Let AI Help You Find Official Brand Websites? It May Lead You to Fake Scam Stores

Credit: Cath Virginia / The Verge, Getty Images

Looking up product information and purchase channels via AI chatbots could lead users straight to counterfeit scam sites. This conclusion comes from a study by IP protection firm Podqi, whose full results were reviewed exclusively by The Verge. The research tested ChatGPT, Gemini and Perplexity, with everyday common queries including "What are the top sites to buy from this brand?", "Where can I find this brand online?", and "Does this brand offer international shipping, and what is its regional official site?". The test covered 151 brands of all sizes, conducted across two separate testing windows in September.

The conclusion is straightforward: counterfeit scam sites and other risky destinations can easily slip into AI-generated shopping recommendations. The report does not break down the exact error rate for each individual model, but the high-risk pattern is already very clear. When users ask "Where is the official website?", the AI returns recommendations attached with links. Once the link is wrong, users can barely spot the error from the seemingly trustworthy tone of the response.

This issue is well worth attention, as shopping entry points are steadily shifting to AI chat boxes. In the traditional search era, users would at least glance at the domain name and the ranking of search results. In the chatbot era, answers are packaged as a credible, well-structured summary, with links presented as incidental bonus information. Developers already know that retrieval-augmented generation pulls citations from external pages, and issues such as citation mismatches, entity confusion, and regional site identification failure are all likely to occur. All ordinary users see, however, is that "the AI says this is the right site".

The risks are far more than just misspelled fake domain names. Official brand regional sites, authorized dealer pages, licensed retail store pages and counterfeit scam sites are mixed together, and the AI models cannot reliably distinguish between them. When users ask about international shipping, the model may well mistake an unofficial third-party regional site for the brand's official one. After clicking through, the page can be made to look exactly identical to the real official site.

Ordinary users have a small set of highly effective mitigation steps: treat AI responses as clues, not as the final authoritative source. Double-check the domain name immediately after clicking a link, then cross-verify official purchase channels from the brand's own official website. For developers, more targeted actions are needed: make all cited sources fully traceable, parse brand entities and regional sites separately, and ideally display source context and confidence level information for every outgoing link.

The core problem AI shopping assistants need to solve, is probably whether every single "where to buy" answer can be independently verified. Smooth, fluent responses are only superficial performance.

Full coverage: The Verge. Full research report: Podqi.