I can find an answer without knowing whether it is right. Search makes this easy to overlook because the answer arrives with the speed and confidence of something already settled. What I have gained, at first, is access to a claim.

There are occasions when this is enough for the immediate task. I look up a word, check a date, or consult a timetable. Even then, success depends on the source being suitable and current. A train time from last year’s timetable may be correctly copied and still fail me today.

The distinction becomes harder to maintain when a system answers many kinds of questions well. Matthew Fisher and colleagues found that searching online for explanations could inflate people’s estimates of their own knowledge, including on other topics.[^1] The experiments concern judgments of knowledge. They do not establish that every search produces the same effect, but they identify a mistake worth guarding against: confusing information that was available to me with information I had learned.

Being right once is also a weak test of knowing. Edmund Gettier’s examples challenged the idea that justified true belief is always sufficient for knowledge.[^2] A belief can turn out to be true through a fortunate accident even when the reasoning behind it is defective. The philosophical problem is broader than internet use, but access to answers gives us many everyday opportunities to mistake a correct result for a reliable method.

Suppose I repeat a historical date from a page that invented its sources. The date may happen to be correct. I have still acquired it through a process I ought to distrust. If I use the same process for the next claim, my earlier success offers little protection.

Checking a source is a skill that can be practised. Sam Wineburg and Sarah McGrew studied how people evaluated websites. Professional fact-checkers often left the page under inspection to investigate it elsewhere, rather than treating its design and internal claims as sufficient evidence.[^3] That approach is called lateral reading. It helps separate what a source says about itself from what can be established independently.

Generated answers make this work especially important. NIST’s generative AI risk profile describes the production of confidently stated false content among the risks of these systems.[^4] Fluency gives a claim an easy passage into our attention. It does not establish the claim’s truth. A reference needs to exist, contain the relevant information, and actually support the sentence attached to it.

Verification takes time. We cannot investigate every statement equally, so we make judgments about what deserves further checking. An unfamiliar claim that carries the weight of an argument deserves more attention than an incidental detail. A source may be reliable within its field and unsuitable for another question. Confidence should follow the quality of this work.

Our own explanations can provide another check. Leonid Rozenblit and Frank Keil found that people often rated their understanding of mechanisms more highly before attempting a detailed explanation than afterwards.[^5] Familiarity with an object can conceal gaps in our account of how it works. I may recognise an explanation, or enjoy reading it, without being able to reconstruct its reasoning.

When knowledge needs to remain available without a device, the way we study matters. John Dunlosky and colleagues’ review found strong support for practice testing and distributed practice across many learning conditions.[^6] Looking something up repeatedly may help with a recurring task, but it should not be treated as equivalent to every form of learning. We need to choose a method that suits what we expect to do later.

Bacon’s Of Studies connects reading with weighing what is read and writing with greater precision.[^7] The connection is still useful. Writing a claim in my own words forces decisions that reading can postpone: what I mean, what follows from it, and which evidence I am relying on. If I cannot make those decisions, the sentence I found has done more work than I have.

Access has enlarged the range of things we can investigate. It has also made the acts of finding, checking, learning, and judging easier to confuse. Each can support the others. They remain distinct enough that we should be able to explain which one we have performed.