--- title: AI Assistants Product Availability: Sources and Freshness url: https://click2buy.com/en/ai-assistants-product-availability/ type: post date: 2026-11-10 description: AI assistants show product availability that comes mostly from what the seller has declared, through a product data feed sent to the platform or on the seller’s product page. Platforms… --- # AI Assistants Product Availability: Sources and Freshness AI assistants show product availability that comes mostly from what the seller has declared, through a product data feed sent to the platform or on the seller’s product page. Platforms make few public commitments on how fresh that data is, and two October 2026 tests found wrong answers and products recommended while out of stock. For a brand that sells through retailers, the question is therefore where the information comes from, and how old it is. * **Key takeaways** - The availability shown comes mainly from merchant feeds, sellers’ product pages and web search, not from a live check on store shelves. - Google says more than 2 billion listings are refreshed every hour, out of more than 50 billion (May 2025 announcement). OpenAI asks for a full feed at least daily. Microsoft and Amazon do not state a refresh rate in the public documentation we reviewed. - A Core dna test found 21 wrong answers out of 40 (prices, Australia, a single day). A First Page Sage test found 4.5% to 7.6% of recommendations were out of stock or no longer sold. Neither test we reviewed covers France. - A brand cannot fix the stock displayed by itself. It can monitor its retailers and make its product information easy to read. ## Where do AI assistants get their stock information? Mainly from feeds that merchants send to the platforms. For Google, the Search Central documentation advises checking that the information in Merchant Center, the tool merchants use to send their data to Google, is up to date. In May 2025, Google said its Shopping Graph, a database of product listings, held more than 50 billion listings, more than 2 billion of them refreshed every hour, and that it powered shopping in AI Mode. This overall figure does not mean every listing is updated hourly. OpenAI has published a product feed specification for ChatGPT. The availability field is mandatory and accepts five values: in stock, out of stock, preorder, backorder, and “unknown*”, to be used when the status is not known. The feed is a full export of the catalog, submitted at least once a day. According to Microsoft, Copilot relies in particular on the feed submitted to Microsoft Merchant Center. According to Amazon, which does not say which markets are concerned, its assistant Rufus (renamed Alexa for Shopping in May 2026) draws on its catalog, customer reviews, community Q&A and information from the web, and the technology “won’t always get it exactly right”. ## Do product pages and web search matter too? Yes. The seller’s product page is the second source. The schema.org *Offer* markup includes an *availability* property (InStock, OutOfStock, PreOrder, and so on). Google warns that dynamically generated markup “can make Shopping crawls less frequent and less reliable”, and that structured data must match the text visible on the page. Merchant Center uses these pages to correct feeds: its automatic updates replace the price or availability in the feed with what is found on the page. Google states that this mechanism “may not work” for products whose price or availability changes frequently, “for example more than once per day”, and then recommends the Merchant API. Web search is the third source. According to Microsoft, Copilot relies in particular on information found on the web. For ChatGPT’s shopping research feature, results come from public merchant sites, according to Search Engine Journal. * ## How often is this data updated? Public commitments are rare. Google gives an overall figure and OpenAI sets a minimum update frequency for feeds. The public documentation we reviewed from Microsoft and Amazon gives none. Assistant Source of stock data Stated freshness Google (AI Mode) Merchant Center and product pages, via the Shopping Graph More than 2 billion listings per hour, out of more than 50 billion (May 2025 announcement) ChatGPT Merchant product feeds, public merchant sites Full feed at least daily Microsoft Copilot Microsoft Merchant Center and the web Not stated in the public documentation reviewed Amazon (Rufus) Amazon catalog, reviews, community Q&A, the web Not stated in the public documentation reviewed ![Google: over 2 billion listings an hour; OpenAI: feed at least daily; First Page Sage test: 4.5% to 7.6% out of stock](https://click2buy.com/wp-content/uploads/2026/11/download-2.png) ## What do tests show about availability errors? When it launched shopping research on November 24, 2025, OpenAI warned that the feature can make mistakes on product details such as price and availability, according to Search Engine Journal. Neither test we reviewed covers France. On October 20, 2026, e-commerce software vendor Core dna, asked ChatGPT twice, for 20 products, “Where is the cheapest place to buy the [product] in Australia right now?”: 21 of the 40 answers were wrong. The price of a Garmin watch came from comparison-site data that ChatGPT itself said was last updated 40 days earlier, and a retailer recommended for Bose earbuds listed them as “out of stock online”. For 11 of the 20 products, the same question asked later the same day returned a different store or price. The test covers a single day. Marketing agency First Page Sage put 480 shopping questions to eight assistants, then checked each recommended product on the retailer’s site within the hour, between September 28 and October 14, 2026. It found 5.6% of