ChatGPT Is Becoming a ShoppIng Discovery  Engine: What It Means for Your Brand This Black Friday

Written By Ryan Edwards

When someone asks ChatGPT this November to find the best Black Friday deal on a robot vacuum, the answer will most likely name two or three products, compare their prices and reviews side by side, and then point the shopper to a store to finish the purchase. I went through what is publicly available about ChatGPT's internal search instructions and its newest model, compared it with how the other leading AI models now handle shopping, and the pattern is consistent enough that I think it tells you where OpenAI has landed for this holiday season and probably for the next year. ChatGPT is not becoming a store. It is becoming the place where people decide what to buy.

GPT-6 changed how ChatGPT shows products

Two shopping rules were taken out of ChatGPT's search instructions in GPT-6.

In the earlier models, GPT-5.3 Instant and GPT-5.6 Sol, the instructions for ChatGPT's search tool included two rules about shopping. In GPT-6, both rules are gone from that tool.

  • How products are displayed. The older instructions told ChatGPT which format to use for products: one large featured product (called a hero), a carousel of options to scroll through, or a side-by-side comparison. Those rules are no longer in the search tool.

  • Products must appear as shopping cards. The older instructions said that when a search returned products, ChatGPT had to show them as shopping cards instead of plain text. That rule is also no longer in the search tool.

We do not know for sure where these rules went. Most likely they moved to a separate set of display rules, called the DIL spec, that has not been made public. [RYAN: confirm what DIL stands for, or whether to name it at all.] In practice, this means OpenAI can change how products look in ChatGPT without releasing a new model, so brands should expect the format of shopping answers to keep changing.

Three shopping behaviors stayed the same in GPT-6

These rules carried over from GPT-5, and each one affects whether a brand gets found.

  • ChatGPT checks product data and websites together. When someone shops, ChatGPT runs a product search and a regular web search at the same time, so its answers mix product feed data with information from web pages. Your product feed and your website both need to be accurate, and they need to match each other.

  • If product results are weak, ChatGPT searches again. Thin or poor product data makes ChatGPT keep looking. A complete product feed gives it what it needs the first time, which most likely makes your product easier to include in the answer.

  • Amazon is not in ChatGPT's shopping data. Amazon listings reach ChatGPT answers only through regular web search. For brands that sell mostly on Amazon, this most likely means your own website and your other retail partners carry more weight in ChatGPT than your Amazon listing does.

ChatGPT tried to sell products directly and stepped back

The checkout experiment was short, and what OpenAI learned from it shapes what it is building now.

In September 2025, OpenAI launched Instant Checkout, which let people buy a product without leaving ChatGPT. At launch it supported single-item purchases from U.S. Etsy sellers, and multi-item carts, promo codes, and sales tax handling were not yet in place. Adoption stayed small. In March 2026, Shopify's president said only about a dozen of Shopify's merchants had gone live, according to Forbes. On March 24, 2026, OpenAI said the first version "did not offer the level of flexibility that we aspire to provide," and that it was "allowing merchants to use their own checkout experiences while we focus our efforts on product discovery."

It would overstate things to say ChatGPT is going back to being a discovery tool, because it was never mainly a purchase engine in the first place. Instant Checkout ran for about six months as an experiment that did not catch on, and OpenAI returned its focus to what ChatGPT was already doing well. Purchases have not disappeared completely either. Merchants can still build ChatGPT apps for deeper experiences, and Walmart is launching one with account linking, loyalty, and Walmart payments. Restaurant reservations still run through booking partners from inside the chat. [RYAN: confirm the OpenTable and Resy reservation detail from your source review. I could not verify it from a public OpenAI page.]

The decision now happens inside the conversation

Discovery is only half of what ChatGPT does with a shopping question.

The way ChatGPT handles products has changed in steps over the last year and a half:

  • 2025: ChatGPT learned to look up products as part of search.

  • Early 2026: products began to appear as picture cards instead of plain text. [RYAN: confirm date]

  • March 2026: products appear side by side with price, reviews, and features, and shoppers can upload an image to find similar items and refine results by chatting, according to OpenAI.

  • Later in 2026: ChatGPT can find local stores and check open restaurant tables. [RYAN: confirm date and source]

  • GPT-6: the rules for how products are displayed moved out of the search tool, most likely into the separate DIL spec. [RYAN: this repeats the new GPT-6 section above. Keep for the timeline, or cut.]

"Discovery engine" undersells what is happening. A more accurate label is a discovery and decision engine, because ChatGPT does not stop at helping people find products. It compares them, summarizes the reviews, and recommends one. The best everyday comparison is a friend who researches everything before you shop, who reads the reviews (often on Reddit), checks the prices, tells you which product or which nearby business to pick, and then sends you off to buy it somewhere else. [RYAN: confirm the Reddit observation comes from your review of the source material.]

The practical consequence for a brand is that the shortlist forms before the customer ever reaches your website. Your site still matters, because that is where the purchase happens, but its job shifts toward confirming a decision the customer has mostly already made. If your product was not in the comparison, the customer most likely never knew it was an option.

The shortlist forms before the customer ever reaches your website.

Black Friday will test three different approaches to AI shopping

OpenAI, Meta, and Amazon are reaching for the same shopper in different ways.

It is no coincidence that these changes are landing just before the holiday season. In my view, Black Friday is make or break for AI shopping assistants, because open questions such as "help me find the best Black Friday deals on running shoes" are high in volume and high in intent, and the assistant that gives the most trusted answer this season will most likely become the habit for next season. The three largest players are making different bets.

