A Search Result Is Not Yet Knowledge

On a Sunday evening, a parent types a deceptively simple question into a search box: “How much screen time is harmful for a nine-year-old?” Within seconds, the browser offers a child psychologist’s interview, two news stories pointing in opposite directions, a parenting forum, a product company’s…

How information becomes a worldview that can still be checked and revised

On a Sunday evening, a parent types a deceptively simple question into a search box: “How much screen time is harmful for a nine-year-old?” Within seconds, the browser offers a child psychologist’s interview, two news stories pointing in opposite directions, a parenting forum, a product company’s blog, a plain-language summary of a study, and several pages that have repackaged the same press release under different headlines. The screen suggests abundance. The parent experiences congestion.

We often treat this moment as if the information-gathering stage were complete and only the final choice remained. That is the first mistake. A results page is not an ordered extract of reality. It is a temporary meeting place for texts produced for different purposes, by different methods, and often from one another. Measurement, testimony, advertising, institutional guidance, opinion and second-hand explanation may all appear in the same visual format. They are not doing the same epistemic job, and they should not be judged as though they were.

The question has already shaped the search. “Why is screen time harmful?” asks for supporting evidence for a role assigned in advance. “What effects have been measured, for which ages, with what content and under what conditions?” leaves more of the field open. The terms overlap, but the two queries enter different informational landscapes. A search engine will not remove the assumption embedded in the first formulation. It may simply serve it with impressive speed.

The search begins before the first result appears

A useful question does not need to sound academic. It needs to reveal what phenomenon, population, time horizon and consequence are actually under discussion. “Screen time” may mean a video call with a grandparent, an endless feed of short clips, collaborative gaming, homework or a cartoon playing in the background. “Harm” may refer to disrupted sleep, reduced movement, attention problems, family conflict, or merely parental discomfort. When these possibilities are packed into one phrase, the results seem to address a single issue while quietly answering several different ones.

Public questions behave in the same way. “Has education deteriorated?” does not tell us whether the speaker means teacher shortages, student performance, social mobility, institutional autonomy or ordinary family experience. Evidence can be found for each. Improvement in one measure does not automatically refute deterioration in another. Vague questions produce vague victories: everyone can find a number to display, and nobody has to state precisely what has been established.

Research on motivated reasoning shows that people do not approach evidence as neutral processors. Desired conclusions influence which arguments we look for, how closely we inspect inconvenient evidence and how soon we feel entitled to stop [1]. Confirmation bias operates at several points: search terms, source selection, interpretation and memory can all favour what already feels plausible [2].

A useful discipline therefore begins with an awkward question: what result would weaken my current view? If no conceivable answer would do so, we may not be investigating a claim. We may be commissioning evidence for a story already written.

Ranking is not adjudication

A search engine is an effective gateway, not a court. It shows what is accessible, what its ranking system considers relevant, what matches the wording of the query and what has been published in a machine-readable form. It does not guarantee that the first result contains the strongest evidence or that the most frequently repeated sentence is the most accurate one.

Ranking nevertheless acquires authority. Most users inspect the first few results; the second page may as well be located in another climate zone. Visual confidence adds to the effect. A polished site, a professional portrait, a graph and an institutional logo can create a strong impression of reliability even when the particular claim under examination rests on little.

AI-generated answers intensify the problem because they replace a visible list of sources with finished prose. Differences in evidence quality, definitions and uncertainty are compressed into a smooth paragraph. The sentence comes to a grammatical end, and readers may feel that the question has ended with it. Fluency demonstrates that the text is well formed. It does not demonstrate that every claim has been checked to the same standard.

The practical question is therefore narrower than “Is this source trustworthy?” It is: trustworthy for what? A ministry statement may be an excellent primary source for what the ministry announced. It is not thereby an independent evaluation of whether the programme worked. A participant’s account may be indispensable for understanding lived experience while remaining insufficient to estimate national prevalence. A peer-reviewed study may be methodologically strong for a narrowly framed causal question and still have limited relevance to a different population or setting.

Sources have functions, not halos.

Credibility is often visible outside the page

When professional fact-checkers assess unfamiliar websites, they often practise lateral reading. Rather than remaining inside the site’s own presentation, they open new tabs to investigate who operates it, where its evidence originated, how independent sources describe it and whether the quoted research can be located [3]. The movement is simple but important. It reverses the direction of attention: the evaluator places the page from the outside instead of allowing the page to define its own context.

