If your assistant can only reach the web after you already searched, is it really the new front door to information?

This explores whether the kind of task someone has decides if they search first or ask an assistant first. The honest answer is that the collection doesn't yet have a study that breaks behavior down by task type. What it does have is a surprising finding about order, which changes the question. A cross-surface panel study found that assistant sessions come after search and browsing 20.6 percentage points more often than before them Do people use AI assistants before or after searching?. That cuts against the popular 'answer engine' story, in which AI becomes the front door to information. Search still tends to open the journey. The assistant usually comes in later, once the person already has some context. The same study also found that sessions using only an assistant are more common than sessions using only search, even among the same people. So many tasks seem to be handed entirely to one tool or the other, rather than moving neatly from one to the next.

Nielsen Norman Group's qualitative work helps explain why task type is hard to see in the data. Every participant kept using traditional search throughout their tasks, often running it alongside AI chat Does generative AI chat actually replace traditional search?. The main obstacle wasn't distrust of AI. People simply didn't know which tasks AI chat is good for. In other words, people may not yet be picking a tool based on the task. Many fall back on search because they're unsure what else to do. The line between the two tools is also blurring inside search itself. Eye-tracking research shows that AI Overviews now take up most of the attention that used to go to the top result: attention on the first-ranked link fell from 31% to 9%, and people trusted both equally Where do searchers look when AI Overviews appear?. When the search results page already contains an AI answer, 'search first or assistant first' stops being a clean choice.

The closest thing the corpus has to a task-based rule comes from the system side, not the user side. Work drawing on conversation analysis asks when an AI agent should stop searching and ask the user instead When should AI agents ask users instead of just searching?. Its answer is that unclear goals and fuzzy scope call for a conversation, while well-defined requests can go straight to tools. Researchers make a related point about assistants that have no record of what they don't yet know about the user Do language models know what they don't know about users?. Reading this across to people suggests a plausible split: tasks with a known target suit search, and tasks where you're still working out what you want suit a dialogue. That fits the pattern of search first, assistant second. People find their footing with links, then turn to an assistant to make sense of what they found.

One more twist: the tasks assistants are most often sold for may be the wrong ones. Most users don't want routine tasks like email and calendar automated, and they value doing those tasks themselves Does the personal assistant model actually serve most users?. So a better question than 'which tool for which task' may be 'which part of a task do people actually want to hand off?' That question remains open in the collection.