Letting AI summarize your research for you might make you understand it less — because the real learning happens in the digging, not the reading.

This explores whether handing the work of searching over to AI leaves people understanding a topic less deeply. The most direct evidence in the corpus says yes. Across seven randomized experiments with more than 10,000 people, those who learned a topic through ChatGPT reported learning less and felt less ownership of what they knew than people who used ordinary web search. Their advice to others was also thinner: independent raters judged it sparser and less informative Does learning from AI summaries produce shallower knowledge than web search?. The likely reason is that the effort of search is where much of the learning happens. When you click through sources, compare them and piece together your own picture, you build understanding. A finished summary skips that step.

The corpus also suggests why summaries might go shallow even when they're accurate. Common words tend to be more general than rare ones, so when an LLM picks the most frequent way to say something, it drifts toward abstraction. The precise, expert-level detail is the first thing to go Does word frequency correlate with semantic abstraction?. So low effort for the reader may also mean less specific content to work with.

Other notes show how quickly reduced effort becomes the default. An eye-tracking study found that when Google's AI Overviews appear, they take over the spot on the page that used to get the most attention. The share of attention going to the top-ranked link fell from 31% to 9%, and people trusted the AI summary just as much as the ranked results Where do searchers look when AI Overviews appear?. People also judge AI answers by surface cues. In one large study, adding irrelevant citations raised users' trust almost as much as adding relevant ones Do users trust citations more when there are simply more of them?. Put together, this suggests that when the work is done for them, readers check sources less and judge credibility by how an answer looks.

There is a counterweight. A Nielsen Norman Group study found that people still used traditional search alongside AI chat, often running both side by side Does generative AI chat actually replace traditional search?. Effort hasn't disappeared, at least not yet. One surprising twist: on the machine side, more search effort clearly buys better answers. Research agents improve steadily as they get more search rounds, much as models improve with more reasoning time Does search budget scale like reasoning tokens for answer quality?, and agents that search the live web beat models that rely only on what they memorized in training Why do search agents beat memorized retrieval on hard questions?. The effort that produces understanding still happens, but the agent does it instead of the person. The answer gets better while the person's understanding may get shallower.

The corpus has one strong direct study and several supporting threads. It doesn't yet have research on interventions, such as designs that bring back productive effort without losing the convenience of AI search.