We asked ChatGPT where to buy 20 Black Friday products in the US. It got about a third of the prices wrong.

Ask ChatGPT where to buy a KitchenAid Artisan stand mixer in the US and it may send you to Best Buy at $219.99. On the day we asked, Best Buy was charging $349.99. The $219.99 belonged to a refurbished bowl-lift mixer listed on the same page.

On November 4, 2026, we asked ChatGPT where 20 popular Black Friday products were cheapest in the US, twice each, and checked every answer against the retailer's own website. When it recommended one of the NRF Top 100 retailers, the price was wrong 13 times out of 38. And 37 of the 40 answers had an ad underneath.

This matters in the weeks before the biggest sale of the year. Adobe expects traffic from AI tools to US retail sites to grow 130% this holiday season. Below is what ChatGPT got wrong, where the wrong prices came from, the ads that now sit under the answers, and what to check before Black Friday on November 27.

This repeats our Australian study from October with US products, US retailers and the same question.

How we ran the test

We asked the question shoppers ask before a sale, about 20 well-known products, and checked every answer by hand.

• The question: “Where is the cheapest place to buy the [product] in the US right now?”, typed the same way every time.
• The products: 20 products US shoppers buy over Black Friday, fixed before the first ask. Eight come from Adobe's US best-seller data for the 2025 holiday season and its 2026 forecast. The other 12 were also in the Australian study, so the same product can be compared across both markets.
• The retailers: we scored the answers against the NRF Top 100 Retailers 2026, ranked by 2025 US retail sales. Banners count toward their parent, so Sam's Club counts toward Walmart.
• The set-up: a free ChatGPT account with saved memory and chat history turned off, a New York connection and a new chat for every question. Each product was asked twice on November 4, 2026, 23 days before Black Friday. Most people use the free version: OpenAI reported 900 million weekly users and more than 50 million paying subscribers in February 2026 (Digital Information World).
• The check: the same day, we opened every store each answer recommended, using the link in the answer where there was one, and recorded the price, the seller and the condition for the exact product asked about.
• The rules: a price counted as right if it was within $1 of the store's own price, for the item asked about, sold by the store. Marketplace sellers, refurbished units, card offers and members' prices didn't count as the store's price.

About a third of the prices were wrong

When ChatGPT recommended an NRF Top 100 retailer and we could check the store's price, it was right 25 times and wrong 13 times out of 38. Seven wrong prices were too low, two were too high, and four quoted a marketplace seller's price as the store's own.

Depending on how you count marketplace sellers and listings that couldn't be bought, the result ranges from 9 of 38 to 16 of 40, which is why we describe it as about a third.

Why the 13 wrong prices were wrong

How far off they were

Macy's doesn't sell the plain Dyson V8, so its gap is measured against its cheapest V8, the V8 Cyclone. The four marketplace mix-ups quoted a seller's real price, so they have no gap to show.

The retailers it recommended

Three of Best Buy's six wrong prices were marketplace listings on Best Buy's own site. Outside the top 100, five of eight checkable recommendations were wrong, including GameStop and J2Games selling a refurbished or pre-owned PlayStation Portal as new.

One cent decided the recommendation

Target was named in 25 answers and recommended in none. Where its price was close, it sat just above the store that won: the LEGO Orchid at $39.99 against Walmart's $39.98, which the answer called “only one cent more”, LEGO Pikachu at $174.99 against $174.98, and AirPods Pro 3 at $199.99 against $199. On a “cheapest” question, the lower number got the recommendation every time we saw it.

Every answer, product by product

Where the wrong prices came from

Six of the 13 wrong prices started on the retailer's own website. The rest came from deal articles and price trackers that copy retailers' prices, as they did in Australia. In each case, ChatGPT took the cheaper number and gave it the store's name.

Listings on the retailer's own site

• Marketplace sellers shown as the store. For the Garmin Forerunner 265, both answers sent shoppers to Best Buy at $365.41, and the second called it the “lowest verified advertised price”. It was sold by marketplace seller Mani International with $99.99 shipping, $465.40 delivered, which is $15.41 more than Target and Garmin were charging. The $117.29 JBL Flip 7 at Best Buy was a seller rated 3.72, and the $539 Switch 2 at Walmart was an international version with a plug adapter. It only appeared when we sorted Walmart's results by price.
• Another product on the same page. ChatGPT said Best Buy's $219.99 KitchenAid was “the genuine Artisan 5-quart model”. It was a refurbished 5.5-quart bowl-lift mixer listed beside the new Artisan at $349.99. Macy's $279.99 belonged to a Dyson PencilVac.
• A card offer quoted as the price. For the Kindle Paperwhite, the answer said Amazon was $149.99. Amazon's page said $199.99, with $149.99 only “upon approval for Amazon Visa”. Another answer listed Target's 5% Circle Card discount as an extra, so the same kind of offer was handled two ways.

