How do we find and fix wrong prices or product facts that AI assistants tell shoppers?
Our answer
Check monthly and fix the sources assistants read. Ask each assistant about your best sellers, recheck after any price change, and make feeds, product pages and retailer listings agree. Keep prices in raw HTML and report wrong ChatGPT product cards.
- Confidence
- Medium
- Strongest proof
- 2 controlled tests
- 7 official rules, 2 datasets, 1 practitioner, 1 opinion or claim
- Sources
- 9 sources
- 13 statements
- OGPA
SEWMEL
New here? Take the 1-minute tour
Why this is the answer
4 reasons and our judgmentErrors cluster on facts that changed. In a test of 15 assistants on public facts such as prices and laws, 242 of 1,500 scored answers were outdated or wrong, all on changed facts, and none of 195 on unchanged facts were (Search Engine World).
Some wrong prices start on someone else's page. Which?, a UK consumer group, found at least one problem in 68% of ChatGPT's lowest-price results for 25 products, some apparently from out of date comparison sites. OpenAI says price changes can reach ChatGPT with a delay.
The platforms take prices from feeds and pages. Google checks Merchant Center prices against your page's HTML and structured data and disapproves mismatches. Microsoft says out of date product data can drop you from AI recommendations.
The size of the problem is disputed. Product.ai, an AI shopping company, found 86% of its 220 shopping questions drew a factual conflict, and wrong prices on high-cost items missed by a median of $300. Google says the test used developer software.
Our judgment
We’d check monthly and again after every price, spec or product change, giving most attention to the facts that changed, because errors in the tests clustered on changed facts and one agency’s monthly routine catches them. We’d fix your feed and product page before third-party listings, because those are the inputs OpenAI, Google and Microsoft document, and Which? traced some wrong ChatGPT prices to stale comparison sites. We’d treat ChatGPT’s report button as an extra, not the fix, since OpenAI promises no correction and arguing in the chat changes only that conversation.
Our judgment, built from Search Engine World, ESBO Ltd, OpenAI, Google, Microsoft and Which?: our reading of these sources, not a finding from any one of them.
What the sources say
13 statements from 9 sourcesGoogle may disapprove Merchant Center products whose feed price does not exactly match the landing page HTML. Prices loaded later by JavaScript cause errors, and mismatches may lead to account suspension.
“Prices reflected in the HTML need to match exactly the prices uploaded in Merchant Center.”
Google says free product listings can show on Search, Images, Lens, YouTube, Gemini and the Shopping tab, but are not guaranteed, and Merchant Center data such as prices should match your store.
“Your products aren’t guaranteed to show across Google, as we rely on the data you provide to us to match your products with what customers might be searching for.”
OpenAI says ChatGPT shopping prices come from third parties, may not be the lowest available, and can lag behind merchant updates. Shopify merchants are already integrated, and users can report wrong product details.
“Prices shown in ChatGPT’s initial response typically reflect the price from the first merchant listed, which may not be the lowest available price.”
Writesonic, an AI visibility tracker, says to audit AI shopping answers by running grouped prompts logged out on ChatGPT, AI Overviews, AI Mode and Gemini, then fix price, stock and GTIN gaps. No results data.
“Most brands find these issues by accident. Someone runs a prompt, sees a wrong price, and fixes that one product.”
OpenAI has opened Businesses in ChatGPT in beta: businesses can claim a listing, manage their information and see how people engage with them, and eligible ones can build an agent that answers customer questions.
“Claim your business, manage your information, and get insights on how people engage with you.”
OpenAI says product feeds in ChatGPT are open only to approved partners, recommends concise, factual product copy, and says optional fields can improve answers but aren't required.
“Use concise, factual copy that helps users understand products.”
Shopify's Knowledge base now sits in the admin under the Agentic sales channel, replacing the standalone app after January 15, 2027. Shopify drafts store facts for AI agents, and nothing goes out until you publish it.
“Nothing Shopify drafted reaches AI agents until you publish it.”
In a Which? test of 25 products at UK prices, ChatGPT got the lowest price and retailer right for only eight, and 68% of results had at least one problem, such as a wrong price or the wrong product.
“ChatGPT correctly identified the lowest price and retailer for only eight products.”
Product AI asked ChatGPT, Gemini, Claude and Perplexity 220 shopping questions in nine categories and found errors like MacBook prices off by $300. Google says the test used developer Gemini, not its consumer app.
“Everything from MacBook pricing being off by $300 in some cases, to SPF sunscreen being off by 10 points, to earbud battery life being off by six hours”
Boris Dzhingarov, whose agency sells digital PR, says arguing with an AI assistant fixes nothing. Put dated current facts on your site and in fresh third-party coverage, and check two assistants monthly.
“Arguing inside the chat changes that one conversation and nothing else”
Product.ai, an AI shopping company, asked four AI assistants 220 shopping questions: 86% produced a factual conflict across answers. Wrong prices on high-cost items were off by a median $300.
“There was almost no middle ground between exactly right and badly wrong.”
Microsoft says AI shopping assistants rely on structured product data, so incomplete or stale data can drop you from recommendations entirely. It points to its Merchant Center feed and Clarity's AI Visibility reports.
“If your data is incomplete or out of date, you're not just lower in the results; you might eliminate yourself from the recommendation entirely.”
Andre Alpar asked 15 AI assistants about 44 facts that had changed: 242 of 1,500 scored answers (16%) were outdated or wrong, with no false alarms on controls. He advises dated pricing and product pages.
“The outdated values arrived without cutoff warnings, often wearing fresh year labels”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it3 figures from cited sources
- Where each source standsA chart placing each source on the question
- The strongest disagreementThe best case against the answer, from a cited source
- How sure we areWhat its confidence rests on, and what would change the answer