If you sell online, the question is simple. When someone asks ChatGPT to help them choose a product, will your store show up? ChatGPT shopping research, launched by OpenAI on 24 November 2025, turns a shopping question into a buyer’s guide built from public store websites. OpenAI says the results are organic. Stores can ask to be allowlisted (added to an approved list) so their products can appear.
Below I explain how the feature works, what OpenAI has said about how it picks products, and the steps I’d take this week as a store owner, right in the middle of Black Friday traffic.
What ChatGPT Shopping Research Is
Shopping research is a mode inside ChatGPT for product choices that need more than a quick answer. ChatGPT does not reply in seconds. It asks a few questions, reads store and review pages for a few minutes, and then gives back a clear buyer’s guide.
OpenAI announced it in its post Introducing shopping research in ChatGPT. The main launch details are:
- Who gets it: logged-in users on Free, Go, Plus and Pro plans, on web and mobile.
- Usage limits: OpenAI is offering nearly unlimited use on all plans through the holidays.
- The model: a version of GPT-5 mini, trained further with reinforcement learning (a way of training by reward) just for shopping tasks.
- Where it works best: OpenAI names electronics, beauty, home and garden, kitchen and appliances, and sports and outdoor.
The timing is not an accident. It arrived in Thanksgiving week, when people compare products the most.
How ChatGPT Shopping Research Works, Step by Step
From the shopper’s side, there are four stages. When I look at a new AI feature, I always map where store data gets used. That is what each stage below shows.
1. The Shopper Starts a Research Session
When someone asks a shopping question, ChatGPT can suggest shopping research on its own. Users can also pick “Shopping research” from the plus (+) menu in the chat box.
OpenAI’s example prompts tell you what kind of question this is for. Finding the quietest cordless stick vacuum for a small apartment. Choosing between three bikes. Finding a gift for a four-year-old who loves art. These are detailed requests with many conditions, not “price of AirPods” lookups.
2. ChatGPT Asks Clarifying Questions
Before it searches, ChatGPT asks about budget, who the product is for, and which features matter. If the user has memory turned on, it can also use what it already knows from past chats.
This matters for sellers. The research is built around clear conditions like “under $300”, “for a small flat” or “quiet”. Pages that state those facts plainly are easier to match.
3. It Researches Across the Web
ChatGPT then reads product pages and other sources. It pulls price, stock, reviews, specs and images. OpenAI says the model was trained to read trusted sites, cite its sources and combine facts from many pages. It also says it tries to avoid low-quality or spammy sites.
While it works, the shopper can steer it. Products appear with options like “Not interested” and “More like this”, and the research changes in real time.
4. It Delivers a Buyer’s Guide
After a few minutes, the user gets a personal guide: top picks, main differences, trade-offs and links to the stores. To buy, the shopper clicks through to the store’s site.
OpenAI has said that, in future, shoppers will be able to buy directly in ChatGPT from stores that use Instant Checkout, where available. If you have not looked at that programme yet, our explainer on ChatGPT Instant Checkout and the Agentic Commerce Protocol covers how it works.
How Products Are Chosen: What OpenAI Has Actually Said
OpenAI has not published a ranking formula. What it has shared is a set of principles across its shopping posts and help pages. Here is what is on the record, and nothing more. I prefer to plan around confirmed facts rather than guesses.
Results Are Organic
OpenAI says shopping research results are organic and based on public retail sites. There are no paid spots inside the buyer’s guide. You cannot buy a place in it.
This matches what OpenAI said before about shopping in ChatGPT. When it expanded product results in ChatGPT search in April 2025, it said products were picked independently and were not ads. When it launched Instant Checkout in September 2025, it said product results are organic and unsponsored. It also said Instant Checkout items do not get a ranking boost.
Matching the Shopper’s Needs Comes First
The whole design is about matching what the shopper asked for. OpenAI measures the feature on “product accuracy”. That means how many of the suggested products really meet the user’s needs, such as price, colour, material or specs.
That tells you what the system is built to do. A product that clearly meets a condition will beat one that might meet it but does not say so.
Source Quality Counts
OpenAI says the model prefers trusted sites and avoids low-quality or spammy ones. It cites where its facts came from, so the shopper can see the pages it used.
When Several Stores Sell the Same Product
OpenAI has not explained this for shopping research. For Instant Checkout, it listed what ChatGPT looks at when several stores sell the same item: stock, price, quality, whether the store is the main seller, and whether Instant Checkout is turned on.
I think it is fair to expect similar logic across ChatGPT’s shopping features. But treat that as an educated guess, not a confirmed rule for shopping research.
Shopper Data Stays With ChatGPT
OpenAI says users’ chats are never shared with retailers. You will see the click and the order, not the chat that led to it.
How Stores Can Appear in Shopping Research
There are three routes. They work together, not as either-or choices.
Request Allowlisting
OpenAI says stores can go through an allowlisting process to make sure they can appear in shopping research results. The announcement links to a sign-up page for stores. It does not give the rules or a timeline. If you sell online, I’d request it now. It costs nothing, and it is the one route OpenAI points to directly.
Be Readable on the Public Web
Results come from public retail sites, so the basics of crawling still apply. ChatGPT’s search features rely on OpenAI’s search crawler (the bot that reads web pages), OAI-SearchBot. OpenAI’s crawler help page says sites that block it will not appear in ChatGPT search answers.
OpenAI has not said which crawler shopping research uses. Still, blocking OpenAI’s search crawler is a clear risk to your visibility. Note that GPTBot is a separate crawler used for model training. You can block it without blocking search.
