文 | 强调Next
商家已经开始为AI付费，消费端还在补贴买订单。
10月20日晚8点，天猫双11开卖，4小时后，124个品牌成交额破亿。三天后，媒体实测发现，同一店铺的一款厨房纸巾，从抖音电商购买要27.9元，经豆包领取补贴和优惠券后，只要17.43元。千问的“双11帮我挑”页面也能领取消费券，并已支持在App内完成选品、下单和支付。
价差10.47元，足以让消费者换一个下单入口。豆包把抖音直播间搬进聊天窗口，又用渠道优惠吸引用户购买。AI开始接管购物流程，电商最熟悉的补贴也跟着进来了。
商家侧走的是另一条路。10月14日，千问办公发布电商版，披露现有付费用户中，20%来自电商行业。消费者用AI购物能拿优惠，商家用AI经营则已经掏钱。
这届双11，AI商业化先兑现的是商家的工作预算。消费者一侧还没发生类似比例的迁移。商家原本就要支付的设计、运营和分析费用，已经成了AI的收入来源。对电商公司而言，这显然比创造一笔全新的消费更容易做成生意。
优惠券进了AI，入口之争换了位置
去年双11，淘宝天猫已经推出“AI万能搜”“AI帮我挑”等六款导购应用。今年，竞争向交易环节推进。媒体记者实测发现，豆包可以在对话中推荐抖音直播间，千问也能在App内完成选品、下单和支付。
这一步改变了AI与电商的关系。过去，用户问完AI还要回到购物App挑货。现在，AI入口本身可以承接交易。谁先接住用户的问题，谁就有机会决定用户随后进入哪家店。
海外已经出现了更具体的成交证据。Adobe对美国零售网站的监测显示，2026年7月，经AI链接到站的访客，购买转化率比非AI来源高60%，每次访问收入高53%，加购率高28%。这组数据统计的是离开AI、进入商家网站之后的行为。AI无需包办付款，也能成为有商业价值的获客渠道。
这也让电商公司的争夺变得更白热化，因为用户可以留在淘宝、京东完成交易，但筛选商品、决定去哪家买的过程，可能已经发生在别处。对豆包这样的外部入口而言，先影响订单去向，就有了争取商家预算的理由。对电商公司而言，支付和履约还在自己手里，也不代表购买决策仍由自己主导。
豆包的补贴尤其能说明这场争夺。上述纸巾经豆包购买，比抖音电商路径便宜约37.5%。用户无需先相信AI比自己更懂商品，也有理由试用它。
字节正在用价格差培养用户购物之前先打开豆包的习惯。补贴支付的是用户改变路径的成本，抖音的商家和直播间则负责承接交易。
两家公司都在把原有电商能力送到新的用户界面。对阿里而言，千问要接住用户离开淘宝之后的消费咨询。对字节而言，豆包提供了一条不必先看短视频、刷直播，就能进入抖音商品体系的路径。
但就现阶段的产品而言，AI购物仍处于早期竞争，带着浓厚的电商色彩。模型负责理解一句模糊的需求，商品库决定能提供什么，优惠决定用户是否愿意在这里下单。单靠对话能力，很难从现有购物渠道手里拿走交易。
这也决定了平台宣传中的“AI成交”需要两面看。一个原本准备买纸巾的人，改用豆包领券购买，首先发生的是订单迁移。字节买到了购物入口的使用机会，消费者省到了钱，但这笔订单还不能被完整记成AI创造的新需求。
抢走一笔订单，比创造一笔需求容易。
AI导购已经进入前一种竞争。对电商公司来说，先把别人入口里的购买需求接过来，就足以构成投入的理由。
好导购，未必是好销售员
今年6月，上海市消保委一份基于4308份问卷的调查显示，84.56%的受访者尝试过AI选购功能，认为AI能精准匹配商品的比例为16.06%。38.79%反馈AI不贴合低价筛选需求、优先推送高价商品，29.71%反馈高低价商品混杂，仍需手动筛选。
消费者已经愿意开口，工具却没替他们省掉足够多的工作。预算说过一遍，筛选还得自己重做，对话越长，AI导购的价值就越容易被消耗掉。
技术进步能够减少这种错误，真正难处理的，是AI导购站在谁那一边的问题。
消费者希望少选错、少花冤枉钱。电商公司需要提高成交，商家希望卖出更多商品。用户已经有一台够用的电脑，AI建议继续使用，完全可能是一条好建议，却给商家带不来订单。选出更便宜的替代品，也可能帮助消费者降低支出，同时压低客单价。
