Does generative AI chat actually replace traditional search?

Nielsen Norman Group ran a qualitative lab study in which participants brought their own real research tasks, to watch how AI-powered search and chat are changing information-seeking habits. Its central finding is that generative AI "isn't close to completely replacing traditional search": every participant still used traditional search — keywords, results pages, visiting content pages — multiple times during the study, and "nobody relied entirely on genAI's responses (in chat or in an AI overview) for all their information-seeking needs." Traditional search and AI chat were often run "in tandem," sometimes even to fact-check each other. Within ordinary search, though, attention is already shifting: the AI overviews that "now top almost all search-results pages" steal attention and often answer the query without a click, a pattern NN/G says a quantitative study by Pew Research confirmed — Google searchers who saw an AI overview were "substantially less likely to click on results links."

The mechanism NN/G gives is habit stickiness plus an incentive threshold, not technology novelty. "Information-seeking habits are sticky" once a method feels reliable and convenient — their example is a participant who defaulted to Google "because it was built into his Chrome browser" — and changing an ingrained habit requires "a significant incentive." GenAI clears that bar often enough to be reshaping behavior, but the clearest obstacle NN/G reports is not resistance, it's discoverability: one participant used ChatGPT routinely for work email yet "had never considered using it for information seeking" until the study put him in front of Gemini for a plumbing question. NN/G's conclusion is that "discoverability remains a major challenge for genAI design — not only the discoverability of the tools, but also the discoverability of its possible functions and how to use it," and that AI overviews work best for "quick definitions and fast facts" rather than the synthesis-heavy tasks where chat has the edge.

This sits beside Which AI design principles for social media have research support?, whose only principle that note found backed by reported results was interface simplicity; NN/G's study gives that principle a concrete field case — a user with months of ChatGPT experience still didn't know chat applied to search until prompted. It also qualifies Do generated interfaces outperform text-based chat for most tasks?: that paper frames structured generated UI as beating conversational chat on preference grounds, while NN/G's participants weren't comparing interface styles at all — they simply hadn't discovered that chat was an option for this kind of task, a prior-to-preference barrier the UI-generation framing doesn't address.

NN/G does not report how many participants it studied, so the behavioral claims rest on an unspecified small qualitative sample plus one cited external figure (the Pew Research click-through finding, not NN/G's own data) — there is no quantitative measure here of how widespread the discoverability gap or the habit shift actually is across the population of search users. The piece also does not resolve whether the current AI-chat discoverability gap is a transient early-adoption lag, likely to close as "Chat" becomes a verb the way "google" did, or a durable design failure that will persist regardless of model capability; NN/G frames it as an open design challenge rather than predicting which.