A study conducted by the intellectual property protection company Podqi revealed that generative artificial intelligence tools, such as ChatGPT, Gemini, and Perplexity, can direct users to untrustworthy, inactive, or fraudulent online stores when they seek tips on virtual establishments.
Podqi conducted comprehensive tests involving 151 brands of different sizes (large, medium, and small) in the fashion, beauty, supplements, lifestyle, and furniture sectors. The tests were carried out in September 2026 and consisted of formulating five central questions related to these brands, using the ChatGPT, Gemini, and Perplexity services.
In total, the three tools generated 6,795 responses, citing 4,437 websites or virtual stores. Of the results obtained, 351 addresses were identified as fake, simulating official brands, selling potentially illegal products, or offering low-quality goods.
Additionally, another 367 recommendations led to sites that were suspended, displayed 'coming soon' notices, or simply failed to load. According to Podqi, such domains pose a risk as they can be acquired by third parties and converted into fraudulent commerce platforms.
AI Tool Performance
The results of the Podqi study indicated significant variations in the performance of the AIs. Perplexity showed the worst performance, with 15% of its suggestions pointing to sites that could put the user at risk. ChatGPT came in second place, registering 8% of responses directing to dangerous addresses.
In contrast, the OpenAI service demonstrated greater transparency, generating risk alerts in 24% of its suspicious responses. Gemini achieved third place, recommending risky sites in only 4% of the analyzed cases.
Company Reactions
Google (Gemini), OpenAI (ChatGPT), and Perplexity commented on the findings of the Podqi study, which was initially published by The Verge. Jessica Levis Boord, a Google representative, stated that the company takes information quality seriously and is pleased with Gemini's good performance among those tested. She added that Google constantly works to improve its models and maintains transparency about possible AI errors by providing source links for verification.
Taya Christianson, an OpenAI representative, declared that ChatGPT's goal is to connect people to safe places to shop and that the company continues to refine its safeguards to reduce the appearance of misleading websites in shopping results, thanking researchers for pointing out areas for improvement.
Finally, Jesse Dwyer, a Perplexity representative, highlighted that the platform has consistently been the most accurate AI for reliable reference searches for four years, emphasizing that while it never claims 100% accuracy, it is the only company focused exclusively on this purpose. More details are available in the AI Search Brand Risk Report 2026.