传统搜索允许这种分歧藏在一排商品里，用户自己翻页、比价。AI直接交出一个结论，也就把选择责任接到了自己身上。它越像消费者的私人顾问，消费者就越会追问：这条建议究竟替谁考虑？
对话界面解决不了这个利益问题。电商公司可以用更自然的语言推荐商品，但要把一次尝鲜变成长期信任，就得接受一部分咨询以“少买”甚至“不买”结束。
这正是消费端AI比商家工具难变现的地方。商家交钱，让AI帮自己经营，两者的任务相对明确。消费者请一个由电商公司提供的免费助手替自己省钱，服务目标从一开始就存在拉扯。模型更聪明之后，这个问题也不会消失。
Token货架，先争客户从哪里付钱
今年9月，天猫上线Token充值入口，智谱、MiniMax等厂商陆续入驻。京东也推出相应服务站。AI服务被装进月卡、季卡和年卡，开始借用电商的店铺与订单体系销售。
天猫给出的双11首日数字是，Token品类成交环比上月增长450%。第一财经10月23日观察到，天猫智谱旗舰店相关商品累计销量约300件，京东一款智谱月卡显示已售100多件。这些店铺还在起量阶段，450%的增速远不足以说明Token已经成为大众爆款。
天猫也没有急着用补贴把它做大。天猫总裁家洛接受采访时表示，暂时不会在这一品类有具体的优惠投入。与消费端用渠道券吸引订单相比，Token货架的投入明显更克制。
这里争夺的是已有付费需求的购买路径。需要模型服务的人，过去到厂商官网付钱，现在可以到淘宝、京东买。电商公司先把交易入口搭起来，再争取套餐比较、购买和续费留在自己的体系内。
这与AI导购争夺消费者购物习惯是同一种渠道生意，只是买家变成了模型用户。Token销量衡量的是电商渠道拿到多少AI服务订单，不能用来判断AI导购懂不懂消费者，更不能代表整个AI商业化的进度。
两种生意混在一起，容易制造一种错觉：AI商品卖得越好，AI就越会卖商品。事实上，天猫完全可以在消费者仍不信任AI推荐的时候，先从AI服务交易中找到新的经营机会。
省了素材钱，贵了买量账
千问办公的付费用户构成，解释了阿里为什么先把行业版本做进电商。
商家不需要等AI证明自己能创造消费需求，才决定是否购买。详情页要做，经营数据要分析，商品要测试。AI只要能以更低成本交付这些工作，就有机会进入原有预算。
千问办公发布材料中的女鞋商家刘祥龙称，借助AI筹备一家3D打印鞋店，周期从30天缩短到7天，相关花费由3万至5万元降至几百元Token费用。这是一个商家案例，其价值在于呈现了经营顺序的变化：部分市场调研和展示素材可以提前完成，商家能够更便宜地验证想法，再决定投入多少生产和备货资金。
商家购买的是一次更便宜的尝试。即使新品最后没卖成，已经完成的素材和分析工作仍然有价格。AI服务商因此可以在商品产生销售之前，先收到钱。
阿里进一步接入生意参谋、千牛、有赞、阿里妈妈万相台，则把工具从内容制作延伸到了经营与投放。商家做完素材，不必离开这套工具体系，就能继续分析表现、调整预算。这让AI有机会持续参与经营，而不只是在开店时用一次。
更深一层的变化，是AI降低制作成本之后，电商竞争并不会跟着变便宜。
一家店一天能做十组素材，对手也能。商品图更容易生产，消费者的注意力仍然有限，商家还要争曝光、争点击。素材供给增加之后，筛选与投放的重要性反而上升。省下来的制作费，随时可能被下一轮获客竞争吃掉。
对阿里而言，这形成了两处收入机会：千问办公承接工具预算，阿里妈妈承接广告预算。商家在前一道工序省钱，能够继续投入后一道工序；AI提高了经营效率，也增加了商家留在这套经营系统里的理由。
因此，“AI帮商家降本”远没有把这门生意说完。效率提高后，收益归谁，才决定AI最终改变了谁的处境。工具服务商先收到了订阅费，电商公司仍握着流量分配，商家则需要在同样变得更高效的对手之间继续竞争。
这届双11，商家是AI最明确的一批付款人。至于他们能留下多少利润，拍图变便宜只是开头，买流量的账还在后面。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Man, stress on Next
Businesses have started to purchase orders for AZpaid and the consumer side is subsidising.
