Ask a chatbot where to buy a brand's products and there's a decent chance it hands you a link to a store that isn't the brand's. That's the headline from a study published last week by Podqi, a brand-protection firm, working with AI search analytics company Gauge.
They asked ChatGPT, Gemini and Perplexity (all logged out) five shopping questions about 151 consumer brands, including Lululemon, Birkenstock and Le Creuset. Across 6,795 answers, 9% pointed shoppers to a risky address. Perplexity did worst at 15%, ChatGPT came in at 8% and Gemini at 4%.
The detail that should bother any ecommerce team: in 57% of those cases the assistant actively recommended the bad site. Risky sites were recommended 6.3 times for every one time they were flagged. In one example, Perplexity described a lookalike page as the official Stüssy store and a source of authentic products.
Black Friday is eleven days away. This is worth an afternoon of your time this week.
What "risky" actually meant
Podqi reviewed 4,437 linked domains by hand and flagged 718. Only about half (351) were fake or counterfeit stores. The other 367 were dead domains: closed or expired addresses that anyone could re-register.
That second number is the one founders can do something about. A lot of those links probably came from old campaign sites, retired regional stores and domains a brand stopped paying for. The model still remembers them. A scammer only has to buy one.
The riskiest questions were about international shipping and regional storefronts, at 16% exposure. When a shopper asks "does this brand have an Irish site?" and the honest answer is no, the model sometimes fills the gap with something plausible. Smaller brands reportedly had about twice the exposure of large ones, which makes sense. There's less authoritative text about them for a model to lean on.
A few caveats. The questions were written by Podqi, not taken from real shopper logs. The study counted appearances, not clicks or lost revenue. And Podqi sells takedown services, so it has a stake in the result. Still, the direction is hard to argue with, and exposure did drop from 12% to 7% between the two September rounds, so the platforms are clearly working on it.
What we'd do on a Shopify store this week
You can't control what a model says. You can make the correct answer very easy to find and the wrong ones very hard to monetise.
- Audit every domain you've ever owned. Old launch microsites, the .co.uk you set up in 2021, the seasonal campaign domain. Renew them and 301 them to your live store. A domain costs about $15 a year. Losing one to a fraudster costs a lot more.
- Answer the regional question explicitly. If you sell internationally through Shopify Markets on subfolders (
/en-ie,/de), say so on a plain shipping page. If you don't ship to a country, say that too. Gaps are where the model starts guessing. - Publish an "official stores" page. List your real domains, your marketplace storefronts and your authorised retailers. Link to it from the footer. It's boring, it's crawlable and it gives a model something definitive to quote.
- Tighten your Organization schema. Most Shopify themes ship some structured data, but check that your
urlandsameAsfields point to your real domain and social profiles, not a theme default or an old domain. - Run the five questions yourself. Where to buy, does it ship to X, is there a regional site. Ask each assistant once a week and log what comes back. It takes ten minutes and it's the only way you'll know.
- Set up lookalike domain monitoring. Registrar alerts or a brand-protection service will catch the obvious typo domains before a holiday campaign drives traffic to them.
Discovery moved. Brand protection has to follow it
For years the fake-store problem lived in paid search and social ads, and brands learned to watch those channels. AI assistants are a new front door, and most merchants aren't watching it yet. The fix isn't exotic. It's housekeeping: own your old domains, document where you really sell, and check what the assistants are telling people.
If your store architecture makes that hard, with scattered domains, half-migrated markets and stale structured data, that's a build problem worth solving properly. We're here to help founders and teams design and build digital products that are built to scale with you, not slow you down. If you're looking to build something, get in contact with us today!