AI engines can cite fake websites: what our analysis of 320,000 domains reveals

BotRank study, November 2026. In late October, Google AI Overviews cited veepeec.com as a source. One extra "c", Veepee's logo and colors. Then a form asking for your phone number and password. The real site is veepee.fr. This is not an isolated case: by analyzing the sources AI engines cite for the brands we track, we identified at least 144 fake websites. The phenomenon is rare. But it is real. And it mostly affects ChatGPT.

Key takeaways

• AI engines sometimes cite fake websites. Out of 320,232 domains cited by AI engines for the brands tracked by BotRank, we identified at least 144 fake sites: 75 confirmed, 69 probable.
• Three kinds of fakes. Perfect clones of brand websites, fake online stores and phishing pages, like this fake Veepee.
• ChatGPT is by far the most exposed. It produces 31% of the answers we analyzed but 91% of fake-site citations. Relative to volume, that is about 6 fake-site citations per 1,000 answers, versus less than 1 for Perplexity and about 0.1 for Gemini or AI Overviews.
• The phenomenon remains marginal. Less than 0.05% of cited domains and about 0.03% of citations. For any single brand, the share peaks at around 0.5%.
• Check before you accuse. More than a third of the domains that looked suspicious turned out to be official sites or legitimate namesakes.

A fake Veepee cited by AI

The domain veepeec.com looks almost exactly like the site of the French flash-sale brand. Only one letter sets it apart. The page shows Veepee's logo, colors and login interface. It asks for a phone number and a password: it is a phishing page, built to harvest customer credentials.

On October 30, 2026, Google AI Overviews listed it among its sources in an answer about online fashion. On November 9, the page was still online. Anyone clicking that source lands on an exact copy of Veepee's login page.

How we ran the analysis

Every day, BotRank asks the main AI engines (ChatGPT, Gemini, Google AI Overviews, Perplexity, Mistral…) the questions customers ask about the brands we track. For each answer, we record the cited sources.

1. The starting dataset

We analyzed 2.5+ million answers, 20+ million citations and 320,000 distinct domains.

2. Spotting suspicious domains

We isolated the domains that reuse a well-known brand name without being its official site. Six patterns keep coming back:

• the brand name with an unusual extension (.mom, .live, .pro, .shop…);
• the brand name glued to "com" (brandcom.com);
• an added word: "official", "shop", "store", "online", "global"…;
• an added country code: "fr", "us", "italia"…;
• a typo or an extra letter;
• an exotic extension on a name close to the brand.

3. Checking each domain, one by one

Every suspicious domain was checked individually:

• DNS resolution: does the domain still exist?
• A visit to the site, with a real browser for JavaScript pages.
• Comparison with the official site: content, links, email addresses, canonical URL.
• Domain registry records: registration date and registrar, compared with the official domain's.
• Security warnings: pages flagged as phishing or malware.

Each domain received a verdict, backed by evidence. A domain is only labeled a "confirmed fake" if we saw the copy, the fake store or the phishing page. It is labeled a "probable fake" when its content can no longer be observed (offline, blocked, hidden) but every clue points the same way: a hijacked brand name, a recent registration, the same registrar and date as proven fakes.

Three kinds of fake sites

Perfect clones

This is the largest group. The fake domain displays the brand's real website, copied live. Example: france-literie.live is an exact replica of the site of France Literie, a French bedding retailer. All 555 links pointing to france-literie.fr in the real site's code are automatically rewritten to france-literie.live. The same mechanism shows up on terdav.life (Terres d'Aventure), bergamotte.mom, loacker.mom and neuhauschocolates.pro.

These clones rarely come alone. On July 16, 2026, eight .mom domains were registered on the same day, with the same registrar, for eight brands that have nothing to do with each other. Brands do not work this way: it is the signature of a single operator.

Fake online stores

Other domains display a store in the colors of a well-known brand: coupons, "free worldwide shipping", fake showcase sites. barebellsus.com poses as a Barebells store, rxbarshop.com reuses RXBAR, godivaofficial.shop imitates Godiva. Four showcase sites named after KIND, Barebells, Fisherman's Friend and Quest Nutrition share the same template and the same servers.

Phishing

This is the rarest group, but the most dangerous. Besides the fake Veepee, tonyschocolonely.live is flagged by Cloudflare as a phishing site.

Why AI engines fall for it

One detail says it all. Maxi Zoo's real website includes copy protection: when it is displayed on any domain other than maxizoo.fr, a script sends the visitor back to the real site. So a human who lands on maxizoo.mom gets redirected.

