Google Data: AI Search Users Have Moved Past Keywords, Your Content Hasn’t

In May, I argued that the new search user represented a genuine behavioral shift while Google’s product updates were largely noise. Now, Google has released a full year of data that confirms the transformation and quantifies it. The report, “How People Are Using AI Mode in the U.S.,” published May 19, 2026, on The Keyword blog, features Shivani Mohan, VP of Data Science & UXR at Google Search. She describes a searcher who no longer fits the persona most SEO teams built their 2025 strategies around.

According to the findings, the average AI Mode query is now three times longer than a traditional search query.

That single statistic invalidates a significant portion of what most SEO teams optimized for last summer. A year ago, the standard assumption for keyword strategies was that users type three to four words and then scan results. Google’s own data now shows that assumption describes only a minority of what AI Mode users are actually doing.

The User Has Evolved, But Content Has Not

The report covers the period from AI Mode’s U. S. launch in May 2025 through April 2026. Several other numbers paint a clearer picture. Follow-up queries in AI Mode have grown more than 40% on average per month, indicating that users are not landing on a single answer and leaving; they are staying in the conversation and digging deeper. Multimodal interactions now account for more than one in six AI Mode searches, meaning voice, image, or video input rather than typed text. Image-input searches have increased more than 40% month over month since launch.

The top five keywords in AI Mode searches are: 1. Information, 2. Identify, 3. Find, 4. Explain, 5. Summarize. The top five opening words are “what,” “how,” “I,” “is,” and “can.” Notice the third entry: “I.” People are narrating personal context into the search bar. Not “running shoes for flat feet.” Something closer to “I have flat feet and my knees hurt, can you help me find a running shoe that will not make it worse?” The Health and Wellness example in the report is even blunter: “I hate cardio. Give me a routine that avoids it but still works.”

That is not a keyword. That is a person talking to someone who might actually help them.

What the Content Gap Looks Like

The report organizes AI Mode behavior into five categories: Explore, Decide, Learn, Create, and Do. Brainstorming-related queries have grown 30% faster than the overall pace of AI Mode queries. Planning queries have grown 80% faster. Queries beginning with “which” have grown 40% faster over the past six months, suggesting that AI Mode has become a genuine decision-support tool for everyday purchases, not just a discovery layer.

This is the gap most content strategies have not addressed. Content built for a user who types [best running shoes 2025] and lands on a listicle does not serve a user asking, “I’m training for my first 5K and I’ve never bought running shoes before, which pair should I start with and how do I know if they fit right?” Both queries express shoe-buying intent. Only one of them describes what the AI Mode user is actually doing.

The practical problem is that most teams are still writing for the shorter query. They are optimizing page titles, meta descriptions, and H2 structures for three-to-four-word keyword targets that represent a shrinking share of how people actually arrive at answers.

3 Things to Do Differently Now

First, audit your top 10 pages against how a person would actually ask for that information in a conversation. Take the primary keyword for each page and rewrite it as a natural-language prompt the way a Google AI Mode user would actually type it. If your content does not answer the longer-form version of that question, it has a gap that a competitor who does will eventually fill.

Second, treat follow-up questions as a content signal, not an analytics footnote. The 40% monthly growth in follow-up queries tells you that users are not satisfied with a single answer. If you know what your site’s most common entry-point questions are, the follow-up question is now as strategically important as the entry point. For most sites, that follow-up question inventory does not exist yet.

Third, start preparing your visual assets for multimodal indexing. One in six AI Mode queries is already non-text, and image-input search is the fastest-growing query type in the system. Alt text written for accessibility and alt text written to serve a user who has photographed a product and is asking AI Mode what it is and where to buy it are different things. The image context around your product and informational assets needs to catch up to where the queries already are.

Google now has more than 1 billion monthly active users on AI Mode globally, and the platform’s query volume has doubled every quarter since launch. The behavioral shift I wrote about in May is no longer a forecast. It is a data set. The question for practitioners is not whether to respond to it but how quickly they can close the gap between the content they published last summer and the user who is searching right now.