Old deal prices and price trackers

Half of all quoted prices came from comparison sites and price trackers.

Answers often cite several sources at once, so this is a guide to the mix, not an exact attribution.

• A dated price that was already out of date. For the AirTag 4 pack, the first answer quoted Amazon at $89 from a deal “reported November 2”. The second quoted $69.99 from a deal listing “verified on November 4”. Amazon was charging $79.99 both times.
• A deal that moved store. B&H sold the Bose earbuds at $229. ChatGPT attached $229 to Focus Camera in both answers and to Best Buy in the second. Both were charging $299.
• A 90-day low read as today's price. Both Oura Ring answers opened with eBay at “currently $267”. The tracker it cited showed $267 as its 90-day low and $309.98 as the cheapest new listing that day.
• Sources most retailers have never heard of. The Price Pickr, Price Bug, Plumzoo, BricksValue, Brickista, Brickfact and askprice.io were all cited for prices.

Refurbished as new, out of stock as in stock

Feeds help, if they carry the right details

Six links came through a retailer's own product feed to ChatGPT, tagged utm_medium=feed. Five quoted the right price. The sixth was J2Games, whose feed sent the pre-owned Portal's price with nothing to show it wasn't new. OpenAI says shopping data comes from “first-party and third-party providers” and lets merchants apply to send a feed directly (OpenAI Help Center, feed specification).

Feeds also open the door to smaller stores. Swiftronics, an online store with no reviews and two units left, took the top “lowest listed price” card for the Sony headphones through a link tagged openai_catalog.

The labels don't track accuracy

Wrong answers carried “Best verified price”, “cheapest confirmed price” and “verified on November 4, 2026”. The caveats, such as “check the seller” or “verify at checkout”, sat in small print below a green label.

ChatGPT reads your product page, the marketplace listings on your site, your feed and the deal sites that copy your prices. When a shopper asks for the cheapest price, the lowest number on any of them can become your price in the answer.

Ask again and the answer changes

For 11 of the 20 products, asking the same question in a new chat gave a different store or a different price. The second answers were more accurate, with 5 wrong prices out of 20 against 8 out of 18 the first time. They fixed four answers and broke one.

Getting the same answer twice didn't make it right. Nine products held, and three of those held on a wrong answer: the PlayStation Portal at GameStop, the Bose earbuds at Focus Camera and the Garmin at Best Buy.

If you track how ChatGPT presents your products, ask each question several times and record every answer. A rank tracker records one result, and a shopper who asks ChatGPT gets one of several possible answers.

Retailers can now pay to appear under the answer

No retailer can pay to change what the answer says. OpenAI says “Ads do not influence the answers ChatGPT gives you” (OpenAI). The space directly under the answer is a different matter. OpenAI began testing ads in the US in February 2026 and opened self-serve buying with cost-per-click bidding on May 5. Ads show on the Free and Go tiers, which is most of ChatGPT's 900 million weekly users.

In our test, 37 of the 40 answers carried an ad. Here is what that looked like next to the wrong prices.

A retailer missing from the answer still showed up under it

The first LG C6 TV answer didn't mention Best Buy. It sent shoppers to Amazon at $1,349.99, a price Amazon wasn't charging. Directly underneath sat a Best Buy carousel of LG OLED TVs.

The ad was more accurate than the answer above it

Both JBL Flip 7 answers sent shoppers to Best Buy at a price Best Buy wasn't offering: $117.29 from a marketplace seller the first time, and a $99.43 listing that couldn't be bought the second. Both times, Best Buy's own ad below the answer showed the speaker at $129.95, its real price. For a shopper reading the screen top to bottom, the correct Best Buy price was in the ad.

It also works the other way

Under both Sony WH-1000XM6 answers, the answer had Best Buy right at $378. Best Buy's carousel underneath showed the same headphones at $458, the “comparison value” on its product page. The shopper saw two Best Buy prices for one product, $80 apart.

The most frequent advertiser was a price-checking tool

Capital One Shopping, a tool that compares prices and tests coupons, advertised under 12 of the 40 answers, more than twice as many as any other advertiser. Six of those 12 answers recommended a wrong price or the wrong item. Shoppers who asked ChatGPT for the cheapest price were offered another tool to find it.

Rivals can buy your shopper's question

OpenAI says ads are matched to the topic of the conversation, past chats and past ad interactions. In our test that put SharkNinja under both Dyson answers, Spotminders, a rival tracker brand, under the AirTag, Temu under the Bose earbuds and Back Market's refurbished AirPods under the AirPods Pro 3. Almost half the ads, 17 of 37, were product carousels from a single advertiser, the format Digiday reported OpenAI launching in August.

