AI Sends Some Shoppers to Fake Stores. Does It Send Yours to the Real One?
- Author
- By Paz.ai Team
- Publication date
- Published
- Reading time
- 4 min read
Ask ChatGPT, Gemini or Perplexity where to buy a brand and about one answer in eleven points the shopper at a site the brand does not run. That is the headline from a study by brand protection firm Podqi, published with The Verge in mid November. Across 6,795 answers about 151 brands, 9 percent showed a risky site, and the engines recommended those sites 6.3 times as often as they warned against them.
Most AI visibility work asks one question: does the assistant name us when a shopper asks what to buy. This study raises a second one that matters more in the four weeks before Christmas. When the shopper already wants your brand and asks where to get it, does the answer send them to your store?
What the study found, and what it does not say
Podqi asked the three engines the same five questions about each brand on the same days, in two waves during September, then reviewed every domain the answers named. A site counted as risky if it appeared to impersonate the brand, sell suspected counterfeits, or was dead with nothing to buy. Of 718 risky domains, 351 were fakes and 367 were dead.
The engines differed. Perplexity showed a risky site in 15 percent of answers, ChatGPT in 8 percent and Gemini in 4 percent. When a risky site did appear, 57 percent of answers recommended it as a place to shop, 34 percent listed it without comment and 9 percent warned the shopper. ChatGPT warned most often, in about a quarter of its risky answers.
The question shaped the risk. "Does the brand ship internationally, what are their regional websites" produced a risky site 16 percent of the time, the highest of the five. "What are the top websites for buying the brand" came next at 11 percent. The Verge reports that answers about smaller brands carried risky sites about twice as often as answers about large ones.
Read the limits before acting on the numbers. The prompts were written by the study authors, not drawn from real conversations. The firm sells brand protection and one of its clients was in the sample. Fieldwork was a September snapshot, and the risk rate fell from 12 percent in the first wave to 7 percent in the second. Dead domains count as risky even though nobody bought anything there. Google AI Mode and Copilot were not tested. The study does not measure lost orders.
The decision this puts in front of you
Before Black Friday, decide whether your team spends a few hours checking the where to buy answers for your own brand, and who owns the fixes if a lookalike shows up. The cost of ignoring it is specific: a shopper who already chose you lands on a fake or a dead page, and you see nothing in analytics because the visit never reached you.
This is a different job from getting into a product shortlist. The inputs overlap, but the check is about your brand name, your domains and your official stores, not your product attributes. The person who owns the brand site and the person who owns the catalog both need to be in the room.
The practical work
1. Run the five questions for your brand
Ask ChatGPT, Gemini and Perplexity the study's questions with your brand name: top websites for buying, where to find it online, where else to buy it besides your site, and whether you ship internationally with regional websites. Log every domain each answer names, which engine named it, and whether the answer recommended it, listed it or warned about it. Flag every domain you do not operate or authorize.
If you sell into more than one country, repeat the regional question for your top three markets. That was the riskiest question in the study, and 36 percent of the risky sites in those answers used a country ending or a region name.
2. Publish one page the engines can copy
Give the assistants a single crawlable page that lists every official regional store and every authorized retailer, and says plainly which countries have no regional store. Link it from the footer and the help center. The study's authors recommend this because the regional question is where invented and lookalike addresses slip in, and a direct answer on your own domain is the easiest thing for an engine to cite.
Make sure that page and your product pages are readable by the AI crawlers. Google's documentation on AI features and your website is the reference for its surfaces, and the access checks in our crawler audit cover the OpenAI and Perplexity bots. A bot wall on your store makes a lookalike with open pages the better source.
3. Make your own offer the obvious answer
Assistants name stores they can verify. Keep price, availability and ship to countries explicit on product pages, send a current feed to Google Merchant Center, and keep your Shopify catalog or direct feed flowing to ChatGPT so the engine has your offer in front of it. How ChatGPT picks products and then merchants explains what that feed supplies. Where a platform lets you claim a business profile, claim it, so the engine has an owner record to match against.
4. Inventory your domains
Half of the risky domains in the study were dead, and a dead domain can be registered by anyone. List the country code domains you own and the ones you do not, keep old brand domains renewed and redirected to the main site, and watch for region and add on variants such as a country prefix or a shop suffix on your name. If a domain you abandoned still appears in answers, that is the first one to recover.
What to leave for later
Do not rewrite product descriptions in response to this study. It tested where engines send shoppers, not which products they pick, and the two questions have different inputs. Leave takedown services and monthly monitoring until the first check shows a problem. Skip Copilot and Google AI Mode for this pass, since the study did not test them, and add them when you repeat the check after the holidays.
Treat the percentages as a September reading from one vendor's prompt set, not as your exposure. Your own check across the three engines is the number that matters.
How Paz handles this
Paz tracks the shopping questions that matter for your brand, including where to buy and regional availability, across ChatGPT, Google AI Mode, AI Overviews, Gemini, Claude, Grok and Copilot, and reports which pages each answer cites. When an answer points away from your store, a Paz expert traces the cause to the pages, feeds or missing facts behind it, ships the fix within your approved guidelines, and measures the change against the baseline.
If you want a read on where the assistants send your shoppers before Black Friday, talk to us.
How AI-ready are your products?
Check how ChatGPT, Google AI, and Perplexity evaluate any product page. Free score in 30 seconds.