recommendations out of stock or removed from sale for ChatGPT, 7.6% for Copilot and 4.5% for Rufus. The market is not specified, the selection of questions is not detailed, and the differences are too small to rank the assistants. ## What can brands do? A brand that sells through retailers does not declare the displayed stock itself: each retailer’s feeds and pages are what count. There are four things a brand can do. First, know which retailers feed Google Merchant Center, Microsoft Merchant Center or OpenAI’s feeds, since these feeds are the first source described by the platforms. Second, keep the feed, the product page and its availability* markup consistent, since, according to Google, an offer can be rejected when the two differ. Third, query the assistants regularly on key products, repeating each question: answers varied from one session to the next in the Core dna test. Fourth, look at the information the brand publishes itself. Some assistants, such as Copilot and ChatGPT’s shopping research, rely in part on what they find on the web, and a “Where to buy” page can be one of the pages they find. [Click2Buy’s Where to Buy](https://click2buy.com/en/where-to-buy/) can contribute: for a brand, it shows the retailers that sell its products. This may help the brand’s products be seen and read by AI assistants, without guaranteeing that they cite them: none of the platforms whose documentation we read commits to this. The tool shows where to find a product. It does not provide stock information: availability remains that of the retailers’ feeds and pages. *Disclosure: Click2Buy, which publishes this article, makes and sells Where to Buy.* **What to avoid**: assuming an AI assistant’s answer reflects today’s stock at a retailer, when the source used may be a daily feed, a page crawled earlier or a comparison site that has not been updated (in the Core dna test, one price came from comparison data 40 days old). **Good practice**: for priority products, compare actual availability at each retailer with what is declared in the feeds and on the product pages, then with what the assistants display, repeating the tests. ## Summary The product availability shown by AI assistants depends on the data sellers declare and on when the platforms pick it up. The method for measuring [online product availability](https://click2buy.com/en/measuring-online-product-availability-the-real-challenge-no-one-talks-about/) is detailed in a previous article, and the arrival of [Google Universal Cart](https://click2buy.com/en/google-universal-cart-2026-how-retailers-are-adapting/) raises the same question of data quality. It also arises during demand peaks, such as [Black Friday](https://click2buy.com/en/product-availability-black-friday-2026-whats-at-stake/). Click2Buy’s [Where to Buy](https://click2buy.com/en/where-to-buy/) may help assistants find where to buy a product, though no platform guarantees it. ## Frequently asked questions ### Where does ChatGPT find a product’s availability? OpenAI documents a product feed submitted by merchants, with a mandatory availability field and a full upload at least daily, and says shopping research draws on public merchant sites. OpenAI also warns that the feature can make mistakes on price and availability. ### How often does Google update the stock shown in AI Mode? Google says more than 2 billion of its more than 50 billion listings are refreshed every hour, with no guaranteed frequency per listing. The data comes mainly from Merchant Center and merchants’ product pages. Google states that its automatic corrections may not work for products whose price or availability changes frequently, for example more than once per day. ### Can a brand correct wrong stock shown by an AI assistant? To our knowledge, no platform documents a procedure that lets a brand directly correct the stock displayed. When the product is sold by retailers, availability comes from their feeds and product pages: the fix goes through the retailer involved, once you have identified where its actual stock, its declared data and the assistant’s answer diverge. This conclusion is based on the public documentation we reviewed. ## Sources - [Search Engine Journal, launch of shopping research in ChatGPT (November 24, 2025)](https://www.searchenginejournal.com/chatgpt-adds-shopping-research-for-product-discovery/561840/) - [Core dna, ChatGPT test on 20 Black Friday products (October 2026)](https://www.coredna.com/blogs/chatgpt-black-friday-recommendations) - [Google, shopping in AI Mode, Google I/O announcements (May 2025)](https://blog.google/products-and-platforms/products/shopping/google-shopping-ai-mode-virtual-try-on-update/) - [Google Search Central, AI features and your website](https://developers.google.com/search/docs/appearance/ai-features) - [Google Search Central, merchant listing structured data](https://developers.google.com/search/docs/appearance/structured-data/merchant-listing) - [Google Merchant Center, automatic product updates](https://support.google.com/merchants/answer/12157888?hl=en) - [OpenAI, product feed specification](https://developers.openai.com/commerce/product-feeds/spec) - [Microsoft Advertising, agentic commerce and Copilot Checkout](https://about.ads.microsoft.com/en/solutions/technology/agentic-commerce) - [Amazon, introducing Rufus and Alexa for Shopping](https://www.aboutamazon.com/news/retail/amazon-rufus) - [First Page Sage, comparison of AI assistants for shopping (October 2026)](https://firstpagesage.com/seo-blog/which-ai-is-best-for-shopping/) ![Photo of Hugo](https://click2buy.com/wp-content/uploads/2026/10/Photo-Hugo-C.png) Hugo, Marketing at Click2Buy “Before relying on the stock an AI assistant shows, you need to know where the information comes from and how old it is.”