> ## Content Index > Fetch the complete content index at: https://www.sayeric.com/llms.txt > Use this file to discover other available public pages before exploring further. # Agentic Commerce is going to see more "lean-in" behaviors from consumers - URL: https://www.sayeric.com/agentic-commerce-is-going-to-see-more-lean-in-behaviors-from-consumers/ - Published: 2026-09-09T22:29:39.000Z - Updated: 2026-09-13T03:15:40.000Z - Description: A sourced, chart-by-chart look at how human behaviors are shifting to be more engaged with the devices since AI chat, live shopping, and AI search assistants took off. - Author: Eric Cheung Search queries, AI prompts, and live shopping chats all point the same direction, but not the direction most people think. Here's the sourced version, including the counterargument that usually gets left out. In 2014, Baidu's chief scientist Andrew Ng predicted that within five years, half of all searches would happen through images or speech instead of text. It didn't happen. Voice search leveled off somewhere around 12 to 15 percent of queries and stayed there. The keyboard survived a prediction that should have killed it. Three years into the LLM era, a version of that story is going around again, except the claim now runs the other way: people are typing more than ever. Longer prompts. Longer queries. Whole sentences into a chat box instead of three words into a search bar. Live shopping chats supposedly full of typed questions. Everyone writing about agentic commerce, myself included, has a stake in some version of this being true. So I went and checked. The problem is that "people are typing more" is actually three different claims stacked on top of each other, and they don't hold up the same way: 1. More people are typing natural language somewhere. 2. Each person is typing more, in total. 3. Each thing they type is longer and more expressive than what they used to type. Start with the simplest comparison across those three claims: how many words does it actually take to ask for something, depending on where you're asking. Words per surface, typical query Google's own numbers, plus Perplexity's and a range estimate for LLM prompts Google's traditional query length has barely moved in a decade. Everything built since 2023 sits well above it. The AI Mode figure is inferred from Google's own stated 3x multiplier over a traditional query, not a disclosed absolute number, so treat it as an estimate rather than a platform-confirmed figure. ## The search box didn't change. A new box showed up next to it Here's the part that should worry anyone who's spent a career optimizing for the keyword box, including me. I spent fifteen years in this industry. The box itself barely budged. Similarweb's tracking of the average US Google query held flat at 3.33 to 3.36 words from mid-2024 through mid-2025\. Then it started climbing, reaching roughly 3.5 words by spring 2026, right as AI Mode scaled. Google made the shift explicit instead of leaving people to guess at it. In May 2026, Google's VP of data science, Shivani Mohan, put a number on the gap: the typical AI Mode search runs about three times longer than a typical Search query. The same post included a detail worth noticing on its own: more than one in six AI Mode searches now involve something other than typed text, whether that's an image, a voice clip, or a live back-and-forth. Image-based AI Mode searches were growing more than 40 percent month over month. Translation for anyone who's done keyword research for a living: the work wasn't wrong. It was measuring a box that was never going to get much bigger. The new demand didn't replace that box. It showed up next to it, already three times longer. Average Google search query length, US Words per query, Similarweb, selected months Flat for a full year, then a rise concentrated in the window where AI Mode usage scaled. These are the specific months Similarweb published a figure for, not a smoothed monthly series, so the line between points is a straight interpolation, not measured data. 