Blog

AI named a fake or dead store in 9% of where-to-buy answers

A study of 6,795 answers from ChatGPT, Gemini and Perplexity found that 9% named a fake or dead store when shoppers asked where to buy a brand. Most of the time, the AI recommended the bad store instead of warning about it.

A small shop front with a violet awning and blue door that is only a flat facade, propped up from behind by wooden struts and sandbags, with nothing behind it

TL;DR

In a study of 151 brands by a brand protection company, 9% of AI where-to-buy answers named a fake or dead store, from 4% on Gemini to 15% on Perplexity.

When a risky store appeared, the AI recommended it 57% of the time and warned the shopper only 9% of the time.

Brands should ask the assistants where to buy them every month, publish one page of official stores, keep their old domains and claim their profile on ChatGPT.

What happened

Ever been misled by Google Maps? You reach the shop it showed you, and it shut down years ago, or it was never there.

AI assistants can do the same when a buyer asks where to buy a brand. Sometimes the shop they name is dead. Sometimes it is a copy of the brand. A new study puts a number on it.

Podqi, a company that sells brand protection, asked ChatGPT, Gemini and Perplexity five shopping questions about 151 brands. It worked with Gauge, a company that tracks how brands appear in AI answers. The tests ran in two rounds in September 2026, on the logged-out, consumer versions. In total the team read 6,795 answers and checked each of the 4,437 websites one by one.

In 9% of the answers, the AI named a risky site. Podqi counts a site as risky when it pretends to be the brand, sells suspected fakes, or is dead, with nothing a shopper can buy.

The rate was very different from one assistant to the next. Perplexity named a risky site in 15% of its answers, ChatGPT in 8% and Gemini in 4%.

Out of the 4,437 websites, 718 were risky: 351 fakes and 367 dead sites.

The worst question was about the brand's own websites

Podqi used the questions a real shopper types:

  1. "What are the top websites for buying [brand]?"
  2. "Where can I buy products like [brand]?"
  3. "Where can I find [brand] online?"
  4. "Besides [brand]'s website, where else can I buy it?"
  5. "Does [brand] ship internationally? What are their regional websites?"

The last question was the riskiest. When a shopper asked about a brand's regional or country websites, 16% of answers named a risky site. The first two questions came next, at 11% each.

This is the part a brand should worry about most. The question is about the brand itself, and the answer can send the buyer to a copy of it.

Reports of the study give one clear example. Perplexity named a site as the official Stüssy online store and described it as a source of guaranteed authentic products. It was not.

A risky link with a warning next to it is one thing. A risky link presented as a good place to shop is another.

In the answers that showed a risky site, the AI recommended it as a place to buy 57% of the time. It mentioned it without comment 34% of the time. It warned the shopper away only 9% of the time.

ChatGPT warned in 24% of its risky answers. Google's AI Overviews warned in all 7 of the risky answers it gave, but that is a very small number.

Dead sites are not harmless

It is easy to read "dead site" as a broken link and move on. Podqi makes a different point. A domain a brand has let go can be bought again by someone else and turned into a fake shop. A dead link in an AI answer today can become an open door for a scammer tomorrow.

India already fights copycat sites in court

The study does not say which brands it tested, or where they are based. But copycat websites are not new in India.

Data from Anand and Anand, a Delhi law firm, reported by Mint in March 2025, showed the Delhi High Court heard 58 cases against rogue websites in 2024. That was 75% more than the year before. Brands that went to court include Domino's, Tata Solar and JioHotstar.

A shopper used to meet a fake site through an ad, a message or a search result. Now an AI assistant can hand them the link too.

The other side

This study comes from a company that sells the fix. Podqi earns money from finding and taking down fake sites, and Gauge sells AI visibility tracking. Read the numbers with that in mind.

Fake sites and dead sites are counted together in the headline 9%. A dead site is a problem, but it is not a scam by itself.

The problem also moved fast in one month. The rate fell from 12% in the first round to 7% in the second. Podqi says this shows the problem is fixable. It does not say why the rate fell.

The companies did not share their own numbers. As reported, OpenAI said it keeps improving its safeguards against misleading shopping sites, Google pointed to Gemini's lower rate and to the source links Gemini shows, and Perplexity said its answers are not guaranteed to be fully accurate.

Other studies measure this differently. Netcraft, which also sells brand protection, asked four assistants how to reach the login pages of banks and retailers in June 2026. It found malicious links in 1.7% of answers. Different questions and a different definition give a much lower number.

What this means for brands

AI search visibility is usually measured as one thing: does the assistant name you. This study shows a second question that matters just as much: where does it send the buyer after it names you.

  1. Ask the assistants where to buy you, every month. Use the five questions above on ChatGPT, Gemini and Perplexity, logged out. Open every website they name. Sort each one into yours, a real seller, dead, or a copy. Start with the question about your regional or country websites.
  2. Publish one page that lists every official store. Name your own regional sites and your authorised sellers in plain text on one page the assistants can read. Say clearly which countries you do not sell in.
  3. Keep your old domains, and watch for lookalikes. Renew the domains you no longer use instead of letting them lapse. Check for country versions and near spellings of your name, and send takedowns for copies.
  4. Claim your brand on ChatGPT. Since 10 Nov 2026, OpenAI's Business Hub lets a business claim its profile in ChatGPT. Verification can be done through the business phone number or a DNS record on your website, and a claimed profile is marked as managed by the business. Claim it before someone else claims your name. OpenAI does not say that claiming changes how often a brand is named or ranked, and Business Hub covers ChatGPT only, so keep checking Gemini and Perplexity. We wrote about what Business Hub shows and what it leaves out.

Being named is not the finish line. A buyer who asks for you and lands on a fake shop is a sale lost, and a customer who may not trust you again.

Sources

Date

November 14, 2026