People keep typing search terms into AI tools built to just answer them directly — why does the old habit stick?
Why do ingrained search habits persist even when AI alternatives are available?
This explores why people keep searching the way they always have (typing queries, scanning ranked links) even when AI tools that answer directly are right there. The corpus has little direct evidence on this, but what it has points somewhere unexpected.
This explores why people keep searching the way they always have (typing queries, scanning ranked links) even when AI tools that answer directly are right there. To be upfront: the collection has almost no research on human search habits or why people resist switching tools. What it does have makes the question more interesting, because it suggests the habit may be changing less than it seems, and that AI systems develop their own stuck search habits for reasons we can actually see.
Start with the one study that watched real searchers. In an eye-tracking study from RMIT and Microsoft, AI Overviews took over the spot where people's eyes land first on the results page. Attention to the top-ranked link fell from 31% to 9% Where do searchers look when AI Overviews appear?. Yet people trusted the AI summary and the ranked links equally. So searchers didn't trade their old method for a new one. They kept the habit (go to a search page, scan from the top) and let the AI answer move into it. This suggests habits persist partly because the new tool arrives inside the old ritual, so nothing forces anyone to rethink how they search.
The second angle is about thinking rather than behavior. The Rose-Frame work describes three mental traps that make each other worse: mistaking a fluent answer for the reality it describes, treating a quick gut feeling as careful reasoning, and having existing beliefs confirmed back to you Why do people trust AI outputs they shouldn't?. This doesn't directly explain loyalty to old habits, but it shows that people's choice of where to look follows ease and familiarity more than an assessment of which tool is better. That same pull toward the familiar plausibly keeps old routines in place.
Now the sideways turn: AI search agents get stuck in habits too, and here the mechanism is well understood. When search agents are trained with reinforcement learning, they converge on a few narrow strategies that earned rewards before and stop exploring alternatives Does reinforcement learning squeeze exploration diversity in search agents?. Training on varied examples of how others search keeps them flexible. Breaking these patterns can take an outside view: in one system, an outer AI loop read the inner loop's code, spotted its repetitive patterns, and wrote new search methods that improved results 5x Can an AI system improve its own search methods automatically?. Training on messy exploration, including failures and backtracking, also produces sturdier behavior than training only on clean solutions Can models learn better by training on messy exploration paths?.
The machine findings are an analogy, not evidence about people. Still, they give a useful lens: habits harden when one approach keeps getting rewarded, and they loosen when you see a variety of other approaches or step outside the loop to look at it. If you want the human side of this question answered properly, the collection doesn't have it yet. Studies of how people adopt new tools and how familiar routines hold them in place would be the gap to fill.
Sources 5 notes
Eye-tracking data shows AI Overviews receive significantly longer fixation times, reducing attention to the first-ranked result from 31% to 9%. Trust ratings between AI Overviews and ranked results remained equally high despite this attention shift.
Rose-Frame identifies map-territory confusion, intuition-reason conflation, and confirmation-bias reinforcement as traps that multiply their distorting effects when they co-occur. Evidence from cross-linguistic overreliance and architectural transformer biases confirms the compounding mechanism operates universally.
RL training compresses behavioral diversity in search agents through the same entropy collapse mechanism documented in reasoning—policies converge on narrow reward-maximizing strategies. SFT on diverse demonstrations preserves exploration breadth, suggesting diversity-preservation techniques are essential for RL search scaling.
An outer loop successfully read inner loop code, identified bottlenecks, and generated new Python mechanisms at runtime, discovering combinatorial optimization and bandit methods that broke the inner loop's deterministic patterns and improved performance on GPT pretraining by 5x.
Research shows that training on messy trajectories—failed attempts, self-correction, and backtracking—teaches more robust reasoning than training only on shortcut solutions. This approach models o1-style deep reasoning as search internalization rather than solution memorization.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- RLVMR: Reinforcement Learning with Verifiable Meta-Reasoning Rewards for Robust Long-Horizon Agents
- Bilevel Autoresearch: Meta-Autoresearching Itself
- Beyond Hallucinations: The Illusion of Understanding in Large Language Models
- An Eye Tracking Study: Are AI Overviews Changing Search Behavior?
- Dream-RSI: Recursive Self-Improvement through Evolving Worlds
- Vector Policy Optimization: Training for Diversity Improves Test-Time Search
- Recursive self-improvement of AI research agents
- Jointly Reinforcing Diversity and Quality in Language Model Generations