Back to blog

How Artificial Intelligence and Evolving User Behavior Are Changing Search

How AI answers, longer queries, and changing audience habits are reshaping search intent, clicks, and the value of a loyal audience.

By Published Updated

Inflect cover reading "Attention is not action," with NOT in yellow; eye-shaped search signals enter a machine and one reaches a bell.

Introduction

The landscape of online discovery has splintered across multiple formats and ecosystems. While many professionals blame artificial intelligence summaries for falling referral traffic, the habit of finding an answer without visiting another website has been building for years rather than arriving overnight. AI is speeding up the change, but the larger question is where people now look for information and what they expect a search to help them do.

SparkToro's 2026 clickstream research found that 68% of United States Google searches ended without any click, compared with 49% in its 2019 study. The panels differ, but the two findings trace the same long-running habit of getting what one needs without leaving the results page. Ashley Liddell was already describing the broader change in 2022; her idea of Search Everywhere Optimisation recognizes that discovery is no longer confined to one search engine.

People also turn to visual and socially driven spaces such as TikTok, Instagram, and YouTube for answers that come with a demonstration, a recommendation, or a sense of who is giving the advice. SparkToro's study of search activity across 41 desktop websites confirms that meaningful search activity happens beyond conventional engines, even though Google remains the largest destination. Google, meanwhile, is answering more questions inside its own results. These places can serve different parts of the same inquiry: a video shows how something works, a search summary frames the basics, and a publisher's own work offers the detail or judgment a reader may still want.

Because user goals are scattered across content types and platforms, banking entirely on standard search results page visibility is no longer enough. Publishers and brands have to understand the communities, formats, and experiences in which a question first takes shape. That is not an argument for abandoning the website, but for making the work on it useful enough to be sought out after an initial answer has appeared elsewhere.

Shifting Behavior and Generative Responses

To grasp how audience habits are changing, we can look at Google's 2026 research on AI Mode. The feature has passed one billion monthly active users globally, and Google's analysis of United States usage shows a move toward conversational exploration. An average AI Mode question is three times the length of a conventional search. People ask it to "explain" or "summarize," sometimes beginning with "I" rather than reducing a personal need to a few keywords. Follow-up questions have risen by more than 40% on average each month in the United States, while more than one in six AI Mode searches now uses a non-text input.

The language matters more than the length by itself. A person who can describe the problem in full can also say what the first answer missed, ask for a different explanation, or bring an image into the conversation. The old search box encouraged people to compress a need into a phrase; this interface invites them to reveal more of it. For a writer, that fuller question is a clue to what needs explaining, not a prompt to add another paragraph of keywords.

Furthermore, the connection between fuller questions and generative responses appears in a study of AI Overviews in Spanish publishing. Clara Soteras and MJ Cachón, working with University of Barcelona researchers Mari Vállez and Carlos Lopezosa under the CUVICOM initiative, examined 2,699 search terms. Four-word searches brought up an AI Overview in 48.1% of the cases studied, and questions beginning with "why" did so particularly often. The finding is not that a four-word headline has some magic advantage, but that a more fully expressed need is more likely to be met by a generated explanation on the results page.

The topics behind those questions matter just as much. In the Spanish study, 75.2% of the queries that triggered an AI Overview concerned evergreen subjects, while "why" questions brought up an overview in 92.3% of the cases tested. These are often the questions a reader asks when a bare fact has failed them, wanting to know why something happened and what it means for them. A useful article can answer by laying out the causes, the evidence, and the parts that remain uncertain, giving the reader something more reliable than an answer that sounds complete because it is short.

Breaking news behaved differently in the same study, with AI Overviews appearing for 34.6% of the evergreen queries examined but only 1.1% of breaking-news searches. Almost half of the results showed neither an overview nor Google's Top Stories carousel. A developing story needs fresh reporting and updates that cannot always be assembled from an existing set of sources. That is why a publisher should not treat all search demand as a single AI-visibility contest. On an enduring question, interpretation may earn the visit; on a live event, being the source of the new information may matter more.

