top of page

Brand Impersonation in AI Search: How to Find the Fake Websites AI Assistants Cite

2 days ago
5 min read

This guide was produced by AI, TELL ME!, a full-cycle AEO & GEO agency with our own AI search monitoring platform.

Summary

Two fake Russell & Bromley websites sat among the sources of a ChatGPT answer in June 2026, and brand impersonation in AI search now has a measured rate: in a September 2026 test of 151 brands by the brand protection company Podqi, 9% of shopping answers from ChatGPT, Gemini and Perplexity contained a risky website. More than half of those answers (57%) recommended the site as a place to buy, and fewer than one in ten warned the shopper. A Fudan University experiment shows how little it takes to get a planted page into an answer, with eight of ten AI search tools citing an invented concept within a week. A seven-step check then finds which domains an assistant names for a brand, and a short list of pages closes the gaps that impostors use.

Cover for a guide on brand impersonation in AI search with three figures: 9% of shopping answers carried a risky website, 57% of those recommended it as a place to buy and 8 of 10 AI search platforms cited an invented concept within a week. A mock AI answer beside them lists an official site, an authorised retailer and a fake domain.

Brand impersonation in AI search happens when an LLM (large language model) names a website that imitates a brand as its official site or as a place to buy. The shopper who follows the link lands on a store owned by someone else, and the brand sees none of it in a rankings report, since the answer sits in the buyer's private chat.

What brand impersonation in AI search looks like

A fake website reaches an answer as a cited source, a named official site or a recommended shop, and the assistant usually adds no warning. In June 2026 the UK scam-checking service Ask Silver found fake sites imitating Russell & Bromley and Dunelm among the sources ChatGPT gave for shopping questions, as Retail Gazette reported. A query about Russell & Bromley bags returned links to cloned stores on domains such as russell-and-bromley and russellbromleyonlineuk, with discounts of up to 80%.

Next bought Russell & Bromley in January 2026 after its administration, and the Guardian reported that the legitimate store sits inside the Next website, which leaves the name without a domain of its own. OpenAI said it had removed the fraudulent sites from its search index.

How common are fake websites in AI answers?

In a September 2026 test by Podqi, 9% of shopping answers from ChatGPT, Gemini and Perplexity contained a risky website. Marketing Tech News reported the study on 16 November 2026: 6,795 responses about 151 brands, five shopping questions per brand. The responses named 4,437 domains, and 718 of them were classed as risky, with 351 fake or suspected counterfeit and 367 inactive.

Podqi's risky-site rate by platform was 15% for Perplexity, 8% for ChatGPT and 4% for Gemini. In the answers that carried a risky site:

  • 57% recommended it as a place to buy;

  • 34% mentioned it without comment;

  • 9% warned the user.

Podqi's questions about regional websites drew the highest rate, at 16%, and 36% of the risky sites used country-specific endings or regional names.

Podqi sells brand protection and wrote the questions itself, and the test measured no clicks, purchases or losses, so the 9% describes how often risky sites appear and leaves the cost to shoppers unknown.

Why fake pages get into AI answers

An ordinary post on a site that assistants often cite can reach an answer within days, according to a Fudan University study. In the paper From Public Posts to AI-Search Citations, submitted to arXiv on 8 October 2026, 8 of 10 AI search platforms cited a fabricated concept within seven days of posting. A $14 purchase of GEO (generative engine optimisation) posting produced 13 public posts, and one AI search platform cited that content within an hour.

In the Fudan experiment one article on a heavily cited platform had more citation impact than over 20 matched posts on rarely cited ones. The experiment planted an invented concept, so it evidences the entry route and leaves the frequency of impostor domains untested.

How to check which domains AI assistants name for a brand

A brand finds its exposure by logging every domain that appears in assistant answers to buyer questions and sorting the domains by owner. Seven steps cover it:

  1. Write the questions that end in a click, such as where to buy, the official site per country and closing sales, using the prompt types from the guide to checking whether ChatGPT recommends a brand.

  2. Run each question at least three times on every assistant, since answers vary between runs.

  3. Log every domain an answer names or cites, whoever owns it.

  4. Sort the domains into the brand's own sites, authorised retailers, independent publishers and unknown ones.

  5. Look up each unknown domain for a recent registration date, spelling close to the brand name, a country ending with no official regional site behind it and prices far below the brand's own.

  6. Report confirmed fakes to the registrar, the host and the assistant's provider, and log the date.

  7. Re-run the same questions after each report and record whether the domain still appears.

Steps 5 and 6 belong with legal and brand protection teams, and the other five are monitoring work. A monitoring report that counts mentions and leaves out the domains misses the part that sends the buyer somewhere, and the guide to AI hallucinations about brands covers the matching check on what an answer says.

What a brand publishes so the official site is the obvious answer

A brand closes the gaps by publishing its official record in one place and getting independent pages to repeat it. Podqi advises publishing accessible lists of official regional sites and authorised retailers, and the table pairs each gap with a page to publish.

Gap

What the studies show

What the brand publishes

Regional sites

Podqi's questions about regional websites drew the highest risky-site rate, 16%

A page listing every official regional site and authorised retailer

Retired or merged brand

The legitimate Russell & Bromley store sits inside the Next website

A notice page or redirect on the old domain that says where to buy

Inactive domains

367 of Podqi's 718 risky domains were inactive

A register of every brand domain with renewal dates

None of the studies measured whether such a list changes what an assistant cites, so the re-run in step 7 is how a brand finds out.

Where an agency fits

AI, TELL ME! is a full-cycle GEO and AEO (answer engine optimisation) agency with 17+ years in marketing, SEO and PR, and its monitoring platform reports which sources the answers cite for a brand's prompts. A free AEO Monitoring scan shows what buyers see today, and GEO Readiness traces the pages behind it.

Delivery runs beside an in-house team through Tailored GEO Consulting and GEO Workshops or as a fully outsourced function, and Third Party GEO places the official domain on independent publications and directories, with PR for GEO as one extra.

FAQ

Can ChatGPT recommend fake websites?

ChatGPT has named fake websites in two published tests, with Ask Silver finding fake Russell & Bromley and Dunelm sites among its sources in June 2026 and Podqi finding a risky website in 8% of its shopping answers in September 2026.

How common are fake websites in AI shopping answers?

In Podqi's test of 6,795 answers about 151 brands, 9% contained a risky website, with rates of 15% for Perplexity, 8% for ChatGPT and 4% for Gemini. The researchers wrote the questions and measured no clicks or losses.

How can a brand find fake websites in AI answers?

A brand runs its buyer questions repeatedly on each assistant, logs every domain the answers name and checks the unknown ones for lookalike spelling, a recent registration and low prices.

Can a brand remove a fake website from an AI answer?

A brand cannot edit an answer and can report the site to the registrar, the host and the assistant's provider, then re-run the same questions to see whether the domain still appears.

Teams that want to see which sources AI assistants name for their brand today can start with a free AEO Monitoring scan from AI, TELL ME!.

bottom of page