💡 The chat box is doing enormous typing volume. Most of it is short. ChatGPT processed roughly 2.6 billion messages a day by June 2025, up 5.8 times from about 450 million a year earlier, according to OpenAI's own research released with the National Bureau of Economic Research. Weekly active users went from 100 million in November 2023 to about 900 million by early 2026. Here's where it gets more interesting than the raw growth number. The median ChatGPT prompt is short: around 13 tokens, roughly ten words, based on an analysis of real usage logs. The bigger averages you'll see quoted elsewhere are pulled up by a small number of people pasting in code or long documents. Most actual prompts look like a slightly longer, more specific version of a Google query, not an essay. That's not a knock on the trend, it's the honest shape of it. People aren't writing more for the sake of it. They're finally able to write down the qualifiers they used to leave out because the box wouldn't take them. ChatGPT weekly active users Millions, OpenAI-disclosed milestones Nine-fold user growth in about 26 months. Message volume grew faster still (5.8x in one year alone), which is exactly why claim 2, "each person typing more," needs its own scrutiny rather than a pass from the headline number. ## Retail's AI assistants make the point better than a chatbot does This is the part I watch most closely, because it's my actual job. Amazon's Rufus (now Alexa for Shopping) was fielding around 274 million queries a day by late 2024, and by Q1 2026 its user base was up roughly 115 percent year over year with engagement up close to 400 percent. Amazon and the analysts tracking Rufus describe the queries the same way every time: longer, more specific, carrying constraints a keyword box was never built to hold. "Running shoes" becomes "best running shoes for flat feet under $100 with good arch support" the moment there's somewhere for the second half of that sentence to go. That second half is the whole thesis. Keyword demand didn't disappear. It was always a compressed version of a longer sentence, and most of us building search and marketplace products for the last twenty years only ever saw the compressed version, because the box only accepted the compressed version. Give people a box that takes the whole sentence, and a meaningful share of them will use it. ## Live shopping chat: real in China, still forming in the US 0:00 /0:10 1× eBay Live: GMV grew roughly 8x year over year in Q2 2026 across seven markets, per CEO Jamie Iannone on the earnings call. Buyers who shop Live spend 70 percent more than comparable buyers who don't, and about half of that extra spend happens elsewhere on the marketplace, not during the stream itself, a real halo effect. Sellers who stream regularly see GMV grow more than 90 percent of the time, and Live sellers grow roughly three times faster than comparable sellers who don't use it. eMarketer puts the broader US live-shopping market at roughly 20 billion dollars in 2026, up about 35 percent year over year, and TikTok has told reporters its US live sales more than doubled in the first half of 2026 alone. eBay Live GMV growth, year over year Multiple vs. same quarter last year, as disclosed on eBay earnings calls Three straight quarters of roughly 7x to 8x year-over-year growth. eBay has not disclosed Live's absolute GMV, revenue, or buyer and seller counts, so read the multiple as directional, not as a scale claim, especially since five of the seven markets had been live for less than a year as of the Q2 2026 disclosure. To be frank though, no platform, eBay included, publishes a clean, comparable time series of typed comments per viewer. What exists is fragmentary: peak TikTok streams reportedly see 500-plus comments a minute, and China's live-commerce audience passed 765 million users back in 2023 with typed "barrage" comments core to the format. The GMV growth above is real and well documented, at least directionally. The typed-engagement claim is directionally plausible and structurally under-measured. I'm not going to manufacture a chart to make it look more complete than it is. ## Now the honest counterargument: the camera is growing faster than the keyboard If I stopped here, this would be a one-sided piece, and one-sided pieces are exactly what got the "voice will hit 50 percent" prediction so wrong in the first place. So look at what's actually growing fastest inside Google, the same company behind the "three times longer" claim above. Google Lens went from about 3 billion visual searches a month in 2021 to roughly 20 billion by late 2024, growing 65 percent year over year through early 2025\. That's a faster