The newsletter cites Adobe's finding that 39% of US consumers have used AI for shopping and Salesforce's estimate that AI influenced 20% of purchases during last year's holiday weekend, while only 11% of shoppers let AI buy for them. It argues the assistant's influence sits in the shortlist, which is built from the reviews, videos and brand searches customers produce after they buy. Marketers should treat post-purchase reviews, customer videos and branded search as inputs to the AI shortlist and build a loop that keeps producing them.
Seven months after OpenAI began selling ads in ChatGPT, Digiday reports that marketers still cannot measure campaigns well enough to commit real budgets, and that measurement is the biggest problem in almost every interview about the business. Accuracast's Farhad Divecha said a UK lead campaign produced form fills that ChatGPT never logged as conversions, and that his agency once counted 20 clicks where the platform reported about 100; Web Guide Partner said conversions can take 24 to 36 hours to appear, against a few hours on Google. Expect to run ChatGPT ads as a test budget until you can reconcile conversions in your own analytics, and expect no demographic or prompt-level reporting.
A Microsoft patent granted on August 25 describes how Bing decides to generate an AI answer inside its results, scoring each query for complexity and producing an answer only for questions that need several sources combined. The query is expanded into sub-searches, and a small model scores those results and keeps the best-ranked links for the answer text; a generated answer can be reused while the underlying results stay stable. Marketers should plan for Bing's answer surface to be decided at the level of individual links and sub-queries, so both the page and the terms it can be found for matter.
Alibaba's Qwen put an application page for its personal agent, Qwen Go, into the Qwen app and says it will admit users in waves. The company says the agent handles tasks across shopping, travel and logistics, and a reporter's test had it pursue a flight delay claim with a platform's customer service, ending in a 70 yuan payout. Brands selling on Alibaba's marketplaces should expect orders, price comparisons and refund requests to arrive from a Qwen agent acting for the customer.
1688, Alibaba's sourcing platform, says more than 100 million annual buyers and more than 1 million factory suppliers can open the Accio AI workbench from their accounts. Buyers get a purchasing assistant that searches suppliers in 128 countries and compares prices, plus a manager agent that runs Taobao, Tmall, Douyin, Pinduoduo and JD stores; suppliers get a shop assistant that handles listings, ad analysis and customer reception. Suppliers selling through 1688 should keep prices, stock and listing data accurate for machine comparison, because buying agents will run the search and the inquiry.
Newegg says it has integrated Google's Universal Commerce Protocol so its products can be bought inside Google AI Mode, Google Search and Gemini. It follows Wayfair, Walmart and Macy's, which have supported UCP since Google launched it in January. Retailers and brands should check whether their catalogs reach Google through a UCP feed, because items that do not will not be purchasable inside those AI surfaces.
Values.media says it has connected to M6 and France Télévisions so that linear TV sponsorship can be bought by software agents, after an earlier test with TF1, according to the group's managing director, Emmanuel Crego. The link lets a buyer's agent handle the transaction through the broadcaster's systems. Agencies and advertisers should settle now which inventory rules, budgets and data their buying agents may act on without a person in the loop.