Is AI replacing search?

As part of getting to know the product and the browser/AI space better, I recruited around 30 people to talk about their browser and AI behavior. Because Magic Box — our name for the omnibox — is essentially a more powerful version of a search box, I found myself thinking a lot about search behavior.

The question is not simply: Will AI replace search?

The more interesting question is: Which search behaviors does AI make obsolete — and which ones still need the web?

AI is replacing the hard parts of search

Generative AI is changing how people look for information, at least in the bubble I’m in. A few months ago, I barely used my ChatGPT Pro account. “Why do I need it? I already have everything I need,” I thought. Now, I don’t even have the Chrome app on my phone.

For many tasks, ChatGPT has replaced Google for me — especially the complicated ones. When I’m trying to understand a topic, compare options, summarize information, or get oriented quickly, AI feels dramatically better than traditional search. Instead of opening ten tabs, scanning long pages, judging credibility, and synthesizing everything myself, I can ask a question and get a structured answer.

Hallucination is still a real issue, so AI doesn’t remove the need for judgment. But the interaction model is different: I can interrogate the answer. I can ask for sources, counterarguments, examples, or a simpler explanation. Instead of hunting alone, I can pressure-test the answer in conversation.

That is AI’s real advantage. It does not just retrieve information. It helps shape, compress, compare, and extend it. Search gives me places to look. AI gives me a starting point for thinking.

Andy Crestodina’s research makes a similar point: AI is not replacing search evenly. People are more likely to use AI when they need more information, more synthesis, or more help making sense of something. Search still wins when the goal is local, transactional, navigational, or exploratory.

Search habits are sticky

My own process of getting to know ChatGPT was an interesting user experience moment. The UI wasn’t that different from what I was used to, but it felt like meeting a new person for the first time — a little awkward, and not knowing what to say. It felt strange to type only keywords or short phrases because the interaction looked like a text conversation.

That’s when I noticed how deeply habitual search behavior is. Once people find a reliable way to get information, that behavior becomes muscle memory. Even within search results, users develop habits. Many people skip sponsored results automatically. They may not know exactly why, but they have learned that skipping ads usually gets them closer to what they want.

The cold, unfriendly chatbox

When people first encounter an AI chat box, it can feel oddly unwelcoming. Am I supposed to chat with it like a person? Ask a question? Give instructions? How much context should I include? What counts as a “good” prompt?

This can feel even more daunting because of all the discourse around prompting. It almost feels like meeting a famous person you have heard so much about — especially the parts about how difficult they are to interact with — and then, when you finally try to talk to them, they just stare back without blinking.

This is where AI UX still feels immature. The challenge is no longer just capability. It’s perception. It’s discoverability.

People understand the box. They do not yet understand the range of behaviors the box supports. And because the interface resembles both search and messaging, the interaction can feel awkward — like trying to talk to a powerful new friend without knowing the rules of the conversation.

The Google parallel

Google trained users to stop browsing portals and start typing what they wanted. AI asks for an even bigger shift: stop translating intent into keywords and start expressing the intent itself.

When Google removed the portal scaffolding people were used to and presented a single search box, the message was clear: forget clicking around; all you need is search. Yahoo let people browse. Google asked people to type what they wanted.

I’m sure that created a kind of blank-canvas anxiety for some users. But Google could get away with it because the reward was immediate and visible: better, faster, superior search results.

AI has a similar but harder challenge. Google’s box asked: Search or navigate?

AI’s box asks something broader: What do you want to know, think through, create, compare, change, or do?

That is a much bigger behavioral leap.

The browser may not stay search-first

For years, the browser has assumed that users begin by navigating: type a URL, search a keyword, open a tab, browse pages.

But if more information-seeking starts with AI, that default may change. The omnibox may become less of a search field and more of an intent field.

As we go through this transition, the omnibox presents a major challenge. A single box that handles URLs, search, summaries, writing, commands, and task delegation is powerful — but also ambiguous. It needs more guidance. It needs stronger affordances.

The browser that wins this transition won’t be the one that simply adds AI to the search box. It will be the one that knows when the user wants a page, when they want an answer, when they want to explore, and when they are ready to hand something off.