Google: AI Queries 3x Longer, 1 in 6 Searches Are Multimodal

If you still picture Google users typing a few keywords and hitting Enter, that image is outdated. In a Google Ads podcast episode, a leader from Google's search behavior research team stated that query lengths have tripled, and one in six searches is multimodal—meaning people search using images, video, or voice, not just text.

This is not a product launch news update. It is a signal about how human question-and-asking behavior is shifting, and consequently, how ads are chosen for display. For marketers, this is a moment to pause and re-examine familiar assumptions about keywords, content, and the customer journey.

What Was Actually Said in the Podcast

The October 28 episode, lasting about 25 minutes, featured Shivani Mohan—who leads the consumer data science and user experience research team for Google Search—in conversation with host Ginny Marvin. The topic: how AI is changing search behavior.

The notable point is that this episode did not announce new products. It simply offered observations on behavior, and Marvin interpreted their implications for advertising accounts. Mohan shared that she started her career in industrial design—once designing bicycles—where there is “only one chance to get the user right,” and later divided research into two halves: one-on-one interviews to understand why people ask questions in certain ways, and data science to quantify observed behaviors on the platform.

Three behavioral changes were noted:

• Longer, more detailed questions, not just replacing old keywords but acting as “entirely new” queries that people previously wouldn’t have thought to ask.
• Asking about images or video—something Mohan noted was much harder to do previously.
• Multi-turn conversations, maintaining context flow across sequential follow-up questions.

Why People Ask Longer Questions: Friction Has Disappeared

Mohan’s explanation was quite straightforward. People have always had countless questions, but previously only a small fraction of those could be translated into keywords to type into a search box. The step of “translating” curiosity into a few keywords has practically vanished, so questions are now spoken more naturally, at greater length, and with more context.

She also gave a very relatable example: a user asks for the weather in San Francisco, and then follows up with “What about Austin?”. The second question is almost meaningless on its own, yet carries the context of the previous one. It is precisely this kind of sequential querying that makes old measurement methods mismatched.

Notably, Mohan stated that her team “penalizes” queries that require manual adjustments, as that is a sign the initial answer missed the user's intent. She asserted that people return to search more often out of expanded curiosity, not because the first attempt yielded no results. However, the podcast episode did not provide the proportion of follow-ups classified as “adjustments” versus entirely new questions.

Two Statistics Spoken Beyond the Original Data

This is the part worth reading slowly. In the podcast, Mohan said “one in six searches is multimodal” and Marvin reiterated “one out of every six searches.” Neither specified a market, a timeframe, or a comparative baseline for the 3x increase figure.

Google’s May 19 report—also published by Mohan—was much more specific: the average query in AI Mode was three times longer than traditional search queries; over one-sixth of AI Mode searches in the U.S. were multimodal; and follow-up queries in AI Mode grew by over 40% month-over-month. That behavioral data was tied to U.S. users.

A segment of AI Mode in a single country is the characteristic of a specific surface that users actively step into. Read without modifiers, it becomes a story about all of Google Search—and no data in the podcast backs up such a broad interpretation.

Why does this distinction matter to media buyers? Because it determines the scale of opportunity. If long queries are a feature of a new surface, you need a distinct strategy for that surface. If it applies to all of Google Search, you must rewrite your entire content system. Two different conclusions lead to two different budget allocation approaches.

How Ads in AI Mode and AI Overviews Are Matched

This is the most practical quote in the episode. Marvin stated that ads within AI search surfaces are matched with both the query and the content of the AI answer, encompassing the entire context, and advised advertisers to position their brands as the “natural next step” in that journey.

Practically speaking: if the system reads the entire AI answer, then landing page quality, product information clarity, and contextual content relevance matter far more than stuffing extra keywords. A landing page that correctly answers the follow-up question a user just asked will hold an advantage over a page that merely repeats the initial keyword.

• Review landing pages currently running ads: do they answer follow-up questions, or do they just pitch a sale?
• Add comparative context, specifications, and use-case scenarios—the things users frequently ask about next.
• For retail, update product data in Merchant Center, including the conversational attribute group introduced by Google. Marvin mentioned “six optional product data fields,” but the podcast did not name these six fields, nor did it state whether they affect ad eligibility or exclusively influence product answers.

Users Still Want Sources, But Click Data Remains Concerning

Mohan suggested that users remain very clear about wanting to hear “authoritative voices and real human perspectives.” She described four types of needs: wanting to see origins to build trust; wanting to compare products through podcasts, review sites, and peer experiences; wanting instructional videos for how-to questions; and shifting from generic content to in-depth content.

However, the demand for sources does not equate to people clicking on those sources. External metrics paint a less cheerful picture: a study by Saharsh Agarwal and Ananya Sen noted that AI Overviews reduce organic clicks by 39.8%; Ahrefs reported a 58% drop in click-through rates for top-ranking pages when AI Overviews appear; and Chartbeat data placed Search at roughly 12% of North American publisher pageviews, down from 21% a year prior.

I believe we should hold both truths simultaneously: people genuinely want trustworthy sources, and the path from an AI answer to a website is growing shorter. The content creator’s job is to become an irreplaceable source—owning proprietary data, distinct experiences, and unique perspectives.

Four Actionable Steps for This Week

1. Rewrite content around real questions. Gather common customer questions from chat, hotlines, and comments, then turn them into subheadings in your articles.
2. Audit product data structures. Specifications, materials, use cases, target audiences—elements that serve comparative questions.
3. Re-measure by surface. Separate reports for regular search and AI surfaces to understand where your money is going, rather than relying solely on last-click attribution.
4. Do not neglect video and podcasting. If users seek out these formats to hear “real human voices,” they are content channels well worth investing in.

An Unresolved Question

There is one detail that makes it difficult for me to piece together a seamless picture. Mohan said queries are growing longer, yet Adthena’s report on 29.1 million queries indicated that over 60% of ad impressions in AI Overviews occurred on 3-to-4-word queries. Short, not long. So are ads chasing new behavior, or clinging to lingering old behavior?

The podcast also offered no performance data for ads across the two AI surfaces: no click-through rates, no conversion rates, and no cost comparisons against standard search engine results pages (SERPs). That is a gap that anyone managing a budget should fill with their own small-scale tests.

What I believe is: there will be no magic button that instantly transforms your ad account into “AI-ready.” There are only small, repeated adjustments—clearer content, richer product data, and more honest measurement. It may sound slow, but that is the part you can control.

As for you, are you noticing your customers asking longer questions, or are they still typing two or three words and clicking the first result? If you believe long queries are merely a quirk for a tech-savvy niche, speak up—I would love to hear a contrasting perspective. And if this article gave you one more thing to do this week, drop a comment and let me know which task you are tackling first.