In practice, this is usually less dramatic than investigative cinema. A health article says “research shows”, so we follow the reference. The underlying study may concern animals while the article generalises to humans; it may examine a narrow population; or the link may lead to another summary that also fails to identify the original data. The claim is not necessarily false. Its evidential route is simply longer and more fragile than the sentence suggests.

Tracing origin also matters because volume can disguise repetition. Five websites with five logos do not amount to five independent confirmations if all five copied the same wire report, relied on the same press release, or cited one another. In that case, we have one claim displayed through five frames.

Social media strengthens the illusion. Repetition creates familiarity, and familiarity can be mistaken for evidence. Frequency may tell us what is spreading or what many people find plausible. By itself, it does not tell us what happened.

Data quality is not intellectual snobbery

It is tempting to believe that truth will emerge automatically from a sufficiently large dataset. Scale can help reveal rare patterns, expose divergent viewpoints and reduce random noise. Yet in a poorly constituted information environment, bias can become a stable pattern as well.

If outrage receives more attention on a platform, the available data do not simply represent public opinion. They have been shaped by competition for attention, moderation rules, user composition, automated accounts, recommendation systems and the quiet majority that stopped participating. A larger sample may measure the same distorted environment with greater precision.

Quality does not mean that only academic journals deserve attention. It means understanding how an item was produced, what it excludes and what it can support. A forum may be excellent for discovering recurring complaints or the language people use to describe a problem. It is a poor instrument for estimating the prevalence of a medical condition. A large news archive may reveal when a topic entered public discussion and how narratives travelled. It cannot by itself determine which narrative is true.

The idea that “enough data will correct bad data” conceals the question of calibration. Without at least some independently checkable reference points, a system may simply identify the most frequent or densely connected story. That story can be socially important. It is not necessarily the best explanation.

Looking for contrary evidence is not balance theatre

Searching for information that challenges one’s position is a powerful safeguard, but it is easy to perform badly. The aim is not to manufacture a “second side” for every question and allocate attention equally. Evidence does not become strong through democratic representation.

The task is to locate the strongest relevant challenge. A random comment is not the proper counterweight to a well-established body of research. A national average does not erase a person’s documented experience. And where the disagreement concerns values rather than facts, an additional graph will not decide how much risk, inequality or restriction people ought to accept.

This distinction matters in political and moral questions. Two people may accept the same budget figures and still choose differently because they assign different weights to short-term security and long-term autonomy. Clarifying the facts remains worthwhile. It reveals where empirical disagreement ends and value choice begins.

Research on misinformation also shows why correction cannot always consist of a simple negation. A false claim may leave behind a causal story that continues to shape understanding after the specific sentence has been withdrawn [4][5]. Effective correction often requires an alternative explanation that reorganises the event rather than merely placing a red cross beside the original claim.

Attention is also a source

Information costs attention. A manipulative page, an untraceable video or the hundredth recycling of a previously debunked claim can demand half an hour while contributing nothing to the question.

The concept of critical ignoring describes the ability to decide when to leave, block or redirect attention towards better sources [6]. This is not the avoidance of unwelcome ideas. The difference lies in the reason. A claim is set aside when it lacks an identifiable proposition, a traceable origin or any new connection to the issue being investigated; discomfort alone is irrelevant.

There is, of course, an opposite failure. “Not worth attention” can become a convenient shield around a preferred worldview. Refusal therefore needs its own rationale: why the material is irrelevant, what verification point is absent, and under what condition it would deserve reconsideration.

Good inquiry is not limitless openness. It is disciplined movement between curiosity and selection.

Context is not decorative background

A statement often changes meaning when its time, place and actors are restored. “It increased by thirty per cent” may be significant, trivial or misleading depending on what increased, from what baseline, over which period and whether the measurement changed. The number may be exact while the story constructed from it is wrong.

Contextualising a claim does not mean softening it until nothing remains. It means drawing a more accurate boundary around it. We separate what the data directly show, what the author infers, and what additional conditions would be required before the conclusion could travel elsewhere.

Consider a dispute with a service provider. The invoice, the contract, customer-service advice and the customer’s memory are different kinds of information. The invoice records what the company demands. The contract indicates what rule it may invoke. A recording may reveal what was actually said. Memory helps reconstruct why someone acted as they did, while years later it may be weak evidence for exact wording. The case becomes intelligible only when these materials are not compressed into one question: “Who is right?”

Context gives information its relations. Without those relations, details survive but their load-bearing function disappears.

Notes are not bureaucracy when they compensate for memory

In a complex inquiry, it becomes difficult to remember where a claim originated, which sources were independent and which questions remained open. A short note, table or visual map is useful because human memory tends to tell a cleaner story after the fact than the path actually taken; a scientific costume is beside the point.

