AI chatbots sent shoppers to fake or dead stores in up to 15% of answers. A sub's purchasing desk should stop using them to find where to buy parts
A brand-protection firm tested ChatGPT, Gemini and Perplexity on where-to-buy questions for 151 consumer brands and found risky sites in about 9% of answers, with smaller brands hit twice as often. Nobody tested construction supply, but the weak spots match how a foreman or purchasing agent hunts for a hard-to-find part.
AI chatbots pointed shoppers to fake, counterfeit or dead retail sites in about 9% of where-to-buy answers in a new test, and up to 15% for the worst performer. For a trade sub, the practical rule is short: an AI assistant can help identify a part, but it should not be how anyone in your shop decides which website gets the company card.
What did the study find?
Podqi, a brand-protection firm, asked ChatGPT, Gemini and Perplexity where to buy products from 151 brands during September 2026, including whether a brand ships internationally. The Verge first reported the results on November 12. The figures below are from Startup Fortune and Digit.
| Measure | Result |
|---|---|
| Answers collected | 6,795 |
| Domains reviewed | 4,437 |
| Domains flagged as risky | 718 (351 fake, 367 dead) |
| Answers containing a risky site, overall | About 9% |
| Perplexity | 15% |
| ChatGPT | 8% |
| Gemini | 4% |
| Risky answers where the chatbot actively suggested the site | 57% |
| Times ChatGPT flagged a risky site as unreliable | 24% |
Podqi counted a site as risky if it impersonates a retailer, sells counterfeits, is an abandoned domain someone else took over, or is a dead domain a fraudster could register. Risky sites were recommended 6.3 times as often as they were warned against. In one case, Perplexity named the wrong page as the official Stüssy site and described it as a source of authentic products.
Two patterns matter more than the headline numbers:
- Smaller brands had roughly double the exposure of large ones.
- Regional storefront questions were the worst category at 16%. The chatbots sometimes produced a plausible regional site for a brand that has none.
Does this apply to construction purchasing?
Not directly, and we will not pretend otherwise. Podqi tested apparel and housewares brands such as Lululemon, Birkenstock and Le Creuset. Nobody measured breakers, valve actuators or cordless tool batteries.
But look at where the chatbots failed: small brands, and "does this company sell here" questions. That is a fair description of construction parts hunting. A service tech needs a control board for a discontinued rooftop unit. A foreman needs a replacement part from a specialty manufacturer with a thin web presence. A purchasing agent is checking who stocks a specific fitting line. Those are small-brand, find-me-a-seller questions, and they are the kind a person now types into a chatbot instead of a search engine.
The cost of a bad answer is also higher on a job than in a closet. A shopper who lands on a fake store loses a card number. A sub that installs a counterfeit electrical or life-safety component has a product that does not match the approved submittal, with the warranty and liability questions that follow.
How much should you trust these numbers?
Some, with caveats:
- Podqi sells brand-protection services, so it benefits from the finding.
- The test covered one month and 151 brands. It is a snapshot.
- None of the three companies disputed the problem or offered competing data. Google said Gemini links to sources so users can verify. OpenAI said it is improving safeguards against fake or misleading results. Perplexity acknowledged its AI is not always fully accurate.
What should a sub change this week?
- Split identify from buy. Using a chatbot to work out a part number, a cross-reference or a spec is reasonable. Choosing the seller is a separate step that runs through the manufacturer's site.
- Check the authorized distributor list. Most manufacturers publish one. If the seller is not on it, call the manufacturer or your regular supply house before ordering.
- Read the domain. Podqi CEO Trevor West's advice is to verify where a chatbot's information comes from before entering card details, and to watch for misspelled domains and unusual extensions.
- Run new vendors through vendor setup. A site a chatbot suggested is a new vendor. It gets the same W-9 and approval steps as any other, even for a $200 part. The same discipline protects you from the payment fraud we covered in October.
- Tell the field. P-card holders and service techs are the ones most likely to do this from a phone in a mechanical room. One paragraph in the next toolbox talk or purchasing memo covers it.
Takeaway: The study does not prove construction supply is infested with fake sites. It shows the tools vouch for sellers they have not verified, most often for small brands. Keep the chatbot for figuring out what the part is, and keep your approved vendor list and the manufacturer's distributor page for deciding who you buy it from.
Construction AI Brief publishes three times a week. Subscribe at constructionaibrief.com and forward this to whoever holds your P-cards.
- Can ChatGPT, Gemini or Perplexity send you to a fake online store?
- Yes. Brand-protection firm Podqi collected 6,795 chatbot answers about 151 brands in September 2026 and found a risky site in about 9% of them: 15% for Perplexity, 8% for ChatGPT and 4% for Gemini. Risky meant a site that impersonates a retailer, sells counterfeits, or sits on an abandoned or dead domain.
- Is it safe to ask an AI chatbot where to buy construction parts or tools?
- Treat the answer as a lead, not a vendor. The Podqi study covered consumer brands, not construction supply, but it found smaller brands were exposed about twice as often as large ones, and many construction parts come from small manufacturers. Confirm any seller against the manufacturer's own distributor list before placing an order.
- Do AI chatbots warn you when a shopping site looks unreliable?
- Rarely, according to Podqi. Risky sites were recommended 6.3 times as often as they were warned against, and ChatGPT flagged a risky site as potentially unreliable in only 24% of the cases where it surfaced one.
- Which AI chatbot was safest for shopping questions in the Podqi study?
- Gemini had the lowest rate, with risky sites in 4% of answers, compared with 8% for ChatGPT and 15% for Perplexity. The test covered one month and 151 brands, and Podqi sells brand-protection services, so read the figures as a snapshot from an interested party.
- How should a contractor verify an online supplier an AI tool suggested?
- Go to the manufacturer's website directly, find its authorized distributor or where-to-buy page, and check that the seller is listed. Read the domain for misspellings and odd extensions, and route any new vendor through the same setup process your accounting team uses for every other supplier.