**AI-powered search tools are providing incorrect business phone numbers more than one-third of the time. According to a recent multi-model analysis, exposing users to scams, misinformation and mounting frustration.**

[The findings highlight](https://www.seerinteractive.com/insights/ai-models-provide-incorrect-phone-numbers-36-of-the-time-heres-what-you-can-do) a growing gap between traditional search safeguards and newer AI search experiences, where authoritative-looking answers may still be wrong.

As AI tools increasingly replace search engines for quick answers, the reliability of basic business information like customer service phone numbers is emerging as a critical and underappreciated risk.

## Why Phone Number Accuracy Matters More Than Ever

Spam and scam calls have conditioned users to rely heavily on technology for protection. 

Modern smartphones flag suspicious calls and traditional search engines have spent decades refining systems to verify contact information for legitimate businesses.

As a result, users have learned to trust phone numbers surfaced directly in search results. If a number appears in a prominent answer box, most people assume it is accurate. 

That trust is now being extended, perhaps prematurely to AI-generated search results.

Unlike traditional search engines, AI models are newer, less transparent and more vulnerable to manipulation, making incorrect contact information not just an inconvenience but a potential security threat.

## What Triggered the Concern About AI Phone Number Accuracy?

The issue gained urgency after an incident surfaced on Reddit showing ChatGPT providing a scam phone number for Chase Bank. 

![Example of AI wrong data](https://7982212.fs1.hubspotusercontent-na1.net/hub/7982212/hubfs/AI%20Models%20Incorrect%20Phone%20numbers-%20Chase%20Bank%20number%20scam%20ChatGPT.png?width=564&height=776&name=AI%20Models%20Incorrect%20Phone%20numbers-%20Chase%20Bank%20number%20scam%20ChatGPT.png)

This was not a fringe example. 

Clickstream data indicates that visitors to chase.com use ChatGPT 13% more often than Bing, suggesting that real customers are already relying on AI tools for banking-related information.

This raised a critical question: was the Chase incident a rare glitch or a sign of a systemic problem?

## Was This Just a One-Off Error?

The short answer is no. An initial internal test looked up contact information for a single brand across six AI models. 

Four models returned phone numbers but only one was accurate. 

That result prompted a much larger experiment to determine how widespread the issue really is.

A scaled analysis examined 178 branded phone-number queries across multiple industries, testing responses from seven AI search platforms including ChatGPT, Perplexity, Google AI Mode, Google AI Overviews, Claude, Meta and Gemini.

![AI search tools](https://7982212.fs1.hubspotusercontent-na1.net/hub/7982212/hubfs/sparktoro-chase-bank-audience.png?width=1200&height=458&name=sparktoro-chase-bank-audience.png)

## How Often Do AI Models Provide Phone Numbers?

AI systems are highly willing to answer phone-number queries. 

Across all prompts, AI models included a phone number 91% of the time when explicitly asked.

From a system perspective, this makes sense. The models are designed to be helpful and responsive, even when certainty is low. 

From a user perspective, however, this behavior creates risk because AI does not reliably know whether the information it is providing is correct.

When phone numbers were not included, models typically cited limitations, redirected users to official websites or stated that phone support was unavailable.

## Which AI Models Are Most Likely to Show Phone Numbers?

The likelihood of phone-number inclusion varied slightly by platform. 

Perplexity and Google AI Mode provided phone numbers 99% of the time, with ChatGPT close behind at 97%.

This consistency reinforces the illusion of reliability. 

When nearly every AI tool confidently returns a phone number, users naturally assume that accuracy is high.

Unfortunately, the data suggests otherwise.

## How Accurate Are the Phone Numbers AI Models Provide?

Accuracy was measured using three verification methods: alignment with Google Business Profiles, matching citations when sources were referenced and comparison against official customer service pages.

The results were uneven. Only 27% of AI-provided phone numbers matched Google Business Profiles, which often list corporate or headquarters numbers rather than customer service lines. 

Citation verification showed a much higher accuracy rate at 93%, indicating that when AI models explicitly cited a source, the number usually existed on that page.

Customer service page matching, arguably the most relevant benchmark revealed the core problem. 

Only 64% of phone numbers matched official customer service pages, meaning 36% were inaccurate.

## Which Industries Are Most Affected?

Industries where phone support is central to the customer experience were the most likely to have phone numbers surfaced. 

Telecom and airlines showed a 99% inclusion rate, while banking, one of the most sensitive categories, had phone numbers displayed 95% of the time.

This is particularly concerning because financial services were also among the most targeted by scams, increasing the potential harm when incorrect numbers are surfaced.

## Which AI Models Are Least Accurate?

Accuracy varied meaningfully by platform. Gemini performed best overall, with an accuracy rate of 89%, while ChatGPT ranked lowest at 68%.

One reason appears to be sourcing behavior. Gemini cited more sources per response and leaned heavily on structured third-party databases such as GetHuman. 

ChatGPT cited fewer sources and showed greater variance in output.

Both models relied on brand-owned sources only about 64% of the time, with the remaining information pulled from third-party and user-generated platforms.

## Why Are AI Models Getting Phone Numbers Wrong?

The core issue lies in sourcing.

Across the dataset, brand-owned sources were cited 41% of the time, while third-party sources accounted for 59%. 

Many of these third-party platforms such as GetHuman, PissedConsumer, BBB.org, and ZoomInfo are vulnerable to outdated data or manipulation.

While some third-party sites showed relatively high accuracy, the reliance on external sources increases the risk of incorrect or misleading information being repeated at scale.

## What Does This Look Like for Real Users?

A real-world example illustrates the confusion. 

Searching for IHG Hotels yields different phone numbers depending on the platform.

Google’s AI Overview shows a branded vanity number. AI models may return a different toll-free number sourced from a consumer complaint site. 

Visiting the official IHG website reveals multiple numbers segmented by country and purpose.

None of these numbers are necessarily fraudulent but inconsistency creates friction. 

Users may waste time calling the wrong line, lose confidence in the brand, or abandon their task altogether.

## Why This Is a Bigger Problem Than Scams Alone

While scams are the most alarming risk, everyday inaccuracies also carry consequences.

![Wil on AI phone number scam](https://7982212.fs1.hubspotusercontent-na1.net/hub/7982212/hubfs/wil-scam-number-test.png?width=825&height=1193&name=wil-scam-number-test.png) 

When customers cannot reach support easily, frustration builds. 

Over time, that frustration translates into negative brand perception, lower trust and lost revenue.

Inconsistent AI answers erode the assumption that surfaced information is authoritative, undermining confidence in AI search as a whole.

## What Can Businesses Do to Reduce Risk?

The most actionable area for brands is their own customer service infrastructure. 

While companies cannot control how AI models scrape third-party sites, they can ensure that official customer service pages are clear, consistent and easy to interpret.

Reducing ambiguity, such as multiple undocumented phone numbers or buried contact details helps AI systems latch onto the correct information more reliably.

## What Should Users Do Right Now?

For consumers, the safest approach remains old-fashioned: verify phone numbers directly on official websites, especially for financial, healthcare, or account-related matters.

Until AI systems demonstrate stronger verification mechanisms, blind trust in AI-provided contact information carries real risk.

## Final Takeaway

AI search tools are fast, convenient, and increasingly influential but this research shows they are not yet dependable for something as basic as a phone number.

With a 36% inaccuracy rate for customer service contacts, the issue is no longer theoretical. 

It affects real users, real businesses, and real outcomes.

Until AI platforms improve sourcing transparency and verification, accuracy not speed may prove to be the defining challenge of the next phase of AI-powered search.