How do you use shopping research in ChatGPT without being upsold?

Shopping research, the buyer's-guide mode OpenAI added to ChatGPT on November 24, landed days before Black Friday, at the start of the holiday rush, and OpenAI is offering nearly unlimited use on every plan through the holidays. OpenAI says the results are organic: products are not placed in the guide because a seller paid, and your chats are not shared with retailers. So "upsold" here does not mean hidden advertising. It means the ordinary drift of any shopping conversation toward the more expensive model, the extra feature and the matching accessory.

That drift has three common causes, and you control all three: a budget you never stated firmly, a request for "the best" instead of "good enough", and suggestions you accept without asking whether you need them.

Where the drift comes from

A soft budget. Phrases like "mid-range" or "not too expensive" leave room for interpretation, and a research tool trying to find the strongest option will use that room. A number is clearer.

"Best" as the goal. Ask for the best camera and you will get the products with the most features. Ask for the cheapest camera that meets three needs and you get a different list.

Your profile. If memory is on, the research can use what ChatGPT knows about you from past chats. That helps when it remembers you play games, and hurts when it assumes you want the top tier because you once asked about a premium brand. Pro users may also see buyer's guides suggested proactively in ChatGPT Pulse, such as accessories for something you discussed. A suggestion is not a need.

A worked example: a laptop for a student

Start shopping research from the (+) menu, or accept it when ChatGPT offers it for a shopping question, and give it a brief like this:

I need a laptop for a first-year university student.
Uses: writing, web, video calls, streaming. No gaming, no video editing.
Hard ceiling: $700 including tax. Do not show anything above it.
Must have: at least 8 hours of battery, a comfortable keyboard,
under 1.6 kg.
In the guide, include:
1. The cheapest model that meets every must-have.
2. Your recommended pick, and exactly what it adds over option 1
   for the extra money.
3. Any feature you think I might want but did not ask for, listed
   separately with its price difference, not built into the picks.
Give model numbers and the date of each price.

While the research runs, use Not interested on anything over the ceiling or with features the student will not use, such as a gaming graphics card, and More like this on sensible options. Early feedback steers the search better than complaints at the end.

When the guide arrives, ask one follow-up question that keeps the comparison honest:

For your recommended pick versus the cheapest option: would a student
with the uses I listed notice the difference in daily use? Answer yes
or no for each difference, with one sentence of reasoning.

If most answers are "no", the cheaper laptop is the right buy. The outcome will vary by store and week; the point is the structure, which makes every extra dollar justify itself.

Watch the checkout, too

For now, you buy by clicking through to the retailer's site. That is where extended warranties, bundles and "frequently bought together" offers appear, none of which come from the guide. OpenAI also plans to let shoppers buy directly inside ChatGPT through Instant Checkout with participating merchants, which it introduced in September for some sellers. Wherever you pay, the same rule applies: add nothing to the cart that was not on your list before you started.

How do you check prices and reviews yourself?

Start with the price. OpenAI itself warns that shopping research can get details like price and stock wrong, and asks people to check the merchant's site. Open the retailer link for your choice and confirm three things: the exact model number (laptops and appliances often come in several versions under one name, with different chips or storage), the final price including shipping and tax, and the return window. Then search that exact model number at one or two other stores you trust. If the gap is large, find out why before you buy; it may be a different configuration.

For reviews, look at the spread rather than the average. Read the most recent one-star and three-star reviews first, because they reveal the faults that matter after a few months, such as a battery that fades or a hinge that loosens. Be wary of a sudden cluster of short five-star reviews posted within a few days, and give more weight to reviews that describe a specific use like yours. Where possible, add one independent review from a publication that tested the product by hand.

Finally, compare the guide's claims with what you found. If ChatGPT said "10-hour battery" and the manufacturer's page says "up to 10 hours", expect less in real use. The guide saves you hours of searching; the last five minutes of checking are still your job.