ChatGPT, Gemini, and Perplexity may direct users to fraudulent websites
What's the story
A recent study has revealed that artificial intelligence (AI) chatbots can sometimes lead users to potentially dangerous or fraudulent websites. The research, conducted by IP protection firm Podqi and shared with The Verge, tested three popular AI engines, ChatGPT, Gemini, and Perplexity, with a range of shopping-related queries about 151 brands. These included questions like "What are the top websites for buying [brand]?" and "Where can I find [brand] online?"
Risky links
All 3 AI engines often cited risky sites
The study found that all three AI engines often cited risky sites.
These were fake websites impersonating real retailers or selling counterfeit goods.
They also included old brand domains that have been re-registered by someone else, and inactive sites that could be purchased and reused by others.
6,795 responses were analyzed in total, with Gemini linking to a risky site in 4% of answers while ChatGPT did so in 8%. Perplexity performed the worst with a staggering 15%.
Brand vulnerability
Smaller brands more vulnerable to this issue
The study also found that smaller brands are more vulnerable to this issue.
Queries related to larger companies had risky sites in answers about half as often as those related to smaller ones.
This could pose a major problem for consumers, especially during the upcoming holiday shopping season when many will be using AI for shopping from unfamiliar brands or sites.
Corporate responses
AI companies respond to study's findings
In light of the study's findings, AI companies have responded.
Google spokesperson Jessica Levis Boord said they take information quality seriously and are happy with Gemini's performance in this test.
OpenAI spokesperson Taya Christianson said their goal is for ChatGPT to connect people with reliable shopping places, and they're working on improving safeguards against potentially deceptive sites appearing in shopping-related results.
Prompt influence
Certain prompts trigger risky sites more often
The study also found that certain prompts triggered risky sites more often than others. For instance, AI showed risky sites more when asked about regional brand websites compared to other categories.
The tested AI engines also sometimes listed fake regional sites for brands that don't have any.
In one case, Perplexity incorrectly identified the wrong page as Stussy's official website and recommended it as "the best starting point for guaranteed authentic items."