What is shopping research in ChatGPT, and how does it build a buyer's guide?

Comparing products online can eat an evening: a dozen tabs, conflicting reviews, prices that change by the hour. Yesterday, Monday, November 24, 2025, OpenAI added a ChatGPT mode built for that chore. It is called shopping research, and instead of a quick answer it spends a few minutes researching and hands you a personal buyer's guide.

It is rolling out now on the phone apps and the web for logged-in users on the Free, Go, Plus and Pro plans. With the holiday season starting, OpenAI is offering nearly unlimited use on all plans through the holidays. CNBC notes the timing, just as holiday spending picks up.

What OpenAI released

OpenAI says hundreds of millions of people already use ChatGPT to find and compare products. Shopping research is designed for the harder end of that: choices with many constraints and trade-offs. OpenAI's examples include finding a quiet stick vacuum that suits a small flat, a choice between three bikes, and a gift for a four-year-old niece who loves art.

It does best, OpenAI says, in categories heavy with specifications, such as electronics, kitchen appliances, beauty products, garden and home goods, or outdoor and sports gear. For a simple question, such as the price of one item, a normal ChatGPT reply is quicker and still the better choice.

The feature also appears in ChatGPT Pulse, currently available to Pro users, which can suggest a buyer's guide on its own. If you had been discussing e-bikes, for instance, a Pulse card might later propose accessories.

How it works, step by step

When you ask a shopping question, ChatGPT offers shopping research automatically; you can also choose it from the + menu. Then:

1. You describe the need. A visual panel opens, and ChatGPT asks clarifying questions: budget, who the item is for, which features matter. If ChatGPT's memory, the details it keeps about you from past chats, is on, it can use those too; someone who talks about gaming may get laptop picks suited to games.
2. It researches while you watch. In the background it reads retail and review sites for price, stock, reviews, specifications and photos, and shows candidate products as it finds them.
3. You steer. Mark items "Not interested" or "More like this," and the search adjusts in real time.
4. You get a guide. After a few minutes, it delivers a summary of the top picks, how they differ, the trade-offs and current details from retailers.

To buy, you click through to the store's site. OpenAI says that later, for stores enrolled in Instant Checkout, the ChatGPT payment feature it introduced in September, you will be able to buy without leaving the chat.

The model behind it

Shopping research does not use the general ChatGPT model. It runs on a version of GPT-5 mini, a smaller, faster member of the GPT-5 family, which OpenAI further trained for shopping with reinforcement learning: the model tries many research tasks and is rewarded when it does them well, so it gradually learns the behavior. Here the rewarded behaviors were reading trustworthy sites, citing reliable sources and combining facts from many pages into one accurate answer.

To measure progress, OpenAI built a test of difficult product searches with many requirements, and scored product accuracy: the share of suggested products that truly meet every requirement, such as price, color, material or specifications. OpenAI says the model does better on this than its other models, but is not perfect.

Here is a request that gives the research a lot to work with:

I need a stroller for city sidewalks and some gravel paths. It must be
light, fold with one hand and fit in a small car trunk. My toddler naps
in it, so a good recline and canopy matter. Budget: up to $450. Compare
the three best options in a table with weight, folded size and price.

Trust, limits and why it matters

OpenAI makes several promises about neutrality. Your chats are not shared with retailers. Results are organic, meaning nobody pays to be included; they come from publicly available retail sites, and the system tries to avoid low-quality or spammy pages. Merchants who want to appear can follow an allowlisting process so the system can read their pages.

The main limit is accuracy. Prices and stock change constantly, and OpenAI warns that shopping research can still get details like price and availability wrong, so it advises checking the merchant's own page before buying.

Why it matters: this is another example of ChatGPT moving from answering to doing research for you, much like deep research did for reports. It also shows OpenAI's push into commerce, which CNBC links to the Instant Checkout launch earlier this fall. For shoppers, the value is time saved on comparison; the risk is trusting a neat summary without checking it.

How do you get a useful buyer's guide from ChatGPT shopping research?

Give it constraints, not just a category. "A laptop" produces a generic list; "a 15-inch laptop under $1,000 for photo editing and light gaming, under 2 kilograms" produces a real shortlist. Answer the clarifying questions honestly, including your true budget.

Use the feedback buttons while it works. Rejecting a few products early teaches the search more than any amount of rewording at the end. If you have memory turned on, glance at what ChatGPT assumes about you, and correct anything outdated.

Before you buy, open the retailer link for your top choice and confirm the price, stock, return policy and model number. Treat the guide as a well-researched starting point, and make the final call yourself.