HSW-0300 · How Studying Works
A learner opens one good article. Three minutes later there are twelve tabs.
Every click made sense. One link defined a term. Another showed a diagram. A third offered background knowledge. Then a comparison article seemed useful. Then a source from that article looked more authoritative.
The learner has found more information and understands the original topic less clearly.
Hypertext navigation creates a dual study job: understand the information and continually decide where the information path should go next. Those jobs can support each other. They can also compete for attention, working memory and coherence.
This article owns that navigation problem. It does not own the paper-versus-screen comparison in Reading Medium Effects. It does not own synthesis across several separate sources in Multiple-Source Integration. Here the question is narrower: what happens when the text itself is a network and the learner must choose the route while trying to learn?
Quick Answer
Hypertext can make learning flexible because readers can move toward definitions, examples, prerequisites and related concepts when they need them. But nonlinear navigation adds decisions that linear text does not require. The learner must select links, keep track of location, decide when to return, integrate information across nodes and maintain the original learning goal.
A 2018 study of first-year secondary students found lower text-based and structural knowledge after networked hypertext than after linear text, with the disadvantage especially evident for students with lower vocabulary. See Comprehension and navigation of networked hypertexts. A 2024 experiment with children in Grades 3–5 found lower comprehension when passages contained hovering hyperlinks that supplied definitions, while scrolling alone did not reduce performance. See Scrolling and hyperlinks: effects on children’s digital reading comprehension.
That is not a universal anti-link result. A 2021 systematic review and meta-analysis of interactive e-text features found an overall positive effect on reading performance across 26 independent studies, with successful studies often combining several helpful features such as questions with feedback, glossaries or collaboration tools. See Interactive features of e-texts: systematic review and meta-analysis. The design and learning job matter.
A hyperlink is not learning. It is a routing decision. The learning value depends on whether the route improves the model more than the decision disrupts it.
1. Linear Text Carries More of the Route for You
In a conventional linear chapter, the author largely controls sequence. Page two follows page one. Definitions tend to arrive before later uses. Examples are placed where the author expects they will help.
Hypertext transfers some of that sequencing authority to the reader. That can be powerful when the reader knows what they need. It can be expensive when the reader is still trying to discover the structure of the topic.
2. Navigation Is a Cognitive Task
Every meaningful click requires a small decision:
- Is this link relevant?
- Should I open it now or later?
- Will I lose my place?
- Is this a definition, example, prerequisite or tangent?
- When should I return?
- How does the new node connect to what I was building?
None of those questions is necessarily difficult alone. A long chain of them can consume the same control resources needed for comprehension.
3. The Learner Can Become a Navigator Before Becoming a Knower
Digital fluency can hide this problem. A student can switch tabs rapidly, search within pages, open references and return through browser history with great technical competence.
That does not prove the learner has built a coherent representation of the subject. Interface skill and domain understanding are different capabilities.
4. Prior Knowledge Changes the Price of Choice
An expert looking at a network of links can predict which branches are foundational, peripheral or redundant. A novice often cannot.
This is one reason prior knowledge repeatedly appears in hypertext research. When the learner lacks a schema for the domain, each link label is harder to evaluate and the new information is harder to place.
Choice is useful when the learner has enough knowledge to make the choice informative.
5. Vocabulary Can Be a Navigation Variable
The 2018 secondary-school hypertext study found that networked hypertexts were particularly challenging for students with lower vocabulary. That makes sense mechanistically: unfamiliar words do not only make sentences harder. They can make link labels less informative, weaken predictions about destinations and increase the chance that the learner follows the wrong branch.
A vocabulary gap can therefore become both a comprehension cost and a routing cost.
6. Helpful Definitions Can Still Interrupt the Model
Hover definitions sound ideal because they reduce search friction. Yet the 2024 child-reading experiment found worse comprehension in conditions containing hovering hyperlinks that supplied word definitions.
