AI Chatbots Push Fake Retail Sites in 15% of Shopping Queries

A new study shows Perplexity, ChatGPT, and Gemini are sending shoppers to counterfeit and scam websites just in time for the holidays.

Research · Source: GadgetReview

What happened

IP protection firm Podqi tested three major AI chatbots to see where they send shoppers. The firm ran 6,795 queries across 151 brands during two testing periods in September 2026. They found that Perplexity, ChatGPT, and Google Gemini frequently recommend fake or risky retail websites. The Verge first reported these findings just ahead of the holiday shopping season. Consumers rely on these tools for product discovery, but the bots are failing them.

Perplexity performed the worst by a wide margin. It surfaced risky sites in 15 percent of its answers. ChatGPT recommended bad links 8 percent of the time. Gemini had the lowest failure rate at 4 percent. Podqi defined a risky site clearly. It includes retailer impersonators, counterfeit sellers, and abandoned domains reregistered by third parties. These are sites designed to steal payment details or ship fake goods.

The chatbots act highly confident even when they are entirely wrong. In one documented case, Perplexity pointed users to an incorrect page for the brand Stussy. The bot explicitly described the fake page as the best source for authentic items. ChatGPT was the most likely of the three to flag a problem. However, it still only issued warnings in 24 percent of the instances where it surfaced a risky site. The safety guardrails are simply not working.

Key facts

Why it matters

Founders embedding AI chatbots into consumer shopping flows are taking on massive liability. If your application uses an LLM API to recommend products, you are now the one directing users to phishing sites. This creates direct consumer protection exposure. You face potential liability under frameworks like Section 5 of the FTC Act. You can no longer trust the base model to filter out fraudulent domains. Enterprise due diligence programs are currently ignoring this unassessed risk.

Smaller brands face a massive trademark and revenue crisis. The study found that smaller companies are exposed to fake AI links roughly twice as often as large brands. Consumers have a harder time verifying lesser-known brand URLs. This means startups and niche retailers will lose revenue to counterfeiters simply because AI models hallucinate their official domains. Queries about regional websites returned risky results even more often because the models invent regional sites that do not actually exist.

For builders

Build URL validation into your AI pipeline

Do not pass raw AI-generated links directly to your users. Engineering teams must implement an output review workflow that checks recommended URLs against known impersonator lists. If you skip this validation step, your company pays the price for user fraud.

Audit your brand presence in AI answers

Run sample queries for your own brand name across ChatGPT, Gemini, and Perplexity. Document any instances where knockoff sites appear in the results. Legal and brand protection teams must treat AI hallucinations as active trademark threats.

Demand vendor liability clauses for bad links

Review your API contracts with OpenAI, Google, and Perplexity immediately. Check if they include warranties or liability provisions covering harmful third-party URL citations. Right now the AI vendors are shifting the blame, meaning enterprise buyers carry the risk.

My take

We are building the future of commerce on models that cannot tell a real store from a cheap scam. AI companies want to replace search engines, but they refuse to take responsibility for the links they serve. Stop trusting the chatbot. Type the domain yourself and read every single letter before you hand over your credit card.

Original reporting: GadgetReview. This is my rewrite and opinion.

More AI news for builders