Scam Phone Numbers in Google AI Overviews

Your card gets declined, a flight gets cancelled, a parcel goes missing - and you do the normal thing: you search "[company] customer service number." Google's AI Overview answers instantly, in bold, with a number and hours of operation. It looks authoritative because it looks like Google. You call, a friendly "agent" picks up, and the conversation feels completely normal.

The question almost nobody asks in that moment: where did that number actually come from?

Often it's correct. But a growing body of reporting - from the Washington Post, from cybersecurity researchers, and from consumers themselves - shows that search and AI-generated answers are increasingly returning phone numbers that belong to scammers, not the company you searched for. And people trust them precisely because an AI, not a random forum post, provided the answer.

Why an "authoritative" answer isn't automatically a real one#

Planting fake customer-service numbers online is an old scam. What's new is where those numbers now surface. They used to sit at the bottom of search results or on obviously sketchy sites. Now they sometimes land in the one bolded answer at the top of the page - the AI Overview - which most people read as Google's own verified answer rather than a summary of whatever the web says most often.

That's the core mechanic to understand: search engines and AI systems don't verify facts, they aggregate patterns. If a fake number appears next to a company's name often enough, across enough pages, the system starts treating repetition as evidence of truth.

How scammers get their number in front of the AI#

This runs on a few layered techniques, and once you see the mechanism the warning signs get obvious.

Classic SEO poisoning. Scammers build pages stuffed with the keywords people panic-search ("bank customer service," "airline help number") and use link schemes to push those pages up in rankings. They target urgent, time-pressured queries specifically - a lost card, a cancelled flight, a missing package - because urgency short-circuits verification.
Seeding numbers wherever AI systems read from. Fraudulent numbers get dropped into review sites, forums, Yelp business pages, YouTube video descriptions, and even PDFs uploaded to compromised university or government (.edu/.gov) websites. A link from a domain that's normally trustworthy lends the fake number borrowed credibility.
Content built specifically for AI, not humans. In December 2025, security firm Aurascape published research describing what it calls LLM phone-number poisoning - a real-world campaign where attackers manipulate public web content so AI systems recommend fraudulent airline support numbers as if they were official. The researchers frame this as a shift from classic SEO to GEO/AEO (Generative/Answer Engine Optimization): instead of competing to rank, attackers compete to become the single source an AI model selects, summarizes, and presents as "the answer." Aurascape found the poisoned content seeded across compromised .edu and .gov sites, WordPress blogs, YouTube descriptions, and Yelp reviews - and found it showing up in both Google's AI Overviews and Perplexity's Comet browser, not just one vendor's product.

Worth noting: even when a model's own answer is accurate, the sources it cites underneath can still be poisoned - for example a Yelp page loaded with bot-generated reviews planting a scam number. It's a systemic weak point in how these systems source information, not a single company's bug.

Why an AI answer is more dangerous than a normal search result#

A traditional search returned ten blue links, and people could at least compare a few sources before deciding. An AI Overview gives one confident, synthesized answer - no comparison, no visible disagreement between sources, and a strong implicit signal that this is settled.

The psychology compounds it. An AI-generated answer reads as neutral and "machine-checked," so it triggers less suspicion than an ad or a forum comment would. Combine that with genuine urgency - a blocked card, a flight about to leave - and critical thinking tends to switch off entirely.

Lily Ray, VP of SEO strategy and research at the marketing firm Amsive, put it plainly to the Washington Post: users would be safer if Google simply didn't show AI Overviews for business phone-number queries at all, since doing so hands scammers a new, higher-trust channel they're already exploiting.

Real cases, not hypotheticals#

The Royal Caribbean case. The Washington Post reported on a traveler who searched for Royal Caribbean's customer service number, couldn't find it in the company's app, and called the number Google's AI Overview surfaced instead. The "representative" was convincing - he knew shuttle pricing and pickup points in Venice, answered questions smoothly, even offered to waive a fee - and the caller handed over his card details. Suspicious charges appeared the next day. The same fraudulent number, reporters found, was also impersonating other cruise lines' support lines and turned up in both Google's and ChatGPT's answers.

A food-delivery case. A widely shared report described someone searching for an Indian food-delivery app's customer care number, getting one from an AI-generated answer, and being talked through screen-sharing and a payment "confirmation" by the fake agent on the other end - a classic remote-access scam wrapped around a number nobody had verified.

The pattern generalizes. Aurascape's research and subsequent reporting describe the same fraudulent-number pattern reaching into airline support lines and other travel and delivery services - not a one-off bug in a single search, but an active, ongoing campaign.

