When AI Shopping Tools Cite a Fake Version of Your Store

In early June, a UK scam-checking service asked ChatGPT for popular Russell & Bromley bags. The assistant did what it is built to do. It returned products, prices, and links to its sources. Some of those links went to fraudulent sites dressed up as the retailer, on domains like therussellbromleyofficial and russellbromleyonlineuk, a few of them advertising 80% off (The Guardian).

The setup that let it happen is the part worth your attention. Russell & Bromley went into administration in January 2026 and was absorbed by Next, so there is no official Russell & Bromley site anymore. Shoppers still search for the brand. Scammers built clones to meet them, and ChatGPT cited the clones as the answer. The same checker found fakes impersonating the furniture retailer Dunelm (Futurism).

For a small or mid-sized brand the threat is not a lost sale. It is an AI engine recommending an impostor under your name, confidently, to a buyer at the exact moment of purchase. OpenAI pulled those specific sites from its index once they were flagged. The hole they slipped through is still open.

Why a clone can score as well as you do

An AI shopping answer is not retrieved from a list of verified merchants. It is assembled in the moment from whatever sources the model judges relevant and trustworthy, then summarized with links. The model does not know your company. It reads signals: a domain that behaves like a store, product pages with structured data, reviews, consistent branding, mentions scattered across the web.

A clone copies every one of those. Same logo, same product shots, same descriptions lifted from your real pages, fake reviews seeded to match. Consumer researchers have a name for the deliberate version of this, data poisoning, where bad actors flood the web with synthetic reviews, forum posts, and manipulated indexing signals so a fraudulent site reads as credible across multiple sources at once. The model is not choosing the scam over you. It is choosing the best-anchored entity it can find, and when your real presence is weak, gone, or mid-migration, the clone is the strongest signal left standing.

That is the uncomfortable part. The fraud did not beat your security. It beat your entity.

The signals that say "this one is real"

Disambiguation is the whole game. You want every machine that encounters your brand to land on one unambiguous answer for which domain, which profiles, and which business are the genuine article. Four signals carry most of that weight.

Signal	What it establishes	Where it lives
Organization schema	This domain is the brand, with a defined name, logo, and identity	Structured data in your site's markup, on the homepage and key pages
sameAs links	These profiles are all the same entity, not separate or fake ones	Inside your Organization schema, pointing to your official profiles
Consistent NAP	One real business with one set of contact facts	Name, address, phone identical across site, Maps, and directories
Verified official profiles	Third parties confirm you, which a clone cannot easily fake	Google Business Profile, LinkedIn, Wikipedia, marketplace stores

The point of doing all four is that they reference each other. Your schema names your profiles, your profiles link back to your domain, your contact facts match everywhere a machine looks. A clone can copy your homepage in an afternoon. It cannot easily forge a web of verified, cross-referencing profiles that all agree on the same business. That web is what a model leans on when two domains look identical, and it is the one thing the impostor has to build from scratch.

A short audit to see if AI already confuses you

You cannot manage what you have not checked. Half an hour finds out where you stand.

1. Ask the assistants directly. Run "best [your product] from [your brand]" and "official [your brand] website" through ChatGPT, Gemini, Perplexity, and Google AI Overviews. Read which domains they cite, not just what they say.

2. Hunt the look-alikes. Search your brand name with the suffixes scammers reach for: official, -uk, -online, -store, -shop, hyphenated spellings. Note any domain that is not yours.

3. Test your own signals. Confirm your Organization schema validates, your sameAs links resolve to live profiles, and your NAP matches across every listing. A broken link here is a gap a clone fills.

4. Watch the discount tell. Search your brand name alongside phrases like "70% off" or "clearance." Extreme markdowns on a domain you do not own are the oldest signal of a fake, and now the AI reads them too.

If an assistant cites a domain you do not control, you have a live problem, not a hypothetical one.

When a clone is already being cited

Move on two fronts at once, because each is slow on its own.

Take it down. Report the fraudulent site to the assistant that cited it, the same path that got the Russell & Bromley fakes pulled from OpenAI's index. File abuse reports with the domain registrar and the hosting provider, submit the URL to Google Safe Browsing, and if the clone uses your marks, run a trademark takedown. None of these is instant, and none is guaranteed, which is why the second front matters more.

Re-anchor yourself. Give the model an obvious legitimate choice so it stops reaching for the impostor. Strengthen and verify your profiles, fix your schema and sameAs, and make sure your real domain is unmistakably the authoritative one. The deepest lesson from the Russell & Bromley case is that the clone won a vacuum. If you are closing a site, rebranding, or migrating, keep the old entity alive with redirects and updated profiles rather than letting it go dark. An empty space gets filled, and you do not get to pick by whom. This is a different problem from the entity fragmentation a rebrand causes on its own, where the pieces are all yours and just scattered. Here a hostile site is competing for the citation, so re-anchoring is defense, not cleanup.

The habit to build before agentic shopping scales

Right now most people use AI to get a recommendation and still click through to buy themselves. That gap is closing fast. ChatGPT lets people check out inside the chat, Amazon's assistant can auto-buy below a price you set, and Google is building a payments protocol for agents to purchase on your behalf. When the agent buys instead of the buyer, the cited link stops being a suggestion and becomes the transaction.

So treat your brand entity as always-on infrastructure, not a project you finish. Domains stay renewed, redirects stay in place, profiles stay verified, schema stays current, and someone owns checking the citations every quarter. The brands that get cloned successfully are almost always the ones that left a door open: a lapsed domain, a dead redirect, an unverified profile, a migration nobody finished. Close those, and the impostor has nothing to stand on.

Your citation is becoming your shelf. You would not let a counterfeit set up next to your storefront and wave customers in. The same vigilance now belongs to how AI describes you.

If you want to know which version of your brand the assistants are actually citing today, that is the first thing we check. Our free AI Visibility Audit runs your brand and three competitors across ChatGPT, Gemini, Perplexity, and Google AI Overviews, and flags any look-alike domain showing up where your name should be. Talk to us and we will tell you whether the engines have you straight.