LLMs Are Time Machines That Don’t Tell You How Far You Went

# What the article argues An LLM takes present-you, the person with a question, and hands back future-you, the person making a decision. It compresses a research journey that used to take hours or days into a single confident answer. That same compression removes signals—what the author calls path metadata—that people used, implicitly, to judge how much to trust what they found.

# How the journey used to inform judgment Before search engines and LLMs, answering a serious question often meant consulting many sources over time. The time spent, the variety and depth of sources, and visible contradictions in what you found were all implicit cues about how robust your answer was. Those cues lived in the process itself: how many books or journal articles you consulted, how much time it took, whether results agreed. They helped shape how firmly you committed to a conclusion.

# What LLMs remove

# Evidence the effect matters The article cites multiple empirical findings:

• Earlier research (2015 Yale experiments) showed internet searching inflates people's self-assessed knowledge—access gets confused with understanding—and that effect existed even when searches turned up nothing.

These different methods—controlled experiments and passive tracking—arrive at converging concerns: summaries reduce follow-up engagement and leave people with weaker understanding.

# Why links alone don't fix it The Wharton experiments included versions where models supplied live web links alongside summaries. Participants still did not explore the links. Once the summary arrived, the sources beside it ceased to be interesting. The presence of a citation in the answer does not reliably restore the travel record or prompt deeper engagement.

# Practical implications for publishers and readers

# Bottom line LLMs speed access to answers but remove the byproducts of research that people used to judge reliability. Multiple studies show that summaries reduce follow-up browsing and can leave people with shallower understanding even when the underlying facts are identical. Returning links or citations alongside a summary does not reliably restore the lost engagement.