AI search engines are sending some customers to fake and dead versions of real brands, and most of the time they present those sites as if they were the real thing. Two studies reported this month put numbers on it: Podqi found a risky website in about 9% of AI shopping answers about 151 brands (Podqi), and Scrunch found AI assistants citing impostor sites for 83 companies in the Fortune 500 (Scrunch).
So if you're the person who gets asked how your brand shows up in ChatGPT, Gemini, Perplexity or Google's AI answers, fake brand websites in AI search are now part of that question too. The good news is that checking for them takes an afternoon and a spreadsheet, and you don't need a new tool to start.
The short version
In Podqi's test, when an AI shopping answer included a fake or dead site, it recommended that site 57% of the time and warned the shopper 9% of the time.
Scrunch found impostors turn up most in refund, cancellation, login and account questions, which is where a customer is about to hand over money, a password or a phone call.
Impostor sites almost never rank on Google's first page (2.2% of the time in Scrunch's check) and almost never appear on public threat feeds, so your SEO reports and your security team's feeds probably won't catch them.
The fix starts with asking the engines the risky questions yourself, logging every domain they give you, and making your own help and official-store pages easy for the engines to match.
What the Podqi and Scrunch studies found
Podqi is a brand protection company, and it ran its study with data from Gauge, an AI search tracking company. They asked ChatGPT, Gemini and Perplexity five shopping questions about each of 151 brands, in two rounds in September, using the logged-out versions of each engine, and then reviewed every one of the 4,437 domains those answers named (Podqi). Of those, 718 were what Podqi calls risky: 351 looked like they were impersonating the brand or selling counterfeits, and 367 were dead domains.
Perplexity had the highest rate, with a risky site in 15% of its answers. ChatGPT came in at 8% and Gemini at 4%, and ChatGPT at least warned people in about a quarter (24%) of its risky answers. The worst question was the one about international shipping and regional websites, at 16%, and 36% of the risky sites in those answers used a country domain ending or a region name in the address. That's the kind of URL that looks perfectly plausible if you're a shopper in Germany or the UK and the assistant tells you it's the local store.
Scrunch looked at a much bigger set of answers. They sell AI visibility software (so they have a product in this fight, keep that in mind), and they went through answers they monitor across ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode and Microsoft Copilot from roughly May through October (Scrunch). About 1 in 1,250 answers with sources cited an impostor, which sounds tiny until you think about the volume; Scrunch estimates that works out to more than a million impostor-citing answers a year in the US alone. They also call the findings preliminary, and the impostor labels came from two AI reviews that had to agree with each other rather than from a person checking each site, so I'd treat the exact rates as rough.
How the engines present the fake sites
The numbers I'd put in front of a CMO are the ones about how the engines present these sites. In Scrunch's data, about 3 in 4 answers that cited a known impostor treated it as the brand's own source, and 0.3% warned the reader that the site might be fake. Podqi's split comes out in the same place from a different direction: of the shopping answers with a risky site, 57% recommended it, 34% mentioned it with no comment at all, and 9% warned against it (MarketingTech News has a good summary of both).
Fake brand sites in AI answers
How often AI sends shoppers to a fake or dead site
Then there's the question of which questions. Scrunch found refund and cancellation questions pulled in known impostors about 7.5 times as often as product comparison questions, and login and account questions about 6.3 times as often. Think about who asks those. It's a current customer, somebody who already bought from you and already trusts you, and they're about to type in an order number or a password or dial a support line. When Scrunch reviewed the impostor sites that were still online, about 6 in 10 had a login form, a support phone number, a discount storefront, gambling pages or crypto content on them.
This also tells you where to look first. In my own research on where AI engines go to decide about your brand, help center pages were 29% of the citations that went to brands' own websites, so the engines already lean on help content when the question is about you specifically. Refund, cancellation and login questions are help center questions, and that's the territory the impostors are fighting you for.
Where to look first
The questions where impostors show up
Tap a question to see what to type into each engine and what to log.
Why your usual reports won't catch it
There are three blind spots here, and my guess is most marketing teams have all three.
