environment we currently inhabit is optimised for immediate, which means the decision to work for later requires swimming against a current that is very strong and very invisible, because the current feels like helpfulness.
The click is available. The question is whether to take it, and what you lose if you always do.
Every tool that makes information easier to find makes the work of genuinely learning that information more voluntary. The ease of access does not reduce the cost of learning; it reduces the cost of not learning. What you do with the difference is a choice that the tool cannot make for you — and that most tools are not designed to help you make well.