This Double 11, AI Is Finally Busy, But Workers Are Anxious

This Double 11, AI Is Finally Busy, But Workers Are Anxious

As large language models enter their third year of competition, Double 11—an iconic consumer event—has become a new stage for big companies and startups to showcase their AI e-commerce capabilities every year.

Comparing the evolution of “AI Double 11” over three years feels like stepping into a large-scale “old wine in new bottles” scene: more AI shopping guides, endless marketing, customer service, and advertising Agent products, digital humans claiming to replace hosts, AIGC creative generation tools that can “produce images in one second”…

E-commerce, which naturally has data and scenario needs, is one of the earliest industries where AI landed and one of the closest to commercialization. This Double 11, e-commerce giants no longer tell stories around single AI capabilities but show a comprehensive AI landscape.

Jia Luo, president of Tmall, defined this year as “the first Double 11 with full AI implementation.” In addition to launching 6 C-end AI shopping guide products, B-end AI tools also basically cover the entire link of e-commerce people, goods, and scenes—from business analysis, commodity operations, crowd operations, content material production to delivery strategy optimization, AI seems to help brands find growth points in the traffic (involution) of the promotion.

JD also said that this year’s Double 11 is the one with the deepest integration of JD AI, large models, robots, and other technologies. It not only released a new e-commerce AI architecture “Oxygen” before Double 11 but also moved the large-scale “Super Brain + Wolf Tribe” intelligent device cluster to the technical stage of Double 11 for the first time, showing its ambition for “embodied intelligence.”

C-end precise AI shopping guides and B-end merchant management assistants seem to have become a reality this Double 11. Some media described: “This Double 11, my AI is busier than me.” “AI saves e-commerce workers,” etc. We talked to consumers and workers in this wave and found that AI has indeed penetrated their actual work and solved some problems.

But some (running-in) and contradictions are also emerging: How to solve the mistakes made by AI? Will AI replace my job? When AI becomes more understanding of my needs, will there be new problems of “over-understanding”?

Consumers: Using AI to “Shop Hard” for Double 11, But Trust Hasn’t Fully “Boarded”

This Double 11, after using DeepSeek to make her shopping strategy, Yangcong (pseudonym)’s shopping experience became more convenient. She tried many prompt words, such as directly throwing her budget and rough shopping plan to DeepSeek, but found that the price prediction given by AI was not accurate and could not provide detailed cross-platform price comparisons.

After training the AI for several days, Yangcong simply handed the last step of her shopping to AI: before paying, let AI calculate her Double 11 shopping discounts.

She first let AI organize the products she wanted to buy into a table, including shopping funds, deposits, final payments, etc., sorted out her spending records in detail, and then let AI help calculate the unit price of the products. “I mostly want AI to help me figure out how much I spend. As long as the final unit price is cost-effective, it means I didn’t lose.”

People like Yangcong who use AI to calculate accounts are just a type of target group for “AI shopping guides.” They have clear shopping goals and roughly know the price range of products on various platforms. Using AI just wants to find the most cost-effective one among the established choices.

Other consumers don’t have detailed plans for Double 11 shopping guides. They are more looking forward to AI being able to handle everything for them, from generating shopping needs to completing purchases.

Double 11 coincides with the “new phone war” of mobile phone manufacturers. Didi (pseudonym), who wanted to change to a new phone with the national subsidy, let Manus, ChatGPT Atlas, and other Agents make detailed shopping guides for her.

Didi wanted to buy an iPhone 17 Pro but didn’t know how to buy it more cheaply. Manus not only provided detailed shopping nodes and discount information but also gave alternative options, comparing parameters of other brands such as Samsung Galaxy S25 Ultra and Xiaomi 17 Pro Max. Using ChatGPT Atlas’s Agent mode, i.e., OpenAI’s new browser, can also achieve the same effect.

More and more people are using AI to shop. OpenAI’s previous research found that about 2% of the 2.5 billion prompts ChatGPT receives every day involve “judging the quality of goods, services, or people.”

A September survey by Adobe found that more than a third of consumers have used AI tools for product research, recommendations, and finding deals.

OpenAI has accelerated its AI e-commerce layout in the second half of this year: on one hand, it cooperates with Shopify, Walmart, etc., and on the other hand, it launches an instant checkout function to open the “shopping-transaction” closed loop.

Domestic giants with e-commerce businesses have never given up (exploring) AI shopping guides. Taotian, which launched 6 products, and Doubao, which is connected to Douyin Mall, have begun to let AI guide traffic for e-commerce.

Although users have needs and giants are accelerating their bets, it does not mean that “AI shopping guides” are a rich mine.

First, there are still many optimizations needed on the user experience side.

For example, Didi mentioned that the discount information provided by Manus was not accurate: “You can buy the 256GB version of the iPhone 17 Pro for 7699 yuan, which is impossible.” She also mentioned that e-commerce discounts are changing at any time, but AI guides cannot adjust in real time.

Other users are worried about privacy issues. A previous survey by KPMG of 1,500 U.S. consumers found that 43% of people are uncomfortable with companies using AI to analyze their personal data to provide shopping recommendations.

Second, technology iteration takes time.

“AI shopping guides” rely on long-term technical evolution of both “people” and “goods.” On the one hand, for “people,” a large amount of user behavior data accumulation is needed. Industry players need to improve model accuracy through multi-dimensional data fusion such as data collection, data compliance, and data storage; on the other hand, for “goods,” it is to iterate the original search and promotion infrastructure through AI. For example, Taotian upgraded its original product library at the beginning of 2025.