For scale: the FTC's 2024 data book recorded more than $12.5 billion in reported consumer fraud losses in the US, a 25% jump over the prior year, and imposter scams - someone pretending to be a company, bank, or government agency - were the single most commonly reported fraud category that year.

The other half of the problem: caller ID spoofing#

So far this is about calling the wrong number. The scam also runs in reverse: the scammer calls you, and your phone displays a real bank or government number.

Caller ID spoofing lets a caller display any number they choose - including one that matches your actual bank. STIR/SHAKEN, the industry framework meant to authenticate caller ID, verifies which carrier handled the call, not who the caller actually is; in one 2026 analysis of thousands of illegally spoofed calls impersonating major banks, retailers, and healthcare providers, more than half still received the framework's higher trust ratings. Reports through 2026 show bank-impersonation spoofing calls continuing to rise for exactly this reason.

So "the screen says my bank" is not proof of anything. Some scammers even run both halves of the play together: plant a fake number online first, then call from a spoofed version of the same or a related number, so the story looks consistent from every angle.

The one rule that defeats this: never trust a call you didn't initiate. If someone claiming to be your bank calls you, hang up and dial the number printed on your card or in the official app yourself.

Warning signs worth reacting to immediately#

Urgency and pressure. "Act now or you'll lose access / the booking / the refund." Panic is the scammer's actual product.
A request to screen-share or install a "support" app. Legitimate customer service never needs remote access to your device.
A request for a one-time code, PIN, or full card number read aloud. Real companies don't ask you to speak a login code.
Payment demanded in gift cards, crypto, or a transfer to a "temporary" account. No legitimate biller works this way.
The number doesn't independently link back to the company. Search the number alone; a real support line resolves back to the official site.
The company doesn't actually offer phone support. Some services are chat-only - a "hotline" for them is fabricated by definition.

How to protect yourself#

Get the number from the source, not the search answer. Go to the company's official site directly, check the back of your card, your statement, the product packaging, or the official app.
Type the address yourself. Don't click through - manually type bank.com or airline.com, and watch for lookalike domains with extra words or hyphens.
Skip the AI Overview and the ads when looking up a support number for anything sensitive - banking, travel, government services. Scroll to the company's own listing.
Cross-check the number on its own. Search the digits by themselves; a genuine number will tie back clearly to the official domain.
If an AI answer is your only option, check its cited sources. A good answer shows where it pulled from. If that's a forum, a Yelp page, or a random PDF rather than the company itself, don't trust it.
Never read a one-time code to anyone on a call, no matter who they claim to be.
Never trust an inbound call you didn't initiate. Hang up, then call back on a number you sourced independently.
Slow down. A few extra minutes of verification is cheaper than a drained account - urgency is the attack, not a side effect of it.
Turn on MFA everywhere it's offered. Even a compromised password shouldn't be enough on its own.
Save the numbers you'll actually need before you need them - bank, insurer, the handful of services you rely on - so you're never searching for one under pressure.

If you've already called a fake number#

Call your bank on its official number immediately and freeze the card.
Dispute the charge. Card payments are usually chargeback-eligible; speed matters.
Change passwords on anything you may have disclosed, and enable MFA if it isn't already on.
If you installed a "support" app or shared your screen, disconnect the device from the internet, remove the app, and scan for malware.
Report it - to your bank, and to your national fraud/consumer-protection body - so the number can be flagged and taken down before it catches someone else.

Summary#

AI-generated search answers can and do surface fraudulent customer-service numbers - verified cases include Royal Caribbean, a major food-delivery app, and an ongoing airline-targeting campaign documented by Aurascape.
This works because AI systems reward repetition across the web, not verified accuracy - and a single confident AI answer invites less scrutiny than a page of search results did.
Always source support numbers from the company's own site, app, card, or packaging - never from a search or AI answer alone, especially under time pressure.
Caller ID isn't proof either - spoofing lets scammers display your bank's real number, so never trust a call you didn't place yourself.
If it happens to you: freeze the card, dispute the charge, and report the number fast.

Want your team trained to catch this kind of scam before it becomes a call? Get in touch - bluwarden runs phishing simulations and awareness training that cover exactly this pattern, alongside the broader smishing and social-engineering playbook attackers are running in 2026.

This guide is general best-practice advice, not a substitute for a security assessment tailored to your situation. If you're actively dealing with fraud, contact your bank and local authorities directly.