Asked AI Overview for a Call Center Number, Got Connected to a Scammer

A while ago, a KOL I was working with got scammed. It started simply: they needed to reach an app's call center and searched Google for the number. At the very top was an AI Overview, the answer summary written by Gemini, complete with a phone number. They called it. The person who picked up wasn't the app's call center but a scammer, and they ended up getting scammed.

What made me curious wasn't just the scammer, but why that number showed up as the answer. My analysis at the time: the number had been spammed all over Facebook and social media forums, again and again, as if plenty of people were already using it. That much repetition made people trust it, and it turns out it made the AI "trust" it too.

After digging deeper, I found this isn't a one-off. Security researchers have documented the pattern, and the way it works is very close to what I saw.

A pattern researchers have already studied

In August 2025, The Washington Post reported on a businessman in the US who searched for a cruise company's customer service number to book a shuttle. The AI Overview showed a number, the "agent" on the line asked for his credit card details, and then charges he never approved appeared. According to Android Authority's summary of that report, a Google spokesperson said the company was aware of the issue, had taken action against certain scammer numbers, and was working on improving results.

In December 2025, the research team at Aurascape published the finding that, to me, best explains the KOL's case. They call it LLM phone-number poisoning. The attackers don't hack the AI. What they poison is the web:

• Fake numbers are planted in many places at once: PDFs on compromised government and university sites, WordPress blogs, YouTube video descriptions, even Yelp reviews.
• The content is written to be easy for machines to read, for example as Q&A or as lists like "[Airline] reservations number: ...".
• The same content is repeated across many domains so it looks credible.

The result: Perplexity answered a question about Emirates' official reservations number with a scam number, and Google's AI Overview also showed fake numbers for the same airline. According to Aurascape, the attackers' goal is to make their content the one piece the AI picks and presents as "the answer".

Why AI falls for it

AI search works by summarizing what is mentioned often and consistently across sources. When one number appears in dozens of places next to a brand name, to a machine it looks like a consensus. In reality, all of it was planted by the same party.

This is the part that gave me chills. I use the same principle when helping clients get named in AI answers: consistent information across the website, Google Maps and the media, as I covered in People Now Ask ChatGPT and Google AI, So Does Your Business Show Up in the Answer? and the Lond by Ciel Sowal salon case study. Scammers pull the same lever, just with fake data.

Two other things make it worse, in my view:

1. It looks convincing. The AI answer sits in a neat, confident box at the very top. Someone panicking about their account rarely double-checks.
2. Often there is no official number. This is my guess: many apps now only offer support through in-app chat, with no phone number. If there's no official number, the only numbers floating around online are the ones other people spread.

Google has made clear in its Google Search spam policies that attempting to manipulate generative AI responses in Google Search counts as spam. So Google does act on this, but enforcement always comes after someone has been hurt.

The scale in Indonesia

Data from Indonesia's Anti-Scam Centre (IASC) doesn't break out fake numbers in AI answers as a separate scheme, so the figures cover every kind of scam. Still, the picture is clear. According to OJK's press release of September 7, 2026 (in Indonesian), from November 22, 2024 to August 31, 2026 the IASC received 668,441 reports, recorded 1,276,688 reported and verified bank accounts, and found 159,472 phone numbers reported by victims.

Scam reports received by the IASC

668K

Phone numbers reported by victims

159K

Reported and verified bank accounts

1.27M

What made me think the most is the gap between the money that was blocked and the money that actually went back to victims:

Victims' funds handled by the IASC(billion rupiah, Nov 22, 2024 to Aug 31, 2026)

• Victims' funds blockedRp771.2 billion
• Funds returned to victimsRp206 billion

Once money has moved, not all of it comes back. So the cheapest fix is still prevention: making sure the number you call really is the official one.

How to check a call center number before you call

1. Get it from a source you open yourself. The Help menu inside the app, the official website whose address you type yourself, or the number on a card, package or receipt.
2. If you see a number in an AI Overview, open its source. If what it cites is a Facebook post, a forum, a comment section or an unclear blog, treat the number as unverified.
3. Match it against the official website. A number that only appears on forums and not on the official site is better left uncalled.
4. Know the warning signs. Real support agents don't ask for OTP codes, PINs or passwords, don't tell you to share your screen or install a remote access app, and don't ask for transfers to a personal account. Being pressured to hurry is a sign too.
5. If it's already happened, contact your bank or e-wallet through its official channel right away to block the transaction, then report it to the IASC. OJK stresses that reports to the IASC can only be made through its official website, iasc.ojk.go.id (in Indonesian), because fake sites impersonating the IASC exist too. OJK's official phone line is 157.

If you own a business or an app

Cases like this hurt two parties: the victim, and the brand whose name was used. A few things you can do:

• Put your official contacts clearly on one page. Also list the channels you don't use, for example "we never contact customers through personal WhatsApp accounts".
• Add structured data. Google's documentation for Organization structured data recommends the contactPoint and telephone properties to state how customers officially reach your business.
• Keep the details identical on every official channel: website, Google Business Profile, app and social media accounts.
• Check regularly what AI says. Ask Google, ChatGPT and Perplexity for "[your business] customer service number". Aurascape also recommends that brands monitor unauthorized numbers showing up in AI answers.
• Report what's wrong. Use the feedback button on the AI Overview, report spam content to Google, and warn your customers through your official channels.

Closing

AI Overview is handy, and I use it often myself. But for anything involving money and accounts, a phone number is the one thing I don't think you should take from an AI summary. Get it from the official app or website, even if it takes a few seconds longer.

If you own a business and want to make sure your official information is what shows up on Google and in AI answers, tell me what you need.