Kai Fu, president of Search and Recommendation Intelligence at Alibaba China E-commerce Group, also mentioned: “The essence of AI is insight into people, shifting from the original description based on standard product keywords to a deeper and more personalized understanding.”

In other words, to make “AI shopping guides” more understanding of people’s hearts, it is necessary to establish a “strong trust” relationship with consumers, but currently, trust has not really “boarded.”

E-commerce Workers: AI Is Busier Than Me, But My Heart Is Tired

For many e-commerce workers, this Double 11, AI is finally busier than themselves.

Tan Si (pseudonym) works in an online education company with about 50 people, responsible for e-commerce course operations. Every Double 11 is a pressure test, and working overtime is the norm. This Double 11, the company introduced AI to private domain mass messaging and front-end customer acquisition consulting, which greatly reduced her pressure.

“In the past, a dozen operation assistants had to manually private message hundreds of thousands of users to send ‘copy + course link + group link.’ Now with AI, only 2-3 people are responsible for mass messaging settings. AI can run multiple accounts at the same time, and you can plan the end time.”

Tan Si also mentioned that in the past, manual private messaging to users would result in account bans if the speed was too fast or exceeded a certain number, but the company is currently testing a consultation AI that simulates assistant replies to communicate with users, reducing the risk of being banned.

However, although AI has reduced her past repetitive labor, Tan Si’s anxiety has not eased.

On the one hand, the company has laid off some outsourcers and even operation staff, making her feel that she could be replaced by AI at any time; on the other hand, after AI improved efficiency, the KPI pressure for Double 11 is heavier. After the operation work was replaced by AI, her work content this Double 11 included delivery on other platforms.

“The physical fatigue has been reduced, but the heart is more tired,” Tan Si said.

Earlier than Tan Si to experience this spreading anxiety are e-commerce advertising optimizers.

Li Nan (pseudonym), 30, who just resigned from an e-commerce agency, told us that advertising optimizers may be the first group to have a direct experience of AI e-commerce.

“If you don’t embrace AI, you will be eliminated by AI,” Li Nan told us. AI automated marketing delivery solutions are everywhere, such as Google’s Performance Max, Meta’s Advantage+, Tiktok’s Smart+ overseas, and Ali Mama’s Wanxiangtai AI Boundless, Douyin’s (Ocean Engine) domestically.

In the past, advertising optimizers might have to manually create and set audience packages, set budgets, set bids, and monitor accounts 24 hours a day to adjust budgets, which was the norm. But now with AI, delivery has become intelligent: “The backend settings have become intelligent. AI can test images, generate materials in batches, and optimizers only need to set budgets and fill in conversion goals.”

But intelligence also has “negative effects.” Li Nan once was responsible for the account of a health industry client. Due to overtrust in data, the final ROI performance was not good. Many optimizers around him who did basic work have changed careers. In the eyes of optimizers, AI does save manpower, shifting from staying up late to monitor accounts in the past to focusing on material control and account operations.

But Li Nan also mentioned that intelligence has a price, such as “black boxization” brought by intelligence. In the past, advertisers could manually control which audience packages to target and which creatives to make, but now after one-click (: entrusted management) of intelligent delivery, transparency has been greatly reduced.

This brings confusion to workers: “In the past, many clients didn’t understand delivery, and now they understand even less about intelligent delivery.” When problems arise, who should take the blame for the AI algorithm—the AI or the worker?

AI E-commerce: A Long Road Ahead

Although this is a Double 11 full of “AI flavor,” there is still a long way to go before the full application of AI e-commerce.

An insider from Taotian told “Silicon Research Lab”: “C-end shopping guide products including AI Universal Search are still being gradually optimized, focusing on polishing the products. There will definitely be certain commercial space in the future, but it needs to wait until it matures.”

To further optimize the user experience and make searches more accurate, on the one hand, users need to use it frequently—which is why major factories are collectively (: making efforts) on “AI shopping guides” this Double 11; on the other hand, the platform needs to connect the data system with the brand around search terms, which takes a certain period of time.

Looking inward, in the technical infrastructure of search and promotion, various parties are exploring recommendation models in the era of large models. Models such as SASRec from UCSD, HSTU from Meta, TIGER from Google, OneRec from Kuaishou, MTGR and UNM from Meituan are constantly optimizing their structures, but they generally face challenges such as data instability, high inference latency, and high storage and computing costs.

Bill Ready, CEO of social platform Pinterest, once said that AI shopping will be a protracted war. “The idea that an agent buys everything for you and you don’t have to do anything—I think it will be a very, very long cycle to realize it.”

In terms of the commercialization of AI e-commerce, giants with e-commerce businesses can naturally take the lead in building an internal cycle of “demand-recommendation-transaction,” but the extent to which AI companies can “monetize through e-commerce” is still a big question mark.

Researchers from the University of Hamburg found in their latest research that new e-commerce channels such as ChatGPT have not had an impact on traditional channels such as Google in the short term.

The researchers looked at real data from 973 e-commerce websites, checked 50,000 ChatGPT recommendations, and compared two dimensions: (profitability) (conversion rate, average order value, etc.) and user activity (bounce rate, visit duration, etc.). They found that traditional channels still outperform ChatGPT.

This further shows that the user penetration and commercialization of AI e-commerce are still a long road. The AIs rushing to carry goods this Double 11 still need to solve more complex challenges in e-commerce scenarios.