--- title: "Where ChatGPT fits in the customer journey vs Google" description: "Later than Google: in a 2026 US and British panel, search usually opened the journey, while people more often browsed before using an AI assistant than after." canonical: "https://underneath.agency/resources/chatgpt-vs-google-customer-journey" published: 2026-10-08 updated: 2026-10-10 publisher: "Underneath (https://underneath.agency/agent)" entity: "https://underneath.agency/.well-known/entity.json" --- Guide · AI search # Where does ChatGPT fit in the customer journey compared with Google? ChatGPT usually sits later in the journey than Google, not at its start. In a 2026 panel of US and British web users, people tended to read web pages after a Google search but before opening an AI assistant. That suggests what buyers read on the open web may travel with them into the AI conversation, though no study has yet proven that link. ## The short version 1. Web pages followed a search in 60.2% of search sessions but came before the assistant in 45.5% of assistant sessions (Iannelli and Ai, 2026). 2. Comparing the same people, search and assistants differed by 20.6 percentage points in this before-and-after pattern, and by 20.3 a month later. 3. Just over half of assistant sessions (51.9%) showed web activity before the first prompt, and a search came before it in 33.8%. 4. Assistant sessions ran longer than search sessions, a median of 22 minutes against 9, and switched between the assistant and the web 5.2 times on average. 5. ChatGPT also reads the web for itself: in our study it ran 3.7 searches per buyer question before answering. ## Does ChatGPT replace Google at the start of the journey? Mostly not: search tends to open the journey, while AI assistants sit deeper inside it. That is the central finding of [Iannelli and Ai](https://arxiv.org/abs/2607.04282), who linked an opt-in panel’s AI assistant activity to the same people’s searches and page visits. The panel covered the United States and Great Britain in February 2026. The assistants were standalone ones such as ChatGPT and Gemini, not Google’s AI Overviews. Two cautions come first. The authors work at Scrunch AI, a commercial company, so this is vendor research. The paper was also under review and not yet peer reviewed when we read it. The table compares where web page visits fell, counting each person equally. | Session centered on | Pages visited before | Pages visited after | |---|---|---| | A Google or other search | 46.7% | 60.2% | | An AI assistant | 45.5% | 38.9% | Around search, page visits mostly came afterward: a query, then the results it led to. Around assistants the order flipped. In the authors’ words, “search tends to open the observed journey, while conversational assistants sit deeper inside it.” That is one reason the evidence does not show [ChatGPT replacing Google search](https://underneath.agency/resources/is-chatgpt-replacing-google). The gap between the two patterns, measured within the same people, was 20.6 percentage points. It held at 20.3 points when the team repeated the analysis on March 2026 data. ## What do people do before they open an AI assistant? Often they have already been browsing; just over half of assistant sessions had web activity before the first prompt. Counting each person equally, 51.9% of sessions showed some web activity before the first prompt. A search engine visit came before the first prompt in 33.8% of sessions and after the last response in 24.4%. Search appeared somewhere in 46.0% of assistant sessions. Sessions with web activity only before the assistant (18.3%) outnumbered those with web activity only after it (10.5%). The popular picture of “ask the AI first, then click out” describes roughly one session in ten. It fits other evidence that [people click fewer websites with AI answers](https://underneath.agency/resources/ai-answer-engines-fewer-website-clicks). The authors are careful about what timing can show. Their data held no page content and no prompt text. So it cannot prove that the pages someone read were why they opened ChatGPT, or that both belonged to one task. ## How is an AI assistant session different from a search session? It lasts longer and mixes more activity: a median of 22 minutes, against 9 for a search session. Inside assistant sessions, prompts and responses made up 59.7% of recorded steps, searches 8.8% and page visits 31.5%. People switched between the assistant and the web 5.2 times on average, against 4.0 switches in search sessions. A search session, in plain terms, is a quick hop from a query to a few pages. An assistant session is a longer working conversation that dips in and out of the web. The authors describe AI assistants as a possible “synthesis layer deeper in the journey,” while search “remains an access layer.” They stress these are broad patterns, not labels proven for each session. ## Does the pattern hold when people research purchases? Yes, within the site categories the researchers could check, shopping included. They grouped sessions by the kind of sites people visited and found the same reversal in every group. It measured 19.8 points for