Number one is Google rankings. Scrunch ran each impostor-citing question through Google within a day and checked the top ten. The impostor made Google's first page 2.2% of the time, compared with 40.9% for the other sources the AI cited, and every known impostor ranked outside the top million websites by popularity, so a rank tracker isn't going to flag any of them.
Two is your security team's feeds. Fewer than 1% of the impostor domains AI cited showed up on any of three public threat feeds Scrunch checked (OpenPhish, Phishing Army and Phishing.Database). If your company relies on those kinds of lists to spot phishing, these sites are mostly invisible to it.
Then last is your analytics. A customer who goes from ChatGPT straight to a fake site never touches yours, so there's nothing in GA4 to see, and tracking ChatGPT traffic is already hard enough when the visit does land on your site.
As for why the engines cite these sites at all, Scrunch's explanation is that impostor pages often matched the wording of the question more closely than the legitimate pages cited next to them, sometimes because they'd copied the brand's own text. That fits with something I've been saying for a while, which is that the engines are assembling an answer from whatever pages match the question, and they don't agree with each other about the same brand. In the AI Verdict Study, 29% of brand and engine pairings gave different verdicts across runs of the identical conversation. This is a nastier version of the same problem, where nobody along the way checks whether the page belongs to you.
How to check for fake brand websites in AI search
I'd go at it in this order.
Number one, ask the risky questions yourself. Use Podqi's and Scrunch's question types as the list: "Where can I find [brand] online?", "Does [brand] ship internationally, and what are their regional websites?", "How do I cancel my [brand] order?", "How do I get a refund from [brand]?", "[brand] customer service phone number" and "[brand] login." Run them logged out in ChatGPT, Perplexity, Gemini, Copilot and Google AI Mode, and log every domain the answer names in a spreadsheet, along with whether the engine recommended it, mentioned it or warned against it. Podqi suggests doing it monthly. I'd run each question more than once, because two identical runs in my own data shared only about 38% of their cited websites on average (Where AI Engines Go).
Two, give the engines a better page to match. Podqi recommends one crawlable "official sites" page that lists every official regional store and authorized retailer, says plainly which countries you don't have a store in, and gets linked from your footer and help center. I'd add that your refund, cancellation and login help pages should answer the question in the same words a customer would use to ask it. That second part is my guess based on Scrunch's point about wording, not something either study tested, but I can't see a downside.
Three, look after your domains. Scrunch recommends keeping old support and campaign domains rather than letting them go, and Podqi recommends an inventory of the country-code domains you own and the ones you don't, plus a watch on variants like your brand name with "us-" or "-support" bolted on. Remember that 367 of Podqi's 718 risky domains were dead, and anyone can re-register a dead domain and put whatever they want on it.
Then last, get legal and security in the room. Hand them the spreadsheet, report the confirmed fakes to the hosting providers and registrars, and pursue takedowns. In a lot of companies marketing is the team watching the AI answers, so marketing may see this before anyone else does.
Self-check, 2 minutes
Would you catch a fake version of your brand?
Six questions. Answer for your brand as it stands today.
0 of 6 answered
Where this fits with the rest of AI visibility
Most of what I write about is the upside, like why visibility in AI answers is the number to watch now and how to get mentioned and cited more often. Fake sites are the downside of the same system, and I'd argue they belong in the same monthly report. If you're tracking how often ChatGPT recommends you, you can track which domains it hands people when they ask how to reach you, and it's mostly the same work.
If you'd like a starting point on the broader question of how the engines describe your brand, the free Quick Check is a reasonable first step, and the $199 Audit Kit has the fuller do-it-yourself version. My book, Explainable, goes into why the engines trust some sources over others, which turns out to matter a lot here.
If you run these questions for your own brand and find something, I'd love to hear about it, and I'd especially like to hear if any of the engines warned you properly. Find me on LinkedIn or drop me a note through the site.
Jarred Smith is the author of Explainable: Why AI Recommends Some Brands & Ignores Others, an Amazon bestseller on AEO, GEO, and SEO. He's a marketing leader with nearly 20 years of experience across healthcare, public media, retail, and environmental services. Find him at jarredsmith.com.