The crossover: What happens when Google Search and LLMs meet?

The disruption of traditional search by LLMs has been one of the biggest shifts and talking points of the last few years. To cite just a few changes during this time: ChatGPT added real-time web search, Google made AI Overviews the default, and GEO became a marketing term that didn’t even exist four years ago.

But has consumer behavior changed? For this analysis, we looked at what happens when people use LLMs and Google in the same sitting, and what that means for brands considering where to show up in a search journey that contains more than one platform.

How common is LLM/Google crossover behavior?

Let’s start with the big picture: of our population of Android users searching Google on a browser in September, 27% used the ChatGPT or Claude apps at some point during the month and 12% used the ChatGPT, Claude, or Gemini websites at some point (we’re excluding the Gemini app throughout, since it’s triggered any time Google Assistant is used on Android, which is fundamentally different behavior to the other AI apps).

We’ll call a Google search and an LLM visit happening within 5 minutes of each other a “crossover.” Now, how many of those people had an LLM/Google crossover? For LLM apps, it’s 9% of all Google searchers (1 in 11), and for LLM websites, it’s 6% (around 1 in 17). So despite LLM apps reaching more than twice as many people, the app/web gap shrinks a lot on this crossover metric.

Among people who crossed over between Google and an LLM at all, nearly half did it just once in the month, and about 1 in 8 did it ten or more times.

Now zoom in to the level of a single search: how often does a search have an LLM crossover nearby? For any LLM (website or app), it’s 2.6% of searches, versus a normal rate of 1.7% — 1.6 times as likely. For LLM websites specifically, it’s 1.2% of searches. That may sound small, but compared to a normal rate of 0.5% for an LLM visit in any other active 10-minute window, it’s 2.4 times as likely. LLM apps show barely any lift by comparison: 1.5% vs. 1.2% normal rate, just 1.2 times as likely.

That gap is likely because our search data comes from the browser, not the app, so app-based journeys aren’t fully comparable here. For that reason, the rest of this analysis will stick to browser data only, which lets us compare like for like.

What those journeys look like: No single route

When people do move between Google and an LLM, what does that actually look like? Most crossovers are one-way: 65% go in a single direction, Google to an LLM or the reverse. Only around a third (35%) switch back and forth. And almost all crossovers (98%) use just one LLM.

Put in behavioral-science terms: most people here seem to be satisficers, not maximizers, when it comes to search – they settle on an answer and move on rather than comparing multiple sources to find the absolute best answer.

As far as where a crossover starts, it’s close to even: 52% start with Google, 48% with an LLM. But where they end shows a bigger difference: 60% end at Google, 40% at an LLM. This means that however people enter, they’re more likely to end at Google than at an LLM.

That lines up with another pattern: using an LLM doesn’t cut down how much people search in the moment. Grouping searches into “runs” (a run ends after a 5-minute gap), runs with an LLM nearby had 2.7 searches on average, vs. 2.3 without one. People using an LLM aren’t searching less – if anything, they’re searching slightly more.

What does this mean for brands? There’s no standard order to design around, but search is still happening, and crossovers tend to finish back at Google. That means search visibility is still crucial, even for LLM-driven traffic.

Everyday tasks over broad topics

What types of Google searches are more likely to have LLMs around them? Translation searches are 4.4x as likely to have an LLM nearby, and searches about creating or making things are 3.5x as likely. This is unsurprising: these are two tasks LLMs can do quickly. This makes these types of searches a competitive moment in the search/LLM battle: if a result is not right there in search results, people may be more likely to turn to an LLM.

Price-hunting searches (on varied topics such as travel, tech, food) also show a lift: 2.1x as common near an LLM. This is a small signal that commercial research is increasingly shared with an LLM. One interesting point: 87% of those price-hunting crossovers end with a Google search (vs. 60% of all crossovers). An LLM is more often a consideration step, but they still come back to search. For brands, this is yet another point of evidence that search visibility is still key.

Outside these three functional search types, topic is not a strong predictor of when an LLM will be included in a search journey. Tech and how-to searches are slightly more likely to show up in LLM-including crossovers; searches about local places and events less so. Crossover vs non-crossover searches also look the same structurally: similar word lengths, similar use of question words.

So task types distinguish crossovers more than broad topics or structural features do. If you’re trying to work out what types of searches will end up including an LLM, look for the task, not the category.

So what?

Our data suggests that things aren’t necessarily “one or the other” – people don’t use Google or LLMs, they use both – and that huge behavior shifts don’t happen en masse. Here’s what that means in practice:

1. Search visibility still matters even when people are using LLMs

2. No LLM owns the “paired with search” behavior more than another

3. Don’t assume one journey order; plan for people arriving either way

A note on the data

This analysis is based on Android users who searched on Google in their phone’s browser during September 2026, with the Gemini app excluded throughout (since it’s triggered any time Google Assistant is used, which isn’t the same as someone choosing to open an LLM).