Search’s Cognitive Bottlenecks: What LLMs Solve (And What They Don’t)

Google drops a list of blue links plus infinite scroll.
By contrast, an LLM produces one synthesized paragraph and a handful of citations.
Within cognitive psychology, this can be framed as a reduction in working-memory saturation, and it’s a UX improvement.
People can juggle about four bits of information before attention fragments.
Cutting the visible options from dozens to one feels like progress.
This was rapidly understood and incorporated via AIOs.

The paradox of choice looks solved, yet the story does not end there.
Search UX also tangles with three other bottlenecks that remain more under the radar.

The first is the time-pressure interaction.
Lots of tasks, from checking a stock price to booking a train, happen under thirty seconds.
AIOs help by front-loading a ready answer, but LLM generation latency can still spike under network load.
For rushed users, a three second wait competes poorly with a skimmable list that loads instantly.

The second blind spot is the cost of regretted clicks.
Fewer links mean fewer bad clicks, yet hallucination risk now replaces them.
A single fabricated fact can cost more than five irrelevant links.
Traditional Google at least lets dwell time and pogo-sticking reveal low quality, metrics the ranking team can penalize.
LLM chat hides that feedback loop because the answer arrives before any click evidence.

The final and most dangerous bottleneck is overestimation of competence.
Experiments show that prose-rich answers inflate confidence even when accuracy drops.
Users confuse fluent language with truth and walk away believing they mastered the topic.
Classic SERPs expose multiple voices, forcing comparison and planting a seed of doubt.
LLM chat narrows the narrative and amplifies the illusion.

The takeaway: to truly revolutionize search, compressing options is necessary but not sufficient for cognitive alignment.
Of the four bottlenecks that psychologists have identified in traditional search UX, LLMs strongly mitigate one, partially resolve another, substitute a third without eliminating it, and worsen the fourth.