The Task Changes the Interface

People may not use one AI interface for a whole information journey. As someone learns, the question they are asking changes, and the place that suits the new question may be different from the place that suited the first. The same person can move between search, an AI-generated answer, a conversational assistant and primary sources within a single decision, and move back again.

The technology lead at a 500-person company has been asked to look at security platforms. She starts where most of us would, with a search:

Best cybersecurity platforms for a 500-person company.

What comes back is useful. There may be an AI-generated summary at the top, then vendor names, comparison pages, review sites and the security press. Twenty minutes of reading tells her which categories exist, which vendors keep appearing and what the vocabulary is.

It also shows her that the question was too broad. What she needs to ask is closer to this:

We're about 500 people, mostly on Microsoft 365 and Azure. We already use CrowdStrike. Our security team is small, our cyber-insurance requirements matter, and we can't take on a platform that needs two dedicated administrators. What should we actually be evaluating?

She could type that into a search box. But it is long, conditional and likely to lead to several follow-up questions, which makes a conversational interface attractive. She opens one and keeps going. Why would you favor one approach over another? Assume ransomware recovery matters more to us than advanced threat hunting. What trade-offs does that create? Give me the questions I should ask the three vendors we’re meeting next week.

The search didn’t fail. The task changed.

She May Move Again

The conversation leaves her with a shortlist and a sharper sense of what matters. It also leaves her with claims to check. Is that integration officially supported? What does the current pricing page say? Has the certification been renewed? What are the contract terms?

For those she goes to the vendor’s documentation, an official integration page or a standards body, or back to search to find them. Then, quite possibly, back to the conversation to ask what the verified facts mean for the choice.

What began as “what are my options?” became “which of these fits my situation?” and then “is this specific claim true?” Those are different information problems. She isn’t a search user who became a chatbot user. She is one person working on one decision, and her need changed three times in an afternoon.

Not a Kind of User

It’s tempting to sort people by product: Google users, ChatGPT users. There may one day be good evidence that different groups favor different tools. I don’t have it, and I think the more useful observation sits somewhere else. Interface choice can be situational. The person who begins in search and the person who later reasons through trade-offs in a conversation may be the same person, twenty minutes apart.

Mark recently looked at the machine side of this: the question changes what an AI system needs to retrieve. There is a human counterpart. As the information need changes, does the user stay in the same place?

Discovery Changes the Question

Part of the reason she moved is that the early reading made her better at asking. At the start, all she could ask was what exists. Once she knew the categories, she could ask what fits her constraints. Then what she might be overlooking, then whether a specific claim was accurate, then what to do next.

That isn’t one query repeated in different places. Her understanding changed, so the problem did.

Discovery changes the question, and the changed question can change where the user chooses to ask it.

What Each Interface Makes Easy

Three things usually get collapsed into “the AI”: the model that produces the language, the retrieval that brings outside information into an answer, and the interface the person actually uses. A preference for one environment at a given moment may have as much to do with the third as with the first.

Different interfaces make different behavior easier. Designers call these affordances. A search box can encourage a compact query, and a results page makes it easy to scan alternatives and click outward to several sources. A conversational interface makes it natural to add personal context, ask a follow-up, revise an assumption and keep the thread across many turns. A primary source, whether that is documentation, a regulator’s site or a product database, may be the better place to establish exact wording, official status or current availability.

None of that is exclusive. A search product can hold a long exchange, and a conversational assistant can look things up. But the interface does more than deliver an answer. It can influence what the user finds natural to ask next.

That is also why I hesitate to think of these simply as channels. A channel usually describes a route through which information or attention moves. An AI interface can also become an environment in which the problem itself is worked through: the user asks, learns, reframes, tests an idea and returns to the evidence.

The Same Pattern in a Different Decision

None of this is particular to software. A couple considering a second home in a coastal market they don’t know might begin with best areas to buy a second home and the name of the region. Search, and an AI summary within it, can be useful for that: neighborhood names, broad price ranges, local agencies, a first look at listings.

Then the real question surfaces, and it is personal.

We'd use the house for six weeks in summer and the odd weekend. We care more about walkability than acreage, we don't want heavy maintenance, and we'd rather have water access than a big pool. What kinds of areas should we be comparing?

That invites a conversation, including the uncomfortable parts: What are we undervaluing? What would make us regret choosing the more remote option?

When they want to know what is for sale today, or what comparable houses sold for, they are back with property search, brokerage inventory, maps and sale records.

Sometimes the User Moves Back

The movement isn’t a line. An executive trying to understand a regulatory change might search for the regulation by name, because search can be useful for locating the official text, the effective dates and the recent commentary. She might then describe her company to a conversational assistant and ask which parts are likely to affect it, what assumptions it is making, and what she should take to counsel. When a specific requirement matters, she returns to the regulator’s guidance and to her lawyers.

A conversation can be useful for reasoning around evidence without replacing the source of it. And each return trip tends to produce a better question for the next one.

I’ve been calling this a multi-interface journey. It’s a description, nothing more formal: an information task that moves between digital environments as the user’s needs change. It might take in search, an AI-generated answer, a conversation, maps, a product database, documentation and a company’s own site. The number of stops matters less than the fact that discovery can be spread across them rather than contained within one.

Where a Question Starts

For many people search remains an almost frictionless place to begin. The browser address bar, long habit and AI answers now built into results can make it a natural first move. The usage data we looked at in an earlier piece suggests people are adding AI tools while keeping search. It can’t show which questions go where, but it fits a picture in which the tools serve different parts of the same job.

The easiest place to start isn’t necessarily where the whole decision happens. Distribution can determine where a question begins without determining where the thinking ends. The reverse holds too: someone who starts in a conversation may leave it for fresh or authoritative evidence.

What I Can and Can’t Claim

This idea came from watching different AI systems behave differently on the same questions. That is evidence about systems. It says nothing about journeys. Citation data can’t show how often people switch interface, in which direction, or why, and it can’t separate the effect of the model from the retrieval or the design.

So I hold the specifics loosely. I wouldn’t say that search is for orientation and conversation is for reasoning, or that people start in one named product and finish in another. The products are converging. Search supports follow-up, assistants search the web, research tools have become conversational, and any feature comparison written today will date quickly.

What interests me is the more durable point. Information tasks differ, interfaces make different things easy, and there is less and less reason to assume one interface must contain the entire journey. Which interfaces people start with for which tasks, when they switch and what pulls them back to primary sources are open questions, and worth studying properly.

One Decision, Several Places

We tend to ask which AI interface will become the place people go for answers. For the technology lead, there wasn’t one place. Search began her question. A conversation reshaped it. A vendor’s documentation settled part of it, and she went back to the conversation with a better question than the one she started with.

If the task keeps changing, there is no reason to assume the interface has to stay the same.