But the bots that crawl the web to feed AI engines generally do not run this kind of script. They see the copy, with the brand's full content. And they can cite it as a source.

More broadly, a perfect clone is hard for a bot to tell apart from the original: same content, same text, same images. Only the domain name changes. Recent extensions (.mom, .live, .pro…) and freshly registered domains are not enough, on their own, to rule a source out.

ChatGPT, by far the most exposed

Not all AI engines are equally exposed to fake sites. ChatGPT produces 31% of the answers we analyze but accounts for 91% of fake-site citations. Gemini, which produces as many answers, accounts for only 2%.

BotRank study · Fake sites cited by AI

ChatGPT accounts for 91% of fake-site citations

Each engine's share of the answers analyzed and of fake-site citations (confirmed or probable).

Share of answers

Share of fake-site citations

Fake-site citations per 1,000 answers

ChatGPT

31%

91%

6.2

Perplexity

14%

6%

0.9

Gemini

30%

2%

0.12

AI Overviews

16%

1%

0.15

Mistral (Le Chat)

8%

0%

0

Source: BotRank.ai study, data as of November 9, 2026. 2.56 million answers analyzed, 5,366 fake-site citations. Copilot, Claude and AI Mode, under 1% of answers, are not broken out.

Our data cannot explain this gap with certainty. One hypothesis: the way each engine selects its sources. Gemini and AI Overviews rely on Google's index, which has long filtered out much duplicate content. ChatGPT relies on its own search tools, which seem to filter these copies less.

Rare, but real

Proportion matters. Fake sites remain the exception in AI answers.

But a single answer is enough to send a customer to a copy of a brand's website, or even to a page that steals their credentials. And when we checked, in early November 2026, about half of the fake sites we identified were still online.

One important caveat: 144 is a floor. We only checked domains flagged as suspicious using the patterns described above. Fake sites built differently will have slipped through.

Check before you accuse

A domain that looks like a brand is not necessarily a fake. More than a third of the suspicious domains turned out to be legitimate:

• Official sites. lindt.mom and lindt-eshopglobal.com do belong to Lindt: they are managed by the same brand-protection registrar as lindt.com.
• Namesakes. kinders.com has nothing to do with Kinder: it is the site of an American barbecue sauce brand, registered since 1998. Floracom is not a fake Flora: it is an Italian florist network.
• Generic sites. Local business directories, comparison sites or fan sites reuse a brand name without trying to pass for it.

Hence our rule: no domain is called a fake without observed evidence.

What brands can do

• Monitor domains that reuse their name. Exotic extensions, an extra letter, "official", "shop", a country code: these are the most common patterns.
• Look at which sources AI engines cite when they talk about them. A fake site does not show up in Google Analytics: you have to find it in the AI answers themselves.
• Report clones. To the domain's registrar, to the host and, for phishing, to browser reporting services.
• Protect their site against copying. A redirect like Maxi Zoo's protects human visitors, but not bots: it is not enough.

FAQ

Can AI engines cite fake websites?

Yes. Out of 320,232 domains cited by AI engines for the brands tracked by BotRank, we identified at least 144 fake sites: clones of brand websites, fake online stores and phishing pages.

Which AI engine cites the most fake sites?

ChatGPT, by far. It accounts for 91% of fake-site citations while producing only 31% of the answers analyzed. Relative to volume, that is about 6 fake-site citations per 1,000 answers, versus less than 1 for Perplexity and about 0.1 for Gemini and AI Overviews.

Is it common?

No. Fake sites account for less than 0.05% of cited domains and about 0.03% of citations. But the risk is real: a single answer can send a user to a cloned site or a phishing page.

How can I tell whether a site cited by an AI engine is fake?

Check the site's exact address before typing anything: an extra letter, an unusual extension or a word added to the brand name should raise a flag. When in doubt, go to the brand's official address rather than the suggested link.

How can a brand find out whether AI engines cite a fake of its site?

By tracking the sources AI engines cite on the questions that concern it. That is what BotRank measures, engine by engine.

Sources

• Data: BotRank.ai study, data as of November 9, 2026. Sources cited by ChatGPT, Gemini, Google AI Overviews, Perplexity, Mistral, Copilot, Claude and AI Mode for brands tracked with an active subscription.
• Domain checks: DNS resolution, site visits, RDAP registry records and Cloudflare warnings, on November 4 and 9, 2026.
• Reuse of the data: free, with the credit "BotRank.ai study, November 2026" and a link to this article.