By: Andrew Carlson

In an age increasingly defined by the seamless integration of artificial intelligence into everyday life, the promise of instantaneous, authoritative answers has captivated billions of users worldwide. Yet beneath this veneer of technological sophistication lies a growing and deeply consequential problem: the erosion of accuracy at an unprecedented scale. A recent analysis has cast a harsh spotlight on Google’s AI-generated search feature, revealing a troubling pattern of misinformation that is reverberating across the digital ecosystem.

According to a report on Thursday by The New York Post, Google’s AI Overviews—prominently displayed summaries generated at the top of search results—are producing vast quantities of erroneous information, raising fundamental questions about the reliability of automated knowledge systems and their impact on both users and the news industry.

The findings originate from an analysis conducted by the startup Oumi, which sought to evaluate the accuracy of Google’s AI models, specifically Gemini 2 and its more advanced successor, Gemini 3. By reviewing thousands of search results generated by each model, researchers determined that the systems were accurate approximately 85 percent and 91 percent of the time, respectively.

While these figures may initially appear reassuring, their implications are far more unsettling when considered in context. As The New York Post report emphasized, Google is expected to process more than five trillion searches in 2026 alone. Even a relatively modest error rate translates into an astronomical volume of misinformation—hundreds of thousands of incorrect answers every minute, accumulating into tens of millions of inaccuracies each hour.

This phenomenon represents a profound shift in the nature of information dissemination. Errors that might once have been isolated or obscure are now amplified and delivered with the authoritative tone of algorithmic certainty, often without clear indication of their fallibility.

One of the most insidious aspects of AI-generated search summaries is their presentation. Positioned at the top of search results, AI Overviews effectively supplant the traditional list of links that once guided users toward primary sources. This design choice, as noted in The New York Post report, imbues the summaries with a sense of authority that can obscure their inherent limitations.

Unlike conventional search results, which invite users to evaluate multiple sources, AI Overviews consolidate information into a single, ostensibly definitive response. In doing so, they reduce the likelihood that users will seek corroboration, thereby increasing the potential impact of any inaccuracies.

The consequences of this shift are not merely theoretical. Oumi’s analysis identified a range of errors, from misreported historical dates to incorrect claims about well-documented events. Among the examples cited were inaccuracies regarding the transformation of musician Bob Marley’s residence into a museum and the omission of cellist Yo-Yo Ma’s induction into the Classical Music Hall of Fame.

While such errors may appear trivial in isolation, their cumulative effect is far from benign. Each incorrect answer contributes to a broader erosion of trust in information systems, a development with profound implications for public discourse.

The ramifications extend beyond individual users to encompass the entire ecosystem of news production. As The New York Post reported, publishers have expressed mounting concern that AI Overviews are diverting traffic away from their websites, thereby undermining the economic foundations of journalism.

Danielle Coffey, president and chief executive of the News and Media Alliance, articulated this concern with stark clarity. The organization, which represents more than two thousand news outlets, argues that Google’s approach effectively appropriates journalistic content without adequate compensation or attribution.

“Algorithmically generated responses that pull in data from nearly every source on the internet simply cannot be trusted,” Coffey stated, in remarks highlighted by The New York Post. She further emphasized the disparity between the rigorous fact-checking processes employed by publishers and the comparatively unregulated nature of AI-generated content.

This dynamic has created a paradoxical situation in which the very institutions responsible for producing reliable information are being marginalized by the technologies that depend on their output.

Compounding the issue of accuracy is the growing prevalence of what researchers describe as “ungrounded” responses—instances in which the sources cited by the AI do not substantiate the information presented. According to Oumi’s findings, the proportion of such responses increased significantly between the two models, rising from 37 percent in Gemini 2 to 51 percent in Gemini 3.

This trend, as noted by The New York Post, suggests that improvements in overall accuracy may be accompanied by a decline in the reliability of source attribution. In practical terms, this means that even when AI-generated answers appear credible, they may lack a verifiable foundation.

The implications are particularly concerning in an era characterized by widespread misinformation and declining public trust in institutions. Without transparent and accurate sourcing, users are left to navigate a landscape in which the distinction between fact and fabrication becomes increasingly blurred.

Perhaps even more alarming is the apparent ease with which AI Overviews can be manipulated. The analysis cited by The New York Post includes an example in which a deliberately fabricated claim—asserting that a journalist was renowned for competitive eating—was incorporated into Google’s summaries within a matter of hours.

This vulnerability highlights a fundamental weakness in the system’s reliance on publicly available information, including sources that may be incomplete, biased, or deliberately misleading. By treating such material as credible, AI Overviews risk amplifying falsehoods and legitimizing them in the eyes of unsuspecting users.

In response to the criticism, Google has defended its technology, questioning the methodology of the Oumi study and emphasizing the overall accuracy of its models. A company spokesperson argued that the analysis relied on flawed benchmarks and did not reflect real-world usage patterns.

As reported by The New York Post, Google maintains that its AI systems meet the same high standards applied to its broader search infrastructure. The company also pointed to the inherent challenges of evaluating AI performance, particularly when using datasets that may themselves contain inaccuracies.

Yet such assurances have done little to quell concerns among publishers and industry observers, many of whom view the issue as symptomatic of a broader tension between technological innovation and accountability.

The current controversy is not an isolated incident. Since their introduction in 2024, AI Overviews have been plagued by a series of high-profile errors, including bizarre recommendations such as adding glue to pizza sauce and promoting the supposed benefits of tobacco for children.

These episodes, widely reported by The New York Post, have underscored the unpredictability of AI-generated content and raised questions about the safeguards in place to prevent such anomalies.

The stakes of this debate extend far beyond the immediate question of accuracy. At its core, the issue touches on the future of information itself—how it is generated, disseminated, and consumed in an increasingly digital world.

For users, the challenge lies in navigating a landscape in which convenience may come at the cost of reliability. For publishers, the imperative is to adapt to a changing environment while advocating for fair compensation and recognition. And for technology companies, the task is to balance innovation with responsibility, ensuring that the tools they create serve the public good.

The revelations surrounding Google’s AI Overviews, as chronicled by The New York Post, represent a pivotal moment in the evolution of digital information. They expose the limitations of even the most advanced technologies and highlight the enduring importance of human judgment and editorial oversight.

As the volume of AI-generated content continues to grow, so too will the need for vigilance, transparency, and accountability. The promise of artificial intelligence is immense, but it must be tempered by a recognition of its imperfections.

In the final analysis, the question is not whether AI will shape the future of search—it undoubtedly will—but whether that future will be grounded in truth or obscured by error. The answer will depend on the choices made today, by those who build these systems and those who rely upon them.