
Quick Answer
According to Podqi, a study found that AI chatbots recommended risky shopping sites 6.3 times as often as they warned against them across 6,795 answers about where to buy products from 151 brands. The results show how AI shopping answers can surface risky sites and direct shoppers toward domains they did not authorize.
Podqi reviewed 4,437 domains and classified 718 as risky. Of those, 351 appeared to be fake and 367 were dead domains that anyone can re-register, according to Podqi’s September 2026 research.
In This Article
What Did the Podqi Study Find?
The study used five shopping questions across 151 brands and three AI engines: ChatGPT, Gemini and Perplexity. It produced 6,795 answers, with 2,265 answers analyzed for each engine.
Podqi found a risky site in 9% of all answers. Among answers that included a risky site, 57% recommended it as a place to shop, 34% mentioned it without comment and 9% warned the shopper away.
Perplexity displayed a risky site in 15% of its answers. ChatGPT displayed one in 8%, while Gemini displayed one in 4%, the lowest rate among the three engines.
The findings differed by type of response. Podqi reported that ChatGPT warned shoppers in 24% of its risky answers. Google AI Overviews warned in all seven risky answers it produced, and the overall risk rate fell from 12% in the first wave of fieldwork to 7% in the second.
These results give brand owners two visibility measures to monitor: whether a chatbot recommends an official retailer or another domain, and whether it attaches a warning to a domain the brand does not operate.
Which Questions Produced the Most Risk?
Regional website questions produced the highest rate of risky results. In 16% of answers to “What are their regional websites?”, the chatbot displayed a risky site.
Enterprise-brand results also showed limited warning coverage. Of 125 risky answers about enterprise brands, four warned the shopper and 121 did not, according to Podqi.
The source reporting on the study said smaller brands appeared more likely to be displaced by scam sites in AI responses. The full findings place particular emphasis on the regional-site question, including cases where an engine named a nonexistent regional page as an official website.
How Dead Domains Become a Storefront Risk
Dead domains made up 367 of the 718 risky domains in Podqi’s review. Because a lapsed domain can be re-registered, a domain repeatedly included in AI answers could pose a risk of later pointing to a counterfeit storefront.
Podqi treats a domain as risky when it appears to impersonate a brand, offers suspected counterfeits or is dead. This means a risky result is not limited to a working fake store. It can also be an inactive or lapsed domain that creates exposure when it appears repeatedly or is acquired later.
Podqi CEO and cofounder Trevor West said consumers should verify the source of a chatbot result before entering personal or credit card information, according to The Verge.
Malwarebytes Shopping Scam Survey and AI Shopping Assistant Risks
Malwarebytes’ 2026 shopping scam survey found that 77% of consumers were concerned about being scammed by a retailer using a fake AI website or using AI to misrepresent products or reviews. More than one in four consumers, or 28%, said they had experienced an AI-related shopping scam in the past year.
Malwarebytes reported differences by age group: 42% of Gen Z respondents said they had experienced an AI-related shopping scam, compared with 30% of Millennials, 22% of Gen X and 9% of Boomers and older.
Tom’s Guide reported that nearly one-quarter of US online shoppers planned to begin holiday shopping on an AI platform such as ChatGPT or Gemini, based on Bain Company research. It also reported that fake online stores were appearing at a rate of one every three minutes and had begun appearing in AI assistants’ product recommendations.
Price-range and gift-category searches could also reveal personal characteristics or financial information, including conclusions about a user’s disposable income, Tom’s Guide reported.
How Did the AI Companies Respond?
Google spokesperson Jessica Levis Boord said the company was happy to see Gemini perform best among the products tested. She also said Google displays links to sources beside its responses and is upfront that Gemini can make mistakes, according to The Verge.
OpenAI spokesperson Taya Christianson said the company continues to improve safeguards intended to make deceptive sites less likely to appear in shopping-related results. The company also asked researchers to share examples that can inform improvements.
What the Evidence Shows
Podqi’s September 2026 fieldwork covered 6,795 answers, 4,437 reviewed domains and 151 brands. It found risky sites in 9% of answers, with recommendations outnumbering warnings 6.3 to 1. The engine-level rates were 15% for Perplexity, 8% for ChatGPT and 4% for Gemini.
Some AI answers can lack accuracy, completeness or safety. Warnings may also be missing.
Key Considerations
- Podqi’s findings cover 151 brands, five shopping questions and three AI engines during two fieldwork waves in September 2026.
- A risky domain can appear fake, offer suspected counterfeits or be dead, such as the 367 dead domains among the 718 risky domains reviewed by Podqi.
- Regional website prompts produced a 16% risky-site rate, the highest among the five tested question types.
- Only four of 125 risky answers about enterprise brands included a warning.
- The risk rate fell from 12% in the first wave to 7% in the second.
What to Do Now
- Publish a clear official-sites page listing every authorized regional domain and storefront.
- Track how ChatGPT, Gemini and Perplexity answer your brand’s top shopping and regional-site questions.
- Record every domain each answer surfaces and flag domains you do not operate.
- Check whether each result recommends, mentions or warns against a domain.
- Review repeated dead domains and confirm that they cannot redirect shoppers to an unauthorized storefront.
For broader visibility tracking, BizScoreAI’s AI Scan tests how an AI agent reads a website and what it finds.
FAQ
Which AI Chatbots Recommend Risky Shopping Websites?
A Podqi study found that ChatGPT, Gemini and Perplexity all displayed risky sites. Perplexity displayed them in 15% of answers, ChatGPT in 8% and Gemini in 4%.
How Often Did the Chatbots Recommend a Risky Site?
Among answers that displayed a risky site, 57% recommended it as a place to shop, 34% mentioned it without comment and 9% warned the shopper away. Across the full set of 6,795 answers, risky sites were recommended 6.3 times as often as they were warned against.
Did ChatGPT Warn Users About the Risky Sites It Displayed?
ChatGPT warned shoppers in 24% of its risky answers, the highest rate among the three engines tested by Podqi.
Which Question Produced the Most Risky Results?
The question “What are their regional websites?” produced a risky site in 16% of answers, making it the riskiest of the five question types tested.
How Many Risky Domains Were Dead?
Of 718 risky domains reviewed by Podqi, 367 were dead. Podqi notes that anyone can re-register a lapsed domain, which could later support a counterfeit storefront.
SEOScanPro
SEOScanPro runs a full technical audit of a site and shows the measured result behind every check. Open the AI visibility report.
Sources
See our editorial disclaimer for how our articles are produced.
Run a free scan to see your AI Visibility Score, SEO rating, and local citation accuracy.