At 8 p.m. on 20 October, Skycats sold 11, and four hours later 124 brands were sold at billions. Three days later, the media discovered that a kitchen napkin in the same shop, purchased from the Douyin electrician, would cost $27.9 and only $17.43 after Doubao received a subsidy and coupon. The Double 11 page will also receive the coupons and have supported the completion of the selections, orders and payments in the App.
The difference of $10.47 is enough to replace the consumer with a single entry point. Doubao moved the Douyin live booth into the chat window and used the channel to attract users for purchase. AI started taking over the shopping process, and the most familiar subsidy for the electrician came in.
The business side is going another way. On 14 October, the publish electrician version of the office revealed that 20 per cent of the current paid users came from the electrician industry. Consumers can get a discount for their purchases and businesses can pay for their businesses using AI.
This double 11, the commercialization of AI began with the work budget of the business. A similar rate of migration has not occurred on the consumer side. The design, operation and analysis costs that businesses would have to pay have become sources of revenue for AI. This is clearly easier for electric companies to do business than to create an entirely new consumption.
The coupons are in AI, the entrances are in different places.
Last year's double 11, the Treasure Cat has launched six pilot-purchase applications like "Ai Can Search" and "Ai Help me Pick." This year, competition moved towards the trade chain. Media journalists have found that Doubao can recommend Douyin during the dialogue, and that thousands of questions can also complete the selection, listing and payment in App.
This step changed AI's relationship with the electrician. In the past, the user had to go back to shopping for App. Now, the AI portal itself can take over the deal. Whoever catches the user first has the opportunity to decide which shop the user will then enter.
More specific evidence of a deal has emerged overseas. Adobe's monitoring of retail websites in the United States showed that in July 2026, visitors who had been linked to the site by AI had a 60 per cent higher conversion rate than non-AI sources and a 53 per cent higher income per visit and a 28 per cent higher purchase rate. This set of data measures behaviour after leaving AI and entering the business website. AI can also be a commercially valuable source of access without having to make a charter payment.
This has also made the competition for electricity companies more ferocious, as users can stay in the shop and finish the deal in Kyoto, but the process of sifting goods and deciding where to buy them may have taken place elsewhere. For an external entry such as Doubao, there is a reason to seek a commercial budget by first affecting the direction of the order. Payments and compliance are in their own hands for electric companies, and do not mean that purchasing decisions remain in their own hands.
Doubao's subsidies are particularly indicative of this competition. The above-mentioned paper towels were purchased through Doubao, which is about 37.5 per cent cheaper than the Douyin electrician path. Users need not first believe that AI knows more about commodities than they do, and there is reason to try it.
byte is opening up Doubao's habits before buying with poor prices. The subsidy covers the cost of the user ' s rerouting, while Douyin ' s commercial and live booths are responsible for the transaction.
Both companies are bringing the original power supply capacity to the new user interface. In the case of Ali, he asked to take up consumer consultations after the user left the shop. For byte, Doubao provides a path to Douyin's commodity system without first looking at short video and going live.
However, at this stage of the product, AI purchases are still in the early stages of competition, with a strong electrician colour. The model is responsible for understanding the vague demand, the commodity bank decides what to offer, and the preference determines whether the user is willing to file here. The ability to speak alone makes it difficult to take trade from existing shopping channels.
It also determines that the “AI deal” in the Platform's advocacy needs to be seen in two ways. One of the men who was supposed to buy paper towels moved to Doubao coupons, the first being the relocation of orders. byte bought access to the shopping mall and the consumer saved money, but the order was not fully recorded as a new demand created by AI.
Taking an order is easier than creating a demand.
AI-guided purchases have entered the former category of competition. For an electrician company, it is sufficient to transfer the purchase demand from someone else's entrance to justify the input.
Not necessarily a good salesman.
In June this year, a survey conducted by the Shanghai City Commission for the Protection of the City of Shanghai, based on 4308 questionnaires, showed that 84.56 per cent of respondents had tried the AI-selection function, and 16.6 per cent believed that AI could accurately match commodities. 38. Seventy-nine per cent of feedback AI did not match low-price screening needs, priority was given to high-priced commodities, and 29.71 per cent of feedback was mixed and still required manual screening.
Consumers have been willing to speak, but the tools have not saved them enough work. As the budget had said, screening had to be done on its own, and the longer the dialogue, the more easily the value of AI-guided purchases would be consumed.
Technological advances can reduce such mistakes, and it is really difficult to deal with the question of whose side AI leads the purchase.
Consumers want fewer mistakes and less money. Electrician companies need to improve their deal, and traders want to sell more. Users already have an adequate computer, and AI recommends continued use, which may well be a good proposal, but does not bring an order to the business. The selection of cheaper alternatives may also help consumers to lower their expenditures while keeping their prices down.