The same research shows why a familiar traffic ranking cannot settle the question of generative visibility. On searches that included each publisher's name, an AI Overview appeared 38.9% of the time for 20 Minutos and 11.1% for El Español, although El Español led the traffic ranking used by the researchers. Those figures describe when an overview appeared, not whether either publication was cited in it or received a visit. Even so, they show that reach and presence in an AI-shaped result are not interchangeable measures, especially when the wording of the reader's question changes what appears on the page.

What People Expect Search to Do

The familiar division of intent into informational, navigational, and transactional searches is less useful once a person can ask for help with a whole task. In Google's United States AI Mode research, people use the interface to compare options, plan travel, organize schedules, learn a subject, and make something new. Google groups those activities under five verbs: explore, decide, learn, create, and do. The verbs matter because they describe what the person hopes to leave with, not merely what kind of page a search engine might return.

Publishers have been asking a similar question for years through the User Needs framework described by FT Strategies, developed at the BBC by Dmitry Shishkin. It treats a news update as only one reason people turn to journalism: readers may also want perspective, education, inspiration, or help understanding what follows from an event. A publication that knows which need a story serves can choose the reporting and format more deliberately, rather than assuming the audience wanted another account of what happened.

Generative search is now trying to satisfy some of those needs before a reader reaches the publisher, raising the standard for the next step. If the obvious facts have already been supplied, an article needs the judgment, original reporting, or practical usefulness that makes it worth opening. The challenge is not to out-summarize a search summary; it is to do the work the summary cannot do for the reader who still wants to understand more.

Attention Is Not Action

Even when the results page offers that next step, being noticed is not the same as being chosen. A Spanish eye-tracking study by Laika followed 22 participants through desktop Google searches and recorded what they looked at as well as what they clicked. Everyone shown image results or product listings looked at them, yet nobody clicked those elements. The AI Overview's "Explore Further" control attracted attention without receiving a click. A visually prominent feature can interrupt a scan of the page without giving anyone a reason to act.

The elements that did receive clicks tell a more useful story, provided we are clear about where the click goes. Traditional organic results were clicked by 95.5% of participants who saw them; the main AI-generated answer block registered clicks from 86.4%, but interaction with Google's answer is not the same as a visit to a publisher. The source links inside that answer were clicked by 13.6%. This is a small observational study, not a forecast for every market, yet it makes an important distinction visible: somebody can notice an answer, use it, and still never reach the people who produced the underlying work.

For publishers, an answer may make a publication's name familiar, an organic link may bring a reader to the full story, and a useful story may bring that reader back without Google's help. These are different stages of the same relationship, so treating them all as "visibility" hides the difference between a passing glance and an audience that has begun to trust the source.

Rethinking Analytics and Audience Connection

Search volume and website visits still matter, but neither can describe that relationship alone. In Heiko Scherer's account of the shift from reach to belonging, a publisher must ask not only how many people noticed a story, but who comes back, who stays, and who cares about the publication behind it. Reach can introduce a reader to the work; it cannot tell us whether the reader found a reason to make the publication part of their habits.

Reciprocity offers a harder test of that relationship: page views and subscriptions record useful behavior, but they do not tell the whole story of whether an audience would recommend the publication, respond to a request, or support it without an immediate reward. These are difficult things to measure at scale, which is precisely why they are easy to ignore. They also reveal a kind of loyalty that a burst of referral traffic cannot establish on its own: a relationship in which the reader values the source, not just the answer they came to collect.

Organizations need to observe where their audiences' questions begin and build for the needs they find there. That may mean a sharper editorial offering, structured educational material, a community worth participating in, or a tool that helps people act on what they have learned. These ways of serving an audience sit alongside reporting and search visibility, giving readers a reason to remember who helped them, return when the next question arrives, and choose the original source even when a platform has already offered a summary.

Conclusion

Search is spreading across platforms and becoming more conversational, visual, and closely tied to what people want to accomplish. The studies here do not suggest that every answer will end in a website visit, or that every subject will be handled in the same way by AI, but they do show why publishers cannot rely on being seen alone. The strongest response is to understand the question behind the search, do the reporting or useful work that gives it depth, and make the source worth returning to after the first answer has faded.

Sources