Why Does ChatGPT Recommend Unavailable or Nonexistent Products?

Why does it happen?

The last case is the most serious. In June 2026, the Guardian reported that a scam-checking service had found ChatGPT citing two cloned websites for the footwear retailer Russell & Bromley. The retailer had gone into administration and been absorbed by Next, so no official store remained. Shoppers who clicked through risked buying from fraudsters.

How do you check for it?

Brands and retailers can test questions such as:

• "Best [product type] from [brand]"
• "Where can I buy [product name]?"
• "Is [product name] still available?"
• "[Brand] [model] vs [competitor model]"
• "[Brand] new range 2026"

Run them in ChatGPT with web search on and off, and in Gemini, Perplexity and Google's AI Mode. Record products that are discontinued, out of stock, never made, or not sold in the shopper's country, with the cited sources. Also check any shopping results or product cards, and where they link.

What should brands fix?

Discontinued product pages

Do not simply delete pages for retired products. A deleted page leaves old reviews and lists as the main sources. Keep the page, or redirect it to a page that says clearly:

• the product is discontinued, and when;
• the successor or closest alternative, with a link;
• any support information for existing owners.

Mark availability in structured data as discontinued where appropriate. Schema.org includes a "Discontinued" availability value.

Accurate availability in feeds

Product feeds to Google Merchant Center and, where you participate, to ChatGPT's merchant programme should update availability promptly. OpenAI's product feed specification requires an availability field for each item. ChatGPT's shopping results also draw heavily on Google Shopping, so errors in one feed can surface in several places.

A current range page

Publish one page listing the products you currently sell, by line and model, and keep it up to date. It gives assistants a clear answer to "does [brand] make [model]?", and a way to spot invented models.

Naming continuity

When you rename or reformulate a product, say so on the new product page: "Formerly [old name]". That links the old name to the current product.

Retailer and list corrections

Ask major retailers to mark discontinued items and link to successors. Ask publishers of frequently cited lists to update them, especially "best of" articles that still rank.

What if assistants invent products that never existed?

Invented models usually follow your naming pattern. A clear range page, consistent model numbers, and product pages that state the full model name help answers that search the web. Report invented products through ChatGPT's feedback option, with a link to your range page. Answers drawn only from a model's training data may still invent products until newer models are released.

What if answers list defunct competitors but leave you out?

A related pattern affects newer products, especially in software and fast-moving consumer categories. An answer lists several competitors, some of which have shut down or stopped development, and omits your active product. The cause is the same: the sources are old. Lists and reviews written two or three years ago include the products that existed then. Answers from training data reflect the same period.

The fixes are to become part of the current sources:

• get into recently updated category lists and review platforms;
• publish comparison pages that name the alternatives buyers consider now, with dates;
• make launch and release announcements easy to find and cite;
• check that your site is readable to AI crawlers, since a new product with a blocked or script-only site has almost no presence.

Over time, as newer sources outnumber the old ones, answers shift. Search-based answers shift first.

How should defunct or relaunched brands respond?

If a brand has closed, been sold or relaunched:

• keep the official domain active if possible, with a clear notice about where the brand now lives;
• monitor for cloned sites using your brand name, and report them to hosting providers, domain registrars and the relevant authorities;
• report fraudulent sources through ChatGPT's feedback option when they appear in answers;
• tell customers, through official channels, which sites are genuine.

Letting the official domain lapse makes it easier for scammers to fill the gap. Related issues for businesses are covered in why ChatGPT recommends businesses that have closed.

What should shoppers do?

If you are shopping with an assistant's help:

• check that product links go to the brand's official site or a known retailer;
• confirm the product and price on that site before buying;
• be cautious of large discounts on sites with unfamiliar addresses, especially for brands you know have closed;
• search for the model number separately if it seems unfamiliar.

What might an illustrative case look like?

Consider an illustrative skincare brand that reformulated and renamed its best-selling serum a year ago. Asked for its best products, ChatGPT lists the old serum by name, citing two beauty lists from 2024 and a retailer page still showing the old product as "temporarily out of stock". It also lists a "night repair" version that the brand never made. The brand restores the old product page with a "Formerly [old name]" link to the new serum. It adds "formerly" wording to the new product page, publishes a current range page, and asks the retailer and one list publisher to update. A month later, search-based answers mostly name the current serum. The invented night version still appears occasionally in answers without web search.

What do people ask about unavailable products in AI answers?

Why does ChatGPT say a product is in stock when it is not?

Its information may come from an outdated feed, a cached listing or an old article. Accurate, frequently updated feeds reduce the problem.

Should we delete pages for discontinued products?

Usually not. Keep a page explaining the product is discontinued and pointing to the successor.

Can we get invented products removed?

You can report them, and clear range information helps search-based answers. Answers from training data change only with newer models.

How often should we check?

Monthly, and after launches, discontinuations and renames.