One plausible interpretation is that the extra feature competed with the construction of the passage-level model. That does not prove digital glossaries are generally harmful. It demonstrates a boundary: locally useful information can impose a global integration cost.
7. Scrolling Is Not the Same Problem as Linking
The same 2024 experiment found no negative comprehension effect from scrolling alone relative to the clicking condition. That is important because “screen reading” often bundles several features together.
If we want useful study advice, we should separate medium, scrolling, linking, pop-ups, multimedia, search and notifications rather than treating every digital feature as one mechanism.
8. Hyperlinks Can Signal Importance Even When They Should Not
Web-reading research has found that linked sentences can be judged as more important and can attract different reading behaviour. A visual link is therefore not neutral decoration; it can become an attentional signal.
But authors add links for many reasons: attribution, optional background, commercial routing, navigation, definitions, related reading or source transparency. The presence of a link does not mean the linked concept is central to the learner’s current goal.
9. The Branching Problem
A useful link can contain three more useful links. If every relevant branch is followed immediately, the learner creates a tree faster than working memory can integrate it.
This is the branching problem: local relevance repeatedly wins against global coherence.
The repair is not “never click”. It is to give branches an admission rule.
10. Use a Link Admission Rule
Before opening a link, classify its expected job:
- Blocking prerequisite: I cannot understand the current argument without it.
- Verification: I need to check a claim or source.
- Useful example: helpful, but not required now.
- Extension: valuable after the current model is complete.
- Curiosity branch: interesting but outside the present goal.
Open blocking prerequisites and necessary verification. Park the rest.
11. Open Later Is a Study Skill
Students often treat delaying a link as losing information. It is usually the opposite: delay protects the current representation while preserving the route for later exploration.
Use a short branch list rather than a growing tab bar. Record the link or topic plus one reason for returning. The branch has been captured; it no longer needs to occupy working memory.
12. The One-Model Rule
During a study pass, maintain one visible model of the topic being built.
- a causal chain;
- a concept map;
- a comparison table;
- a claim–evidence structure;
- a worked problem state;
- a short outline of the argument.
Every clicked page must either improve that model or be parked. This prevents the browser’s structure from becoming the learner’s structure by accident.
13. Mathematics: Links Can Break the Problem State
A student solving an unfamiliar Mathematics problem clicks a formula reference, then a worked example, then a video. Each resource may be correct. The danger is losing the exact state of the original problem: what was known, what had been tried and why the previous method failed.
Before leaving the problem, write one line: current state + exact missing information. Then search only for that missing piece. Return and continue before opening another branch.
14. Science: Hyperlinks Can Mix Explanatory Levels
A learner studying respiration might move from a school-level explanation to cellular biochemistry, then to medical pathology, then to a molecular animation. More detail has entered the session, but the levels may no longer align.
Use the current explanatory question as a level filter. A deeper link is useful only if it resolves a gap the present model actually contains.
15. English and Humanities: Links Can Replace Close Reading With Background Accumulation
When reading literature, history or argument, background links can be irresistible. Biography, context and commentary may enrich interpretation—but they can also crowd out the primary text.
First record what the source itself supports. Then use linked background to test, extend or qualify the interpretation. This preserves provenance and prevents explanation from becoming borrowed authority.
16. Nonlinear Reading Needs Stronger Return Cues
In linear text, returning often means continuing from the next paragraph. In hypertext, the learner can re-enter at the wrong node with the wrong goal.
Before following a branch, leave a return cue: “I am checking what X means because it is the missing link between A and B.” When you return, that sentence restores both position and purpose.
This connects to Study Resumption Cues, but here the interruption is self-created by navigation.
17. Graphic Overviews Can Help, but They Are Not Automatically Helpful
One proposed solution to hypertext disorientation is an overview or map. Research reviewed in the 2018 hypertext paper suggests that overviews can reduce some navigation demands, but benefits vary with prior knowledge, working memory and the structure of the overview itself.
A map that is more complex than the text simply moves the navigation problem into another representation.