A common setup that allows search but opts out of training looks like this:
User-agent: OAI-SearchBot
Allow: /
User-agent: GPTBot
Disallow: /Whether you block GPTBot is a business choice. The point is not to block OAI-SearchBot by accident with a broad rule or a firewall setting. In my experience, this kind of accident is more common than people think.
Product Feeds Through the Agentic Commerce Protocol
The shopping research announcement does not mention product feeds. But OpenAI’s Agentic Commerce Protocol documentation describes a product feed format. It lets stores share structured catalogue data (product details in a fixed, machine-readable format) with ChatGPT. That feed is the base of Instant Checkout.
OpenAI has said Instant Checkout will come to shopping research, so preparing a clean product feed is a sensible step. OpenAI launched Instant Checkout with Etsy sellers and said Shopify stores would follow. If you sell on a platform, check what it offers before you build a feed yourself.
What Stores Should Do Now
None of this needs a new skill set. It needs the product details you already have to be complete, specific and easy for a machine to read. Here is the order I would work in.
- Request allowlisting. Use the merchant link in OpenAI’s announcement. Note the date you applied.
- Check your robots.txt and firewall. Make sure
OAI-SearchBotcan reach product and category pages. Check CDN or bot-protection rules too, not just robots.txt. This is the first thing I check, because one wrong rule can hide a whole store. - Write specs as plain facts. Put dimensions, weight, materials, noise level, battery life, compatibility and age range in text on the page. “Whisper-quiet” does not match a request for “under 70 dB”. “68 dB on the low setting” does.
- Keep price and stock correct. OpenAI warns that shopping research can get price and stock wrong, and tells users to check the store’s site. Make sure your page is the one that is right. Out-of-stock items should say so.
- Add and test product schema. Structured data (code that labels price, stock, reviews and IDs like GTIN) gives every system a clean copy of the facts. Our guide to product structured data for ecommerce walks through the properties that matter.
- Answer the comparison questions. Shoppers ask “which is better for a small apartment” or “is this good for beginners”. Add short sections that say who a product suits and who it does not.
- Get your feed ready. If you plan to use Instant Checkout, clean up titles, variants, images and IDs now, in line with the Agentic Commerce Protocol feed format.
Product Page Content That Helps
The pages that do well in a system like this are the ones that remove guesswork. A few simple examples:
- A stick vacuum: list the noise level in decibels, run time per charge, dustbin size and weight. Add a line such as “Suits flats and smaller homes; not designed for thick carpets.”
- A children’s art kit: state the age range, what is in the box, whether the materials wash out, and the box size for wrapping.
- Running shoes: give the heel-to-toe drop, weight per shoe, the surface they are made for, and how sizing compares with your other models.
Reviews matter too. OpenAI lists reviews among the details shopping research collects. Real reviews that mention real use (“fine on stairs”, “too loud for a nursery”) give the model specific details to match.
Most of this is just good practice. If your pages need a wider tidy-up, our ecommerce product page SEO checklist covers titles, descriptions, images and internal links.
A Worked Example: From Request to Shortlist
Picture a shopper who types: “I need a cordless stick vacuum for a one-bedroom flat, under $300, quiet enough to use in the evening.”
ChatGPT might ask if they have pets, if the floors are mostly hard or carpet, and how much storage space they have. Each answer becomes one more condition.
Now picture two stores selling vacuums at a similar price. Store A’s page says “powerful, quiet cleaning for every home”. Store B’s page lists the decibel rating, a 40-minute run time, a 2.4 kg weight and a note that it suits hard floors and low-pile rugs.
Only Store B gives the research something it can check against the shopper’s needs. Store A may still sell the better vacuum, but its page does not prove it. I see this gap all the time: a good product, but a page that does not show it. That is where most stores can improve fastest.
Common Mistakes to Avoid
- Hiding specs in images or PDFs. If a size only appears in a diagram, a system that reads text may miss it.
- Letting prices drift. A page with an old price, or a schema price that differs from the one on the page, makes your data look unreliable.
- Vague marketing copy. “Premium quality” and “best in class” do not meet any condition a shopper sets.
- Blocking all AI crawlers at once. A blanket AI-bot block may also stop the crawler that powers ChatGPT search.
- Expecting a paid shortcut. OpenAI says results are organic. There is nothing to buy in the buyer’s guide.
How to Measure the Impact
Shopping research sends shoppers to store sites to buy, so you should see the traffic in analytics. Check referral reports for visits from chatgpt.com, and watch the landing pages they arrive on.
OpenAI has not announced store reporting for shopping research. For now, referral data and conversion rates from those visits are the best signals you have. Compare them with other channels before you decide anything. A few days of holiday traffic is a small sample, and I would not judge a channel on it.
What Is Still Unclear
Some questions are still open, and I want to be honest about them:
- OpenAI has not published the rules for allowlisting or how long approval takes.
- It has not said how much allowlisted stores appear compared with other public retail sites.
- It has not given a date for Instant Checkout inside shopping research.
- It has not said which crawler shopping research uses when it reads pages.
Plan around what is confirmed, and check back as OpenAI shares more.
Conclusion
ChatGPT shopping research rewards stores whose product details are specific, correct and easy to reach. OpenAI says results are organic, it does not share chats with retailers, and stores can ask to be allowlisted.
Your next step is simple. Request allowlisting, make sure OpenAI’s search crawler can reach your product pages, and rewrite the specs of your best sellers as plain facts. If I had only one hour, that is exactly where I would spend it. Then read our guide to ChatGPT Instant Checkout so you are ready when buying moves inside the chat.