shopping sites, 21.1 for news and 26.8 for reference sites. This is a rough check, not a study of buying journeys. Only 22.5% of assistant sessions visited sites that could be put in a category. Research on Google points the same way for AI answers in general. [Grossman and colleagues](https://arxiv.org/abs/2604.27790) ran 11,500 queries through Google in December 2025. An AI Overview, the AI summary at the top of Google’s results, appeared on only 17.4% of plain product searches. It appeared on 88.2% of product comparison searches and 92% of product questions. Those two sets were written by AI from real shopping searches, so treat them as a test, not real traffic. The authors conclude that AI search plays a larger role in the research stage than at the final purchase. ## Does what buyers read earlier shape what the assistant tells them? Possibly, but no study has shown it yet. What we do know is that the assistant reads the web itself. In [our hidden-searches study](https://underneath.agency/research/ai-hidden-searches-study), ChatGPT ran a mean of 3.7 searches per buyer question before answering. None of the 509 searches we collected repeated the user’s question word for word. The pages it cites are often not the ones Google ranks highly. In [our rankings study](https://underneath.agency/research/ai-citations-google-rankings-study), 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the question as typed. So a buyer can meet your brand on Google first, then meet a different set of sources inside ChatGPT. ## What should you do about it? Treat AI assistants as a research step in the middle of the journey, not as its front door. Practical steps: 1. Keep investing in search. Search still opens most observed journeys, so your Google presence shapes what buyers see before they ask an assistant. 2. Publish the material buyers compare. Comparison and question searches drew AI answers far more often than plain product searches, so pages that answer “how does X compare with Y?” matter. 3. Check what assistants say about you. They run their own searches and cite pages that often do not rank on Google, so ask them the questions your buyers ask. 4. Measure the whole journey. A buyer who read your page last week may [never click through from ChatGPT](https://underneath.agency/resources/ai-assistant-sessions-without-website-visits), so watch branded searches and direct visits, not only AI referrals. If you want help mapping where AI assistants enter your buyers’ research, see our [generative engine optimization service](https://underneath.agency/services/generative-engine-optimization). ## What does the research not tell us yet? The research shows the order of events, not their cause or the buyer’s intent. The main gaps: - Whether earlier reading shapes the prompt. The panel study kept no page content or prompt text, so the link is a reasonable guess, not a finding. - Business buyers. The panel was an opt-in group of web users that the authors call AI-forward, not a sample of B2B purchasers. - Mobile apps. Assistant use in native mobile apps and through APIs was not observed. - Time and place. The data covers two months, February and March 2026, in two countries. - Google’s own AI. AI Overviews and AI Mode were outside the panel study, because they sit inside the results page. ## Frequently asked questions ### Do people use ChatGPT before or after Google? More often after browsing than before it, in the one large study that tracked both. In a 2026 US and British panel, page visits came before the first assistant prompt in 45.5% of sessions and after the last response in 38.9%. ### Is ChatGPT replacing Google search? The data does not show replacement. Search appeared somewhere in 46.0% of assistant sessions, and the authors describe coexistence while saying their design cannot settle whether AI reduces search overall. ### Where in the buying process do AI answers show up most? At the research stage. In one 2025 test, Google showed AI Overviews on 88.2% of product comparison searches but only 17.4% of plain product searches. ### Does ChatGPT find the same pages as Google? Often not. In our study, 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the same question. ## Sources - Iannelli, M. and Ai, A. (2026), [The New Shape of Search: How Conversational AI Recomposes Information Seeking](https://arxiv.org/abs/2607.04282), arXiv:2607.04282. - Grossman, R., Liu, S., Chen, M. K., Smith, M., Borcea, C. and Chen, Y. (2026), [How Generative AI Disrupts Search: An Empirical Study of Google Search, Gemini, and AI Overviews](https://arxiv.org/abs/2604.27790), arXiv:2604.27790. - Underneath (2026), [The hidden searches AI assistants run before they answer](https://underneath.agency/research/ai-hidden-searches-study) - Underneath (2026), [Do ChatGPT, Gemini, Perplexity and Claude cite pages that rank?](https://underneath.agency/research/ai-citations-google-rankings-study) --- This is the Markdown twin of https://underneath.agency/resources/chatgpt-vs-google-customer-journey. The HTML page is canonical. Publisher: Underneath, https://underneath.agency/agent. Site index: https://underneath.agency/llms.txt.