18. Interactivity Can Help When It Performs a Learning Job
The 2021 meta-analysis of interactive e-texts is a useful counterweight to simplistic “interactive equals distracting” claims. Across the included literature, interactive features showed an overall benefit to reading performance, with successful designs often using features that performed a clear instructional job.
Questions with feedback, well-designed glossaries and collaborative tools are different from arbitrary branching. The correct comparison is not interactive versus static. It is instructionally useful interaction versus interaction whose decision cost exceeds its contribution.
19. The Tab Count Is Not a Learning Metric
Many open sources can create a feeling of serious research. They can also represent unresolved routing decisions.
If ten tabs are open, ask which one currently owns the next cognitive job. Close, park or group the others. The aim is not digital minimalism for its own sake; it is preserving the learner’s ability to know what they are doing now.
20. Search Results Are Hypertext Too
A search-results page is a high-density routing environment. Snippets invite branching before the learner has established an owner source or a stable question.
For broad research, separate discovery from learning:
- Discover candidate sources.
- Select one or two owners for the current question.
- Read deeply enough to build the model.
- Reopen search only when a specific gap remains.
This connects to Search as Learning.
21. AI Answers Can Become Invisible Hypertext
An AI answer can compress several source branches into one fluent response. That reduces visible navigation cost but introduces another risk: the learner may no longer see which claims came from which sources or where uncertainty entered.
If AI is used to navigate a topic, preserve provenance for claims that matter and ask the learner to reconstruct the model without the generated answer. Better routing is not automatically better learning.
22. The Delayed Coherence Test
After a hypertext session, close the browser and answer three questions:
- What was the original question?
- What is the coherent answer or model now?
- Which branch materially changed that model?
If the learner remembers many interesting pages but cannot reconstruct the main model, navigation succeeded as browsing and failed as studying.
23. The Changed-Path Test
Knowledge should not depend on reproducing the exact click path.
Test the topic from a fresh question, a different starting page or a blank sheet. If the learner can explain the structure without retracing browser history, the model has become more independent of its navigation history.
24. Parent and Tutor Guide
When a student has many tabs open, avoid treating the number itself as the problem. Ask what each tab is doing.
- Which page is the current owner source?
- What question made you open this branch?
- What would make you return to the main text?
- Which links are prerequisites and which are curiosity?
- Can you explain the topic if all tabs close now?
The goal is not adult control of every click. It is a learner who can keep a learning objective stable while moving through a network.
25. Evidence Boundary
Hypertext research spans different ages, structures, devices, tasks and interactive features. Results should not be reduced to “linear text good, hyperlinks bad”. Networked hypertext can impose additional navigation load, especially for learners with weaker prior knowledge or vocabulary, but carefully designed interactive features can also support reading and learning.
The safe educational conclusion is architectural: when navigation itself consumes substantial cognitive control, reduce unnecessary branching, make structure more visible and test whether linked exploration improved the learner’s coherent representation rather than merely increasing exposure.
26. The Hypertext Study Protocol
- Write the current learning question before opening branches.
- Choose one owner source or main path.
- Classify each tempting link: prerequisite, verification, example, extension or curiosity.
- Open only prerequisites and necessary verification immediately.
- Leave a return cue before branching.
- Add new information to one visible model.
- Park optional branches outside working memory.
- Close the network and reconstruct the answer.
- Use a fresh question or source to test whether understanding survives a changed path.
27. Return: Do Not Let the Link Graph Become the Knowledge Graph
The web is built to let information branch. Learning is built to let knowledge connect.
Those are not the same structure.
Navigate when a branch solves a real information gap. Park branches that do not. Keep one learning model visible. Return deliberately. Then close the browser and prove that what remains in the learner is more coherent than the path that produced it.
Continue through Reading Medium Effects, Multiple-Source Integration, Search as Learning, Study Resumption Cues, the How Studying Works Numbered Series Reading Index and the How X Works Hub.