OpenAI is betting on recommendation. ChatGPT compares, recommends, and then sends the shopper to the merchant's own checkout.

Meta is betting on an agent that acts for you. Meta released Muse on September 8, 2026, as a personal agent that connects to email, calendar, payments, dining, and shopping, and that can place orders using a one-time card generated for each transaction. Meta says Muse checks with the user before a purchase. On September 20, 2026, Amazon said it had cut Muse off from buying on Amazon.com, citing that Meta never got its agreement and that Muse does not identify itself when it browses.

Amazon is betting on its own store. In May 2026, Amazon folded its Rufus assistant into Alexa+ and launched Alexa for Shopping, which builds buying guides, compares products side by side, shows a full year of price history, and offers Auto-Buy to purchase an item once it reaches a target price. Amazon says Rufus had more than 300 million users in 2025.

None of these companies publishes how its model chooses which products to recommend, so the honest way to describe the ranking is as likely behavior rather than known rules. What the three approaches share is that the comparison and the recommendation happen inside the assistant, and the brand that is not in that comparison is not in the running.

Last season's numbers show AI already shapes holiday buying

The 2025 holiday data gives a baseline for what to expect this year.


Statistic Source Date What it means
AI and agents influenced 20% of global orders during Cyber Week 2025 Salesforce December 2025 "Influenced" covers product recommendations and service chats, so this shows reach, not that AI caused the sale.
AI and agents drove $67 billion in Cyber Week 2025 sales Salesforce December 2025 This was one week of spending connected to AI, before Muse or Alexa for Shopping existed.
Generative AI traffic to U.S. retail sites rose 693.4% from November 1 to December 31, 2025, compared with the prior year Adobe Analytics January 2026 The growth started from a small base, so the rate is more meaningful than the raw share of traffic.
AI traffic to U.S. retail sites rose 670% on Cyber Monday 2025 Adobe Analytics January 2026 Shoppers used AI on the biggest deal days, which is when open "best deal" questions peak.
Rufus had more than 300 million users in 2025 Amazon, reported by Axios May 2026 This is Amazon's own figure, and it shows the size of the audience now inside Alexa for Shopping.

This is Amazon's own figure, and it shows the size of the audience now inside Alexa for Shopping.

Brands get picked on product information, reviews, and listings

The inputs an AI model can check are the inputs that decide the shortlist.

Because the choice happens inside the assistant, a brand needs to give the assistant what it needs to include you in the comparison and to trust what it finds. Three areas matter most.

Complete, accurate product information. ChatGPT presents products side by side on price, reviews, and features, so any field you leave empty or out of date is a field where a competitor wins by default. That means price, availability, specs, variants, shipping times, and return terms in your product feeds and in the structured data (schema, the code that labels product details for machines) on your product pages. During Black Friday week prices change often, and a sale price in your feed that does not match your product page gives a model a reason to doubt both. [RYAN: add one practical operating number from your experience here, for example how often you recheck feed prices against product pages during sale weeks.]

Reviews, including the conversations on Reddit. AI models summarize what people say about a product, and they most likely give weight to sources where real buyers compare options in their own words. A product with few recent reviews gives the model little to summarize. Read the Reddit threads where your category is discussed, and fix the product or service problems they raise, because those threads will keep being read long after the sale ends.

Correct business listings. When someone asks for a nearby store or an open table, the model relies on listings for hours, address, and phone number. Holiday hours are the most common gap, and a store listed as closed on the day it is open is a store the model will not recommend.

What to expect over the next 6 to 18 months

The pieces in place now point to a few likely changes.

Moving the product display rules into a separate rule set means OpenAI can most likely change how products appear without waiting for a new model release, so expect the format of product answers to keep changing and plan to check it regularly rather than once. Checkout will probably come back through partners and merchant apps rather than through OpenAI's own checkout, since that is the direction OpenAI described in March. Agents that buy for people, such as Muse and Alexa's Auto-Buy, will push retailers to decide which agents they allow on their sites, and Amazon's block of Muse is the first public example of that decision. Finally, more buying decisions will happen without a click to your site, so analytics will most likely undercount the role AI plays, and teams will need to watch AI referral traffic and branded search together to see the full picture.

Camino5's [RYAN: tool name] shows how your products and locations appear in AI search answers today. [RYAN: confirm tool and wording]

How to start today

  1. Ask ChatGPT, Gemini, and Alexa for Shopping the "best Black Friday deal on [your category]" questions your buyers ask, and record which brands are named and which sources are cited, because that is your baseline.

  2. Audit your product feed and product page structured data for price, availability, shipping, and returns, and set a schedule for updating sale prices in both places at the same time.

  3. Update holiday hours and addresses on every business listing before Thanksgiving week, because local recommendations depend on them.

  4. Read the Reddit threads and review sites for your category, and fix the issues buyers raise, because models summarize what they find there.

  5. Decide how your site will treat AI shopping agents that try to buy on a customer's behalf, so the decision is made before the traffic arrives.

  6. Track AI referral traffic separately from organic search so you can see how much of your holiday demand starts in an AI answer.

The brands that get picked this season will most likely be the ones whose information was easiest for a model to check and trust.

Sources

Ryan Edwards
Next

B2B SaaS Content in AI Search: What 10,382 Keywords and 350,000 Articles Show