Researchers warn that scammers are poisoning AI answers with fake company contacts

Malicious actors are using fake web content to push scam phone numbers, email addresses, and login pages into answers from ChatGPT, Gemini, and Google AI Overview, according to security researchers.

Ariel Simon writes on Medium that his team found attacks affecting 374 companies, including airlines, banks, travel platforms, and software providers. The reported targets include Delta, Lufthansa, Qatar Airways, Bank of America, Chase, Airbnb, and TripAdvisor.

The campaign does not appear to exploit a flaw inside the AI systems themselves. Instead, it targets the information sources that AI tools retrieve from the public web. Attackers publish pages designed to look relevant and credible when someone asks for customer support, refunds, account help, or a company’s contact details.

How the manipulation works

According to the researchers, attackers distribute similar false claims across social media posts, forum entries, user-generated pages, PDF uploads, review sites, and website-hosting services. The pages often repeat fraudulent contact information, use question-and-answer formats, and include phrases such as “24/7 support” or “updated 2026.”

The content may also disguise phone numbers with symbols, spaces, emoji, or Unicode characters. This can make automated spam filters less effective while still allowing AI systems to interpret the number correctly.

Simon’s team says the results are inconsistent. A fake number may not appear every time an identical prompt is entered. But the likelihood rises when attackers publish more polished content across many websites.

The researchers argue that conventional takedowns cannot keep pace with automated posting campaigns. They also note that archived copies of removed pages can remain searchable.

Google reportedly classified the issue as outside its vulnerability reward program, while OpenAI closed the researchers’ report as not reproducible. The findings underline a practical risk for users: AI-generated contact details should be checked against a company’s official website before calling, emailing, or entering login credentials.