Traditional searches allow such disagreements to be hidden in a row of commodities, with users turning their own pages and prices. By handing over a conclusion, AI had placed the responsibility of choice on itself. The more it resembles a private adviser to the consumer, the more the consumer will ask: Who is it for?
The dialogue interface does not solve this issue of interest. Electrician companies can recommend commodities in more natural terms, but in order to turn a taste into a long-term trust, they have to accept some of the consultations to end “less” or even “no”.
This is where consumer-end AIB business tools are hard to come by. The business has a relatively clear mandate to pay and allow AI to run for itself. Consumers have asked a free assistant from an electrician company to save their own money, and the service objectives have existed from the outset. When the model is smarter, it will not disappear.
Token's shelf. First, the customers pay for it.
In September this year, the Tokentop up entrance, the Zhipu, Mini Max and others, on Skycat, continued to be stationed. The corresponding service stations were launched in Kyoto. AI services are loaded into monthly, quarterly and annual cards and start selling on the electrician ' s shop and order system.
The double 11-day figure given by Skycat is a 450 per cent increase in the number of token-like trade rings over the previous month. On 23 October, the first financial year observed a cumulative sales volume of about 300 items at Skycat Zhipu flagship, and a Zhipu monthly card in Kyoto showed that over 100 items had been sold. These stores are still in the start-up phase, and the increase of 450 per cent is not sufficient to suggest that Token has become a public explosive.
Skycat didn't rush to make it big with subsidies. In an interview with the director of Skycat, Lou, it was stated that specific preferential inputs would not be available for the time being. Token shelf inputs are significantly more restrained than consumer-end coupons attract orders.
The competition is for a purchase path that already requires paid. Those in need of model services, who used to pay for it in the network of manufacturers, can now buy it in Kyodong. Electrician companies first set up the trading entry and then seek to keep the package comparison, purchase and continuation in their own systems.
This is the same source business as the AID procurement competition for consumer shopping, except that the buyer has become a model user. Token sales measures how much of the AI service orders are received by the electrician channel and cannot be used to judge how well the AI-guided purchase understands consumers, much less represents the progress of the entire AI commercialization.
The combination of the two types of business makes it easier to create the illusion that the better it sells, the better it sells. In fact, Skycat can find new business opportunities in the AI service transaction when consumers still do not trust AI ' s recommendations.
Saved asset's money and paid for it.
paid users at the office explained why Ali first made the industry version into an electrician.
Businesses do not have to wait for AI to prove that they can create consumer demand to decide whether to buy. Detailed pages, business data analysed and commodities tested. AI has an opportunity to enter the original budget as long as it can deliver these tasks at lower cost.
Liu Cheunglong, a shoe dealer in the publish office, stated that the cycle had been reduced from 30 to 7 days, and the cost had been reduced from $30,000 to $50,000 to several hundred token costs through AI. This is a business case, the value of which is to present a change in business order: Some market research and demonstration of asset can be done ahead of schedule, and businesses can validate ideas more cheaply and then decide how much to invest in production and preparation.
Business buys a cheaper attempt. Even if the new product was ultimately not sold, there were prices for the completed asset and the analysis. The AI service provider can therefore receive money before the goods are sold.
Further access to business staff, Chibo, Zan, and Ali's mother, Man, extended the tools from content production to business and delivery. Businesses can continue to analyse performance and adjust their budgets without having to leave the tool system. This gives AI the opportunity to participate in the business on a continuous basis and not only once in the business.
The deeper change is that after AI has reduced production costs, electricians do not compete to be cheaper.
A store can make 10 assets a day, and so can the opponent. Commodity maps are more easily produced, consumer attention is still limited, and businesses need to compete for exposure and click. The increased supply of asset increased the importance of screening and delivery. Saved production costs could be eaten at any time in the next round of competitions.
For Ali, this has created two income opportunities: a budget for office delivery tools and an advertising budget for Mother Ali. Businesses save money from the previous process and can continue to invest in the latter process; AI improves business efficiency and increases the rationale for the business to remain in the system.
As a result, “AI’s Businesses Deprivation” is far from completing this business. When efficiency gains are realized, it is the return that determines who AI ultimately changed. Tool service providers first received subscriptions, electric companies still held the flow distribution, and businesses needed to continue to compete among equally more efficient rivals.
This double 11, the merchant is the clearest group of AI payers. As to how much they could leave behind, the cheapness of the picture was only the beginning, and the flow of money was still in the back.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.