The New Shape of Search: How Conversational AI Recomposes Information Seeking
This paper analyzes user behavior across search and conversational AI interfaces to reveal that information journeys are being "recomposed" such that assistants often follow traditional search and browsing rather than initiating them, and are more frequently self-contained than search sessions, without necessarily implying task resolution.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Map of Your Digital Mind
Imagine you are trying to solve a mystery. In the old days, the story was simple: you had a question, you asked a librarian (the search engine), got a list of books, and then you went to read them. This is how we've thought about finding information for decades. But lately, a new character has entered the library: a super-smart, chatty robot assistant that can write stories, solve math problems, and summarize books just by talking to you.
The big question on everyone's mind is: Is this robot replacing the librarian? Are we stopping our search for books the moment the robot starts talking? Or are we using the robot and the books together in a new, weird dance? This paper is like a detective story that tracks exactly how people move between asking the robot questions and clicking on web links. It doesn't just guess; it watches real people (who said "yes" to being watched for research) to see the actual shape of their thinking. The goal is to understand if our digital lives are getting simpler (just talking to the robot) or if they are getting more complex, weaving between talking and reading in ways we haven't mapped before.
The New Shape of Search: A Tale of Two Journeys
So, what did the researchers at Scrunch AI actually find? They looked at the digital footprints of people in the US and UK during February and March 2026. They wanted to see the "shape" of a search session. Usually, we think of a search session as a straight line: you ask a question, you get answers, and you leave. But with these new AI assistants, the path is getting twisted.
The researchers discovered that the "old shape" of searching and the "new shape" of using an AI assistant are actually mirror images of each other.
The Old Shape: The Search Engine as the Doorway
Think of a traditional search engine like a front door. When you use it, you usually start at the door, then you wander out into the house (the internet) to find what you need. The data shows that for regular search, the journey almost always starts with the search query, and then you go looking at websites. About 60.2% of the time, the content you read comes after you search. You ask, then you click. It's a very forward-moving path.
The New Shape: The Assistant as the Deep Dive
Now, imagine the AI assistant as a cozy reading nook deep inside the house. The researchers found that people often arrive at this nook after they've already been wandering around the house. In fact, for AI assistant sessions, the pattern flips! About 45.5% of the time, people were already looking at websites before they started talking to the assistant. They gather some clues, get confused or overwhelmed, and then say, "Hey robot, help me make sense of this."
This flip is the paper's biggest discovery. They call it recomposition. It means the activity is rearranged, not just replaced.
- Search tends to be the start of the journey (the door).
- Assistants tend to sit deeper inside the journey (the reading nook), often used after you've seen some web pages.
The "Contained" Mystery
One of the most interesting things they measured is how often a session stays "contained." A contained session is one where you talk to the assistant and never leave to look at a website.
- For AI assistants, about 34.1% of sessions are totally contained. You ask, the robot answers, and you stay there.
- For regular search, only about 19.5% of sessions are contained. You almost always click out to a website.
This might make you think, "Aha! The robot is replacing the web!" But the paper is very careful here. It says: No, that's not what this proves. Just because you stay in the robot's room doesn't mean you didn't need the web earlier. In fact, the data shows that for the same people, they use the robot more often after they've already been browsing. The robot isn't necessarily killing the search; it's just changing the order of operations.
The "Bridge" and the "Interleaved"
The researchers also found that many people are doing a "bridge" dance. They go: Website -> Robot -> Website -> Robot. About 37.1% of AI sessions are like this, where the robot and the web are mixed together. This is much less common for regular search (which is more like a straight line), but for AI, it's a huge chunk of the activity.
What the Paper Rules Out
It's important to know what this paper doesn't say.
- It doesn't say AI is "winning" or "losing" search. The paper explicitly states it cannot tell if AI is creating new demand or destroying old demand. It just maps the shape of the journey, not the total number of questions asked.
- It doesn't say the robot "solved" the problem. Just because you stayed in the AI session and didn't click a link doesn't mean you got a perfect answer. Maybe you were just chatting, or maybe you gave up. The paper calls this "observed containment," not "task completion."
- It doesn't say the robot is the only way to search. The study only looked at "standalone" assistants (like a separate app or website you visit), not the AI that is built inside the search results page (like Google's AI Overviews).
How Sure Are They?
The researchers are very confident in the shape of the data they found. They checked their work in two different months (February and March 2026) and the results were almost identical. They also checked it across different types of websites (shopping, news, coding) and the "flip" in direction (search first vs. browsing first) happened every time.
They are also very careful about the numbers. They say that if you change how they define a "session" (for example, waiting 15 minutes vs. 60 minutes before calling a session over), the exact percentages change a bit. But the main story—the fact that search usually comes first, and AI usually comes later—stays the same no matter how they slice the data.
The Takeaway
The paper concludes that we are not just swapping one tool for another. We are building a new, more complex ecosystem.
- Search is still the front door, the place where we usually start our journey.
- AI Assistants are becoming a deep, internal tool we use to synthesize what we've already found, or to help us when we get stuck in the middle of our journey.
The "new shape of search" isn't a straight line anymore. It's a loop, a bridge, and sometimes a deep dive. We aren't just asking questions and getting answers; we are weaving between reading, clicking, and talking in a way that makes our digital lives feel more like a conversation and less like a checklist. And that, the authors suggest, is a fundamental change in how we seek information.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.