AI Chatbots Surfaced Fake Retail Sites in up to 15% of Shopping Queries
What happened
IP protection firm Podqi ran a structured study, shared exclusively with The Verge, testing whether three leading AI chatbots recommended harmful retail websites to shoppers. Described in AI chatbots are recommending fake retail websites, study finds, the research used crafted prompts across two testing periods in September 2026. It covered 151 brands including Lululemon, Birkenstock, and Le Creuset. Across 6,795 total responses, Podqi defined a risky site as one impersonating a legitimate retailer, selling knockoffs, or an expired domain reregistered by a third party. Perplexity showed risky sites in 15 percent of answers, ChatGPT in 8 percent, and Gemini in 4 percent. ChatGPT provided a warning about a risky site it surfaced in 24 percent of those cases, the highest warning rate among the three. Smaller brands were affected roughly twice as often as larger ones. In one documented example, Perplexity named an incorrect page as a brand's official website and described it as the best source for authentic items.
Why it matters
- ·Organizations embedding AI chatbots into shopping flows may direct users to fake or harmful sites without adequate output review. This creates consumer protection exposure and potential liability under frameworks such as FTC Enforcement on AI (Section 5 of the FTC Act).
- ·The study reveals significant performance variance across vendors, with risky-site rates ranging from 4 to 15 percent. Enterprises that have not evaluated their chatbot vendor's URL citation accuracy carry an unassessed risk in their vendor due diligence program.
- ·Smaller brands face disproportionate exposure because AI outputs tied to lesser-known names are harder for consumers to verify. This raises brand protection and trademark risk for companies whose names appear in AI-generated answers alongside impersonator sites.
Stay ahead of stories like this
Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.
Recent issues
- The next quarter will test whether compliance programs can keep pace with AI systems that breach containment, reshape capital markets, and enable fraud at a scale no prior control framework anticipated.12 Nov
- State-linked attackers and criminal operators are actively weaponizing AI tools as entry points into enterprise systems, turning internal platforms and open-source frameworks into live attack vectors.5 Nov
Free every Thursday. Unsubscribe anytime.
Governance controls affected
What to do now
- ☐Identify every consumer-facing product or workflow where your organization uses ChatGPT, Gemini, or Perplexity to surface website recommendations or shopping results, and confirm whether any URL validation step exists before those results reach users.
- ☐Ask your AI vendor or engineering team whether the chatbot's output review process specifically checks recommended URLs against known-harmful or impersonator site lists, and obtain written confirmation of that control.
- ☐Review your vendor contracts with OpenAI, Google, and Perplexity to determine whether they include any warranty or liability provision covering harmful third-party URL citations surfaced to your customers.
- ☐Instruct your brand protection or legal team to run sample queries for your brand name across all three chatbots and document any instances where impersonator or knockoff sites appear in results.
- ☐Add AI-generated URL accuracy to your AI output quality monitoring program, with particular attention to any prompts asking about regional websites or lesser-known brand variants, which the study found triggered risky results more often.
What to watch next
Podqi's study was timed ahead of the holiday shopping season, when consumer use of AI for product discovery is expected to increase. Compliance teams should monitor whether the FTC Enforcement on AI (Section 5 of the FTC Act) produces guidance or enforcement action specifically targeting AI-generated harmful citations in consumer-facing products. Vendor responses from Google and OpenAI indicate they are updating safeguards, so teams should request updated documentation of those controls as part of their next vendor review cycle. State consumer protection regulators, particularly in California and New York, may treat AI-surfaced impersonator sites as deceptive trade practices, warranting monitoring of enforcement signals in those jurisdictions.