What it means for your Black Friday plan

You can't pay to correct the answer. You can pay to put your real price directly under it, and if your ad shows a comparison value or an old price, you pay to show a second wrong one. Search Engine Land reports that retailers can now build ads from their product feeds, and that the same product data informs both answers and ads. OpenAI's own pages don't say this, but if it holds, a missing seller name or condition in your feed shows up in both places.

Our account asked 40 shopping questions in a row, so another account may see fewer or different ads. For how to approach the ads themselves, read ChatGPT Ads and eCommerce: What Merchants Should Actually Do Now.

What to do before Black Friday

You have under three weeks. Start by measuring what shoppers are told, because it shows which listings and sources to fix.

Measure six things

Pick 10 to 20 hero products for the sale. Ask ChatGPT where each one is cheapest, three times each in new chats, on a free account, and record:

Use one shipping ZIP code on every site, and click through from inside the answer, because the page the link opens is part of what you're measuring. Our 15 copy-and-paste prompts are a ready-made starting set, and How do you measure AI visibility? covers tracking over time.

Best practices so far

AI shopping answers are still new, and there's no way to guarantee the price ChatGPT quotes. Based on what we've seen in our tests, these practices give it the best chance of finding your right price. Product Feed Optimization for AI Shopping covers how to structure the full catalog.

1. Make your own price the one in the page data. On pages with marketplace sellers, open-box or refurbished units, name the seller and condition for every other offer, using the seller, itemCondition and availability fields of each Offer. Four of our wrong prices were marketplace sellers quoted as the store.
2. Name the exact model. Model, generation, size and bundle belong in the title and the product data. Two products on one page produced two of our wrong prices.
3. Send the same details in your feed. Include condition, and update prices whenever they change, including when each sale starts and ends. Five of the six feed links in our test quoted the right price. The sixth sent a pre-owned price with nothing to show it wasn't new. Do I need a separate product feed for ChatGPT? answers the most common question about feeds.
4. Keep stock current. Three recommended listings couldn't be bought. Availability in the feed should change when the stock does.
5. Put offer terms next to the price. Card offers, member prices and codes need their terms beside the number, on the page and in feeds. ChatGPT quoted an Amazon Visa offer as Amazon's price.
6. Date every deal you brief out. Deal articles and trackers keep quoting a sale after it ends. A dated offer is easier to recognize as expired.
7. Check your ChatGPT ads. If you advertise, make sure the ad shows the price you charge, not a comparison value.

Your Black Friday timeline

What nobody controls

Which source an answer picks, and whether it changes when someone asks again. That's why the measuring has to continue through sale week.

For why ChatGPT quotes the wrong price in the first place, and what our September study of 51 Australian retailers found about product pages, schema and stock, read We checked what ChatGPT tells shoppers about 51 retailers.

All of this depends on the price being the same everywhere it's published: the product page, the marketplace listings, the feeds, the ads and the emails. That's an operational question as much as a marketing one, and it's how Core dna's eCommerce platform is built, with one source that the site, the storefronts and the feeds all read from. For groups running several brands, it's a platform decision.

About this test

• Two asks per product, one day, one free account, one assistant. 40 answers on November 4, 2026. Prices change, and the figures describe what we saw that day.
• Every recommended store was checked on its own website the same day, for the exact product asked about, except Newegg for the Bose earbuds and Amazon for the Breville and LEGO Pikachu.
• A price counted as right if it was within $1 of the store's own price, for the item asked about, sold by the store.
• Members' prices and card offers were recorded separately. Quoting one as the price everyone pays counted as wrong.
• A marketplace seller's price presented as the store's counted as wrong, even where the answer advised checking the seller.
• Refurbished or pre-owned prices presented as new counted as the wrong item. Where the answer said it was recommending a related model (the Xbox Series X digital edition, the Dyson V8 Cyclone), the item counted as a match.
• Four recommendations were left out of the price counts because the listing couldn't be bought or confirmed: Micro Center (Xbox), Best Buy (JBL, second answer), Staples (Portal) and eBay (LEGO Orchid).
• Store locations were set to New York and Brooklyn ZIP codes. Some retailers price by location.
• Ads were recorded but didn't count as a store being named or recommended.

About this study. The AI Retail Readiness Study is run by Core dna, the operations platform for multi-property operators. Our eCommerce platform runs product pages, storefronts and product feeds for retailers, and we run tests like this to see how our clients' products and prices are picked up by AI shopping tools, and where the gaps are between what a retailer publishes and what a shopper is told. We'll run the same 20 products again in Black Friday week, in the US and Australia, and publish what changed.