Google has a new message for advertisers: the shape of search has changed, not just its volume. On October 28, 2026, the Google Ads Decoded podcast published a 25-minute episode titled "How AI is changing Google Search behavior and what it means for advertisers." Shivani Mohan, Search Vice President of Data Science and User Experience Research at Google, told host Ginny Marvin, Google's Ads Product Liaison, that query length "has increased 3x" and that "one in every six searches is multimodal." Marvin closed with the detail advertisers should care about most: ads in Google's AI search surfaces are matched to "both the query and the content of the AI response," and retailers were pointed toward six optional product data fields.
That is a real shift in how Google talks to the ad side. But there is a gap between how those two headline figures were stated on the podcast and how Google itself published them five months earlier, and that gap matters for how much of your account you should restructure in response. Here is what was said, what it rests on, and what we would actually change.
What Google told advertisers
The episode was unusual for the format. Ads Decoded launched on August 29, 2025, and its recent editions were product talks: a Demand Gen product manager calling $100 a day the floor for a campaign on October 1, and a Google Ads product manager on October 15 saying budget changes cause no learning period. This one carried no product news at all. It was claims about how people search, followed by Marvin's reading of what those claims mean for an ad account.
Mohan leads data science and user experience research for consumer-side Google Search, and she framed the shift as an "expansionary moment." Her account: people always had questions, but only a small fraction could be compressed into search-engine keywords. With AI Overviews and AI Mode, that translation step is gone. "We're seeing that people are asking much longer, much more detailed questions," she said, adding that "we can see that the query length has increased 3x." People are also asking "net new types of queries that they would have never thought to ask before," following topics across several turns in a stateful, conversational way, and asking questions about images and video rather than text alone. The result, in her words, is that "people are just asking more questions because it is much more natural and frictionless."
The second headline came near the end, in answer to Marvin's question about what an agency lead should change the next morning: "One in every six searches is multimodal."
Marvin repeated both figures in her closing summary: "with queries in conversational search running three times longer on average and one in six searches being multimodal, users are relying on Google for discovery well before they are reaching their final purchase decisions."
The May report told a narrower story
Here is the part that should temper a Monday-morning restructure: neither figure on the podcast came with a market, a time period, or a baseline. Google published both numbers before, with qualifiers attached.
The May 19, 2026 report marking one year of AI Mode, authored by Mohan herself, put the average AI Mode query at three times the length of a traditional search query. It said more than one in six AI Mode searches in the United States were multimodal. And it reported that follow-up queries in AI Mode had grown by more than 40 percent a month on average in the US. In other words, in May, both figures described AI Mode, and the multimodal share was limited to the United States.
On the October podcast, "one in every six searches is multimodal" arrived without the AI Mode or US qualifiers, and the 3x figure was placed by Marvin in "conversational search." As PPC Land's November 1 report on the episode noted, a share of AI Mode searches in one country describes a single opt-in surface, not Google Search as a whole, and nothing in the episode supports the broader reading. A share of AI Mode searches is a meaningful number. "One in six searches" as the sentence reads is a bigger one.
This is not an accusation of fabrication. Mohan described on the episode how Google tests its published numbers for repeatability and durability, and said figures are checked "very strongly" before release. But those tests govern how a number is produced, not how it gets restated afterwards. When a stat loses its qualifiers between May and October, that is marketing drift, not new data, and advertisers pricing decisions off the broader phrasing should know which version they are buying.
Independent signals point the same way, at smaller scale
None of this means the behavioral shift is invented. Other measurements point in the same direction, if not at the same scale:
- Robby Stein, VP of Product for Google Search, said in July 2025 that visual searching on Google had grown 65 percent year over year.
- The State of Search report from Datos and SparkToro for the fourth quarter of 2025 found the fastest US growth in searches of six to nine words.
- Since September 24, 2026, Google Search Console splits web results into text-based and multimodal search types, though the queries behind multimodal searches are not shown to site owners.
- Search Live, Google's voice and camera mode, has been available to all US users in English since September 24, 2025, removing friction from voice and image queries.
Put together, the direction is consistent: questions are getting longer, less keyword-shaped, and more visual. The honest summary is that the trend is real, the podcast numbers are the trend stated at its most flattering, and the true whole-of-Google figures remain unpublished.
Five practical moves for advertisers
Given all of the above, what would we actually change in an ad account? Five things, in rough order of effort.
- Write ads for questions, not keywords. If queries are getting longer and more natural, ad copy assembled from two-word keywords will match worse than copy that answers a full question. Draft headlines that read like the first sentence of a good answer.
- Fill in the product feed, all of it. Google pointed retailers on the episode toward six optional product data fields, and said AI Mode ads are matched to the content of the AI response, not only the query. In practice that means feed completeness now functions as ad targeting. Attributes that felt optional in classic Shopping are becoming eligibility criteria in AI surfaces.
- Get visual assets into AI-ready shape. If a meaningful share of searches start with an image or the camera, product imagery and video stop being creative garnish and become matchable input.
- Split text from multimodal in your reporting. Since September 24, 2026, Search Console separates the two. Even without query-level visibility for multimodal clicks, knowing which surface produces your traffic tells you which format to invest in.
- Feed the answer layer, not just the ad layer. Ads in AI Mode are matched to the content of the AI response. That response is built from content somewhere, often not yours. Structured, citation-friendly pages on the questions your buyers actually ask are how your brand shows up inside the answer your ad sits next to.
Our read
Google is telling advertisers the truth in a sales voice. The behavioral shift is real and worth budgeting for; the specific numbers are AI Mode figures presented as though they described all of Search. The advertisers who lose are the ones who hear the pitch and either ignore it entirely or rebuild their accounts around a stat that still, as of today, only describes a subset of one surface in some markets. Adjust for the direction, discount the decimals.
If your paid and organic programs are still built for ten-word queries and text-only intents, this is the quarter to fix that. We build content and measurement systems for exactly this shift. Start with The Vivid Company.