A few fields may be enough: what the source claims; what it relies on; which part appears strong; what conflicts with it; and what would change the assessment. The format is secondary. Traceability is the point. If a week later we no longer know whether a number came from an original study, a press release or our own recollection, our reasoning has silently merged levels of certainty.

The PRISMA framework was designed for transparent reporting of systematic reviews, not for ordinary web searches [7]. One lesson can nevertheless travel: record where you looked, what you included and what you excluded. This does not turn a personal search into a research publication. It does make selective memory less powerful.

New information does not sit on a shelf; it rearranges the room

Inquiry matters only if the result changes more than the number of saved links. Sometimes a new fact simply adds detail. At other times it forces us to redraw a connection, separate concepts that had been merged, or admit that a favoured explanation works only under narrower conditions.

Formal theories of belief revision examine how a system of propositions can respond to new information while retaining as much of its previous structure as possible [8]. An ordinary worldview is not a logical database. It includes emotion, experience, identity and moral commitment. Yet the difficulty is recognisable: a new fact rarely fills a single empty box. It often changes the weight of several existing claims.

Coherence is necessary but insufficient. A conspiracy theory can be internally organised. It may become especially resistant when every attempted correction is reinterpreted as further proof of suppression. A usable worldview must therefore be connected and repairable. It should be able to distinguish strong elements from provisional ones and leave room for areas that remain unknown.

That costs energy. New information is sometimes uncomfortable because it requires several old statements to be rearranged at once, even when the news itself is neutral. A clearer picture is often purchased with the loss of a simpler story.

Knowing when to stop

Endless searching is not necessarily rigorous; sometimes it is merely endless. Decisions usually have to be made while uncertainty remains. The practical question asks whether the claims that matter for the decision have been examined deeply enough; complete knowledge is an impossible threshold.

A reasonable stopping point usually means that the main terms have been clarified, the strongest available sources have been located and traced, the most relevant challenge has been considered, and the remaining uncertainty can be named. If ten further results repeat the same material under new headings, the marginal value is small. If a decisive statistic has no identifiable origin, or two strong studies disagree for reasons that remain methodologically unclear, closure would be premature.

A good decision does not always emerge from the most confident sentence. Sometimes it begins with an accurate account of the uncertainty being accepted.

Inquiry is therefore neither link collection nor the search for one flawless source. It is the gradual construction of an account of what each piece of information says, how far it says it, and what it changes in the picture already held. The result may not be dramatic. It may be a more cautious sentence, several discarded explanations and a better-named open question.

That can look modest. It is where knowledge begins.

References

[1] Kunda, Ziva (1990): “The Case for Motivated Reasoning.” Psychological Bulletin, 108(3), 480-498. DOI: 10.1037/0033-2909.108.3.480.

[2] Nickerson, Raymond S. (1998): “Confirmation Bias: A Ubiquitous Phenomenon in Many Guises.” Review of General Psychology, 2(2), 175-220. DOI: 10.1037/1089-2680.2.2.175.

[3] Wineburg, Sam; McGrew, Sarah (2019): “Lateral Reading and the Nature of Expertise: Reading Less and Learning More When Evaluating Digital Information.” Teachers College Record, 121(11), 1-40. DOI: 10.1177/016146811912101102.

[4] Lewandowsky, Stephan et al. (2012): “Misinformation and Its Correction: Continued Influence and Successful Debiasing.” Psychological Science in the Public Interest, 13(3), 106-131. DOI: 10.1177/1529100612451018.

[5] Ecker, Ullrich K. H. et al. (2022): “The Psychological Drivers of Misinformation Belief and Its Resistance to Correction.” Nature Reviews Psychology, 1, 13-29. DOI: 10.1038/s44159-021-00006-y.

[6] Kozyreva, Anastasia; Wineburg, Sam; Lewandowsky, Stephan; Hertwig, Ralph (2023): “Critical Ignoring as a Core Competence for Digital Citizens.” Current Directions in Psychological Science, 32(1), 81-88. DOI: 10.1177/09637214221121570.

[7] Page, Matthew J. et al. (2021): “The PRISMA 2020 Statement: An Updated Guideline for Reporting Systematic Reviews.” BMJ, 372:n71. DOI: 10.1136/bmj.n71.

[8] Alchourrón, Carlos E.; Gärdenfors, Peter; Makinson, David (1985): “On the Logic of Theory Change: Partial Meet Contraction and Revision Functions.” Journal of Symbolic Logic, 50(2), 510-530. DOI: 10.2307/2274239.