AIGenerated
EN · IT · ES

AI chatbots including ChatGPT and Gemini still point shoppers at scam sites

People who ask AI assistants where to buy things can be sent to fake stores that steal their money.

Business and rulesAI chatbots including ChatGPT and Gemini still point shoppers at scam sitesHerald of AI

How the test worked

Podqi conducted the study 1. They asked questions like “What are the top websites for buying [brand]?” and “Does [brand] ship internationally?” across about 151 different brands, ranging from small labels to big names like Lululemon and Birkenstock .

The goal was to see which chatbots would suggest shady links—sites that pretended to be real stores, revived dead ones, or sold knock-off products 1.

Risky Site Recommendations by Chatbot
Gemini4%ChatGPT8%Perplexity15%

percentage of answers that included a website known to be a scam or potentially harmful. Lower is better. Numbers: Podqi. Chart drawn by Herald of AI.

Which chatbot was worst?

Perplexity AI turned out to be the biggest offender, showing risky sites in 15 percent of its answers—that’s roughly one bad link out of every seven suggestions 1. OpenAI’s ChatGPT sat in the middle with 8 percent, while Google’s Gemini was least likely to show dubious storefronts, though still at a worrying 4 percent .

The chatbots didn’t always push the dangerous links outright. ChatGPT gave warnings about risky sites in 24 percent of its responses, the highest rate among the three 1. Still, across all three tools, 57 percent of the answers that included a risky site recommended it anyway, and 34 percent did so without any comment—only 9 percent actively warned the shopper away .

Who gets targeted most?

Smaller brands appeared to be the most vulnerable: when people asked about lesser-known labels, the AI was more likely to steer them toward scam sites rather than the official store 1. Certain types of questions also triggered riskier results—for example, asking about a brand’s “regional website” produced more problematic suggestions .

Why does this happen?

Conventional tools such as Google are susceptible to manipulation as well, yet AI assistants frame their output as definitive conclusions without showing their reasoning — a quality that leads users to accept their responses with greater unquestioning faith. Fooling these systems seems easy: a BBC journalist recently proved he could trick ChatGPT into telling other users he was a world-famous hot-dog champion—a lie he invented to make a point 1.

The problem isn’t limited to one startup. Fraud-detection outfit Ask Silver discovered that ChatGPT directs users toward replica copies of discontinued pages belonging to major brands — pages constructed specifically to pilfer financial account information. Overall, the internet is flooded with fake content—from news to product listings—and the AI models are still struggling to tell what’s trustworthy 1.

Podqi CEO Trevor West reminded users to double-check any link an AI gives before entering personal or payment information 1. Until the chatbots get better at spotting fakes, the safest habit remains: verify the source yourself.

This story reports on a trend for information only. It is not advice or encouragement to do any of it.

Key facts

  • Perplexity AI showed risky sites in 15% of answers, higher than ChatGPT (8%) and Gemini (4%).
  • 57% of answers that included a risky site still recommended it; only 9% warned against it.
  • Smaller brands were more likely to lead to scam recommendations.
  • Asking about a brand’s regional website increased the chance of a risky suggestion.

Sources

  1. AI Chatbots Won’t Stop Recommending Scam Websites to People Trying to Shop — futurism.com

If this story concerns you or your company, or uses something that belongs to you, ask for a correction or for it to be taken down. How requests work

Keep reading