BEEGITAL Insights · 13 November 2026 · Research and online safety
If an AI assistant tells you where to buy something, has it checked that the seller is legitimate? Not necessarily. A study published this week found that ChatGPT, Gemini and Perplexity sometimes showed people questionable shopping websites. Some impersonated real brands; others were inactive addresses. The answer is not to abandon AI shopping research, but to stop treating a chatbot’s link as proof of a trustworthy retailer.
Brand-protection company Podqi, using data from AI-search company Gauge, examined 6,795 answers to shopping questions about 151 brands. The research ran in two waves in September 2026 and was covered by The Verge on 12 November. About 9% of answers included at least one website the researchers classified as risky. That does not mean 9% of online purchases are scams, or that every risky link was recommended as safe.
What did the researchers actually test?
The study used five predefined shopping questions about where to buy each brand, which websites to choose, international shipping and local stores. The same questions were put to the three assistants on the same days; researchers then reviewed the domains cited in the answers. These were researcher-designed prompts, not a random sample of real customers’ conversations.
Of 4,437 different domains reviewed, 718 were classified as risky: 351 apparently fake or suspicious websites and 367 dead or inactive sites. An inactive website is not necessarily fraudulent; it was included because it cannot currently serve shoppers and a lapsed address could later be reused. The definition also covered lookalike shops and suspected counterfeit sellers.
| AI assistant | Visual comparison | Answers |
|---|---|---|
| Perplexity | 15% | |
| ChatGPT | 8% | |
| Gemini | 4% |
Source: Podqi / Gauge, September 2026, 2,265 answers per engine. Bar scale 0–20%. Includes mentions and warnings, not only recommendations.
These percentages measure exposure to risky links or mentions, not completed scams or proven losses. An answer was counted even when the assistant warned the reader about a questionable site. Comparing the numbers therefore requires more care than a simple “which AI is unsafe?” headline.
The most revealing number is not 9%
When a risky site appeared, the assistants recommended it as somewhere to buy in 57% of those answers. In 34% they mentioned the site without comment; in only 9% they warned shoppers. These are rounded figures within the risky-answer subset, not percentages of all 6,795 answers. ChatGPT warned about risky sites in 24% of its own risky answers—the highest warning rate of the three—while Gemini surfaced risky addresses least often. These are results from one specific test, not permanent safety rankings.
Why international shopping was the weak point
The riskiest question asked whether a brand ships internationally and what its regional websites are. A risky address appeared in 16% of answers to that question. Asking for the best places to buy a brand produced a risky address in 11% of answers. This matters when shopping across borders: a URL combining a brand name with a country name can look official, even if it is not linked from the real company’s website.

How an apparently plausible AI link can fool you
Imagine asking an AI assistant for the official local online store of a fashion brand. It returns a domain containing that brand’s name and the word “shop”. The site opens, shows product photos, a familiar logo and an HTTPS padlock. None of those details alone proves that the seller is authorised. This is a hypothetical example, not a transaction documented by Podqi.
HTTPS encrypts the connection; it does not confirm that a merchant is honest. A fake shop can also use a valid certificate, professional-looking graphics and an attractive checkout. The US Federal Trade Commission makes that distinction explicit in its online-shopping advice.
Six checks to make before paying at an AI-suggested shop
- Start from the brand’s real website. Find its main official domain independently, then use its own store locator to reach authorised regional shops or retailers.
- Inspect the full address. Look for odd country endings, spelling changes or extra words. A familiar logo can be copied; a link in a chatbot answer is not a certificate.
- Verify who is selling. Check the business name, usable contact details, shipping conditions and return policy. Search for independent reviews, while remembering that reviews can be fabricated.
- Question extraordinary discounts. An implausibly cheap branded product or high-pressure countdown deserves extra scrutiny, especially ahead of holiday shopping.
- Use payment protections. Prefer a payment method that offers a way to dispute a charge in your country. Avoid unfamiliar sellers demanding only gift cards, crypto or hard-to-reverse transfers.
- Keep evidence. Save the exact store URL, order confirmation, payment receipt and terms. If something goes wrong, contact your payment provider promptly.
These steps reflect the consumer-safety advice of the FTC and the European Commission. They reduce risk, but no checklist guarantees a perfect transaction.
What the findings mean if you own an online store
AI shopping answers can also affect legitimate businesses. If an assistant mistakenly labels a third-party shop as an official retailer, genuine customers may become confused or buy from someone else. The Podqi study did not quantify lost sales or customer losses, so those outcomes remain possible risks rather than measured results.
A practical starting point is to publish a clear list of official domains and authorised retailers, accessible from the website footer. Make the merchant’s identity, shipping rules, contacts and return information easy to find. Periodically ask more than one AI assistant where your brand’s products are sold, and manually check the domains it lists. This initial audit does not need an expensive new subscription.
Already clicked the link or entered payment details?
If you merely opened a suspicious page and provided nothing, close it and go to the retailer’s official site independently. If you entered a reused password, change it through the real service and enable two-factor authentication where available. If you paid or entered card information, contact your bank or payment provider immediately to discuss blocking the card or disputing the transaction. Ignore anyone who demands a second fee to “recover” the money.
Related BEEGITAL guides: how to spot a suspicious QR code and whether a small business really needs a website or online store.
Important limits of the study
Podqi sells brand-protection services and has a commercial interest in this subject. The research was not a government audit or a representative measure of every country, language or AI product version. Its prompts were written by the researchers and responses collected on specific dates in September 2026. The report does not show how many people clicked the links, paid or lost money. It cannot establish the risk of every future AI-assisted purchase.
Bottom line
AI assistants can help compare products and organise research. The final identity check must happen before you pay: find the official seller, verify the address, read the terms and choose a payment method with suitable protections. Convenience should not replace verification.
Sources and methodology
- Podqi — Original research, data and methodology (September 2026)
- The Verge — News report (12 November 2026)
- FTC — Online shopping safety
- European Commission — Protecting online shoppers
In-article photo: FlyD / Unsplash · Featured photo: rupixen / Unsplash. Illustrative images only.