growth curve than anything in the typed-query numbers in this piece. Remember that one-in-six stat from AI Mode? Non-text input is the fastest-growing part of Google's fastest-growing product. There's a version of the counterargument that goes further than "voice and image are also growing." It's this: the long, effortful, ten-plus-word prompts everyone (including this piece) is citing right now might be a transitional artifact, not a destination. Agentic commerce, which is the category I work in, is explicitly trying to make one short instruction do the job that used to take a paragraph of typed constraints. If that works, the honest prediction isn't "people will keep typing more." It's "people will type a little, forever, and an agent will type the rest on their behalf." That's a real possibility, and anyone citing this data, including me, should say so instead of only quoting the numbers that flatter the thesis. Google Lens visual searches per month Billions, Google-disclosed, the fastest-growing input Google has Roughly 6.7x growth in three years, faster than the typed-query growth rates cited earlier in this piece. The strongest evidence against a pure "typing wins" story sits inside the same company making the case for it. **Claim 1:** More people are typing natural language somewhere **Verdict:** Holds up **Evidence:** ChatGPT: 100M to 900M weekly users in about two years. Amazon Rufus: user base up 115% year over year in Q1 2026 alone. --- **Claim 2:** Each person is typing more, in total **Verdict:** Doesn't hold up **Evidence:** Most of the volume growth tracks user growth. Dividing messages by users instead of looking at totals removes most of the drama. --- **Claim 3:** Each thing they type is longer and more expressive **Verdict:** Holds up, and it's the real story **Evidence:** AI Mode queries run 3x longer than traditional Search. Rufus queries are longer and more specific. Perplexity's median query is roughly triple Google's. --- I write something substantial a couple of times a year, plus an annual shopper survey on questions like this one. ## Sources - [eBay Q2 2026 Earnings Call Transcript, Aug 6, 2026](https://www.theglobeandmail.com/investing/markets/markets-news/Motley%20Fool/3815693/ebay-ebay-q2-2026-earnings-call-transcript/?ref=sayeric.com) - ["eBay Live Touts 8x Growth, But Scale Remains Hard To Measure," Aug 2, 2026](https://www.valueaddedresource.net/ebay-live-8x-growth/?ref=sayeric.com) - [Shivani Mohan, "AI Mode: making search more personal, visual and conversational," Google, May 19, 2026](https://blog.google/products-and-platforms/products/search/ai-mode-us-insights/?ref=sayeric.com) - [Chatterji et al., "How People Use ChatGPT," NBER Working Paper 34255, Sept 15, 2025\. (PDF)](https://www.nber.org/system/files/working%5Fpapers/w34255/w34255.pdf?ref=sayeric.com) - [Similarweb, US average Google query length tracking, 2024 to 2026](https://officechai.com/ai/google-search-queries-have-become-longer-since-the-introduction-of-ai-mode-data-shows/?ref=sayeric.com) - [Whatnot, "2026 State of Live Selling Report"](https://blog.teamwhatnot.com/unitedstates/2026livesellingreport?ref=sayeric.com) - [eMarketer, US live-shopping market forecast, via CNBC, Sept 1, 2026](https://www.cnbc.com/2026/09/01/tiktok-whatnot-livestream-shopping.html?ref=sayeric.com) - [Amazon Rufus usage disclosures and Canopy Management analysis, 2025 to 2026](https://canopymanagement.com/generative-search-amazon-optimization-guide/?ref=sayeric.com) - [Google, Google Lens usage disclosures, Google I/O and Google blog, 2021 to 2025](https://blog.google/products-and-platforms/products/search/google-search-lens-october-2024-updates/?ref=sayeric.com) - [Backlinko / DataForSEO, analysis of 306M keywords, keyword length distribution](https://backlinko.com/google-keyword-study?ref=sayeric.com) - [ThoughtTrace, prompt/response token length analysis on public conversation logs, arXiv 2605.20087, 2026](https://arxiv.org/abs/2605.20087?ref=sayeric.com) - [Zhao et al., "WildChat: 1M ChatGPT Interaction Logs in the Wild," ICLR 2024](https://arxiv.org/abs/2405.01470?ref=sayeric.com) - [Andrew Ng, Baidu, on voice and image search adoption, Fast Company, 2014; comScore denial of the "50% by 2020" figure](https://www.fastcompany.com/3035721/baidu-is-taking-search-out-of-text-era-and-taking-on-google-with-deep-learning?ref=sayeric.com)