Internet search and learning can feel almost inseparable. A question appears, we find a useful page, and within seconds the answer seems clear. Yet the ease of finding information can create a misleading impression: because we can access an explanation, we may feel that we have learned it.
Matthew Fisher and his colleagues examined this gap between access and knowledge. Their study predates generative AI, but it belongs squarely within current discussions of AI and cognition. Search engines retrieve sources; AI can also summarize, explain, and reorganize them. In both cases, fluency at the point of access may conceal how little remains when the tool is removed.
Summary of the Study
The researchers conducted five experiments with adults recruited online. Participants prepared for short quizzes on topics such as inflation, photosynthesis, and autism. Some used Google to locate specified webpages, while others received the relevant material directly.
Later experiments tightened the comparison by controlling distractions and study time. In the final preregistered experiment, both groups viewed the same webpages; the main difference was whether participants reached them through Google or a direct link.
Across the experiments, people who searched later remembered less of the material than those who accessed it directly. The result persisted when the researchers gave both groups equal study time and reduced other obvious differences between the conditions.
Just as important, searchers did not adjust their expectations downward. Their confidence remained disconnected from their weaker performance. This makes the study more than a warning about memory. It identifies a metacognitive problem: people may mistake the ability to relocate information for evidence that they know it.
The authors interpret this pattern through cognitive offloading and transactive memory. When information feels readily retrievable, we may encode how to find it while devoting less effort to the content itself. That explanation is plausible, especially because the deficit remained when participants viewed identical pages. It also connects with later research on AI and cognitive offloading.
Still, the experiments demonstrate a learning difference more directly than they demonstrate the proposed mental mechanism. They did not observe encoding itself, so the explanation should be treated as a well-supported interpretation rather than a settled account.
The scope of the evidence also needs care. Participants were adults completing brief, tightly structured tasks online, and learning was measured through factual multiple-choice questions soon afterward. The study does not tell us whether search harms deeper understanding, transfer, inquiry skills, or long-term retention in authentic classrooms.
Nor does it test generative AI. Extending the findings to ChatGPT is reasonable as a research question, not as a conclusion. AI may intensify the problem by making answers more immediate, but it can also prompt elaboration, feedback, and comparison in ways ordinary search does not.
Practical Insights
The educational issue is not whether students should search. Search is essential to contemporary learning. The issue is what happens after useful information appears on the screen.
- Separate finding from learning. Ask students to close the source and explain the idea from memory. Any gaps that emerge show what still needs attention.
- Begin with an initial account. Before searching, students can write what they know, what they are unsure about, and what they expect to find. This creates a basis for noticing how the source changed their understanding.
- Add a retrieval step. After research, have students reconstruct the central argument, draw a concept map, or answer a related question without access to the source. A polished set of notes is evidence of access; independent retrieval offers better evidence of learning.
- Make confidence testable. Students can predict how well they will explain or recall the material before completing a short no-search check. Comparing prediction with performance helps expose the difference between “I can find it” and “I know it.” This matters when easy assistance encourages the kind of metacognitive laziness found in AI-supported learning.
- Use tools according to the learning goal. If the purpose is locating credible sources, search is part of the skill. If the purpose is remembering a concept or building an explanation, repeated retrieval and elaboration must remain with the learner.
For researchers, the next step is to test these effects in classrooms and with generative AI. Studies should examine delayed retention, transfer, conceptual understanding, and students’ ability to judge the quality of retrieved or generated information. They should also compare different tool routines. Searching followed by independent explanation may produce a very different outcome from searching followed by copying, even though both are recorded as internet use.
The practical warning is easy to miss precisely because search works so well. Information can be available, relevant, and understood in the moment without becoming knowledge a learner can later use. Education has to preserve the step between encountering an answer and making it one’s own.
Reference
Fisher, M., Smiley, A. H., & Grillo, T. L. H. (2022). Information without knowledge: The effects of Internet search on learning. Memory, 30(4), 375–387. https://doi.org/10.1080/09658211.2021.1882501
