A Yale study published in 2015 found that people who searched the internet for an explanation came away feeling smarter afterward, and the effect held even when the search failed to produce an actual answer. That result reads, at first, like a quirk of self-deception, a strange little bug in how people evaluate their own understanding. It’s worth treating instead as a description of a product working exactly as designed, doing precisely the job it was built to do.
What the experiments actually showed
The research, led by then-doctoral student Matthew Fisher and senior author Frank Keil, with co-author Mariel Goddu, ran more than 1,000 participants through nine separate experiments. In the core design, one group searched online for an explanation to a question while a separate control group either didn’t search or was given the same information without needing to look for it. Afterward, everyone rated how well they thought they could explain unrelated topics, ones the search had nothing to do with. The searchers rated their own general knowledge higher across the board, not just on what they’d looked up. As Fisher put it, “people who search for information tend to conflate accessible knowledge with their own personal knowledge.”
The detail that makes the finding harder to dismiss as a fluke is what happened when the searches came back empty. Even when the query returned nothing useful, the inflated sense of knowing persisted. Keil, the Yale psychology and linguistics professor who led the lab, described the effect as strong enough to survive its own failure: “the cognitive effects of ‘being in search mode’ on the internet may be so powerful that people still feel smarter even when their online searches reveal nothing.” The confidence wasn’t tracking the quality of the answer. It was tracking the act of searching itself.
The researchers ruled out a few obvious alternate explanations along the way. It wasn’t simply that searchers were more confident people in general, since a separate group of participants who didn’t search at all rated their own knowledge no higher on the same unrelated questions. It wasn’t that the search results happened to contain flattering or reassuring information, since the effect showed up even for topics the search never touched, and even when the search came up empty. What was left, across nine variations on the same basic setup, was something closer to a contagion effect: the sense of having reliable information nearby bled into a sense of holding that information personally, regardless of whether it had actually transferred.
Feeling smart and being smart are not the same transaction
This is the part worth sitting with longer than the headline finding usually gets. A search engine’s business model does not depend on users walking away with more accurate beliefs about the world. It depends on users coming back. A tool that reliably left people feeling less certain after using it, even if that uncertainty were more accurate, would be a weaker product by one of the metrics that most determines whether a search company survives commercially: repeat use. A tool that reliably leaves people feeling resolved, informed, and slightly smarter, regardless of whether the underlying question actually got answered, is a better product in exactly the sense that matters commercially.
Put plainly, the confidence boost isn’t a side effect of the product, an incidental emotional byproduct that happens to ride along with the real service. For a search-based business, it functions closer to the actual thing being delivered, whether or not that’s how the product is described in its own marketing. The information retrieved is almost incidental to the feeling of having retrieved it, which is a strange thing to realize about a tool most people describe as purely informational.
This also reframes what “engagement,” the metric most of these products are ultimately optimized around, is actually measuring. It isn’t a proxy for how useful the tool is. It’s closer to a proxy for how reliably the tool produces the sensation of usefulness, which can rise or fall independently of whether anyone’s actual understanding improved. A product team watching engagement climb has no built-in signal telling them whether that climb reflects better answers or simply a more convincing feeling of having received one.
Why this matters more now than it did in 2015
The original study predates the current wave of AI chat tools that answer in fluent, confident prose rather than a list of links to sort through yourself. Keil never tested chatbots directly, since the study predates them, but the mechanism he described doesn’t obviously require search results to look like a list of links. If simply being in “search mode” was enough to inflate a sense of personal knowledge, a tool that skips the sorting entirely and hands over a complete, confident-sounding paragraph seems, on its face, more likely to deepen the effect than soften it. That’s an inference drawn from the original finding, not something the 2015 study itself tested.
None of this means looking things up is bad, or that the confidence it produces is always misplaced. Plenty of searches do resolve into real, verifiable understanding. The finding is narrower and more specific: the feeling of having learned something and the fact of having learned it can come apart cleanly, and the tools built around search have very little commercial incentive to keep those two things attached. A rising sense of certainty after a quick search is not, on its own, evidence that the certainty is earned.
A useful habit, if the finding is taken seriously, is a small delay between searching and concluding. Asking whether you could actually reconstruct the explanation without the page open, rather than whether the page felt satisfying to read, targets the exact gap the study identified. It won’t eliminate the illusion, since the researchers found it persisting across contexts specifically designed to strip it away. It at least introduces a checkpoint between the feeling and the conclusion the feeling is quietly standing in for — one of the few real defenses against mistaking one for the other.