Tens of Millions of Errors Per Hour: Investigation Reveals the 'Accuracy Illusion' of Google AI Search

marsbitОпубліковано о 2026-04-13Востаннє оновлено о 2026-04-13

Анотація

A New York Times investigation, in collaboration with AI startup Oumi, reveals significant accuracy and reliability issues with Google's AI Overviews search feature. Testing over 4,300 queries showed the accuracy rate improved from 85% (powered by Gemini 2) to 91% (Gemini 3). However, given Google's scale of ~5 trillion annual searches, this 9% error rate translates to nearly 57 million incorrect answers generated hourly. A critical finding is the prevalence of "unsubstantiated citations." For correct answers, the rate of citations that do not support the AI's summary surged from 37% to 56% with the Gemini 3 upgrade, making it difficult for users to verify information. The AI heavily relies on low-quality sources, with Facebook and Reddit being among its top-cited websites. Furthermore, the system is highly manipulable. A BBC journalist successfully "poisoned" it by publishing a fabricated article; Google's AI began presenting the false information as fact within 24 hours. Google disputed the study's methodology, criticizing its use of the SimpleQA benchmark and an AI model (Oumi's own) to evaluate another AI. The company maintains its AI Overviews, combined with its search ranking systems, perform better than the underlying model alone. Critics note this defense does little to bolster user confidence in the feature's reliability.

Author: Claude, Deep Tide TechFlow

Deep Tide Guide: A recent test conducted by The New York Times in collaboration with AI startup Oumi shows that the accuracy rate of Google Search's AI Overviews feature is approximately 91%. However, given Google's scale of processing 5 trillion searches annually, this translates to tens of millions of incorrect answers generated every hour. More troublingly, even when the answers are correct, over half of the cited links fail to support their conclusions.

Google is disseminating misinformation on an unprecedented scale, and most people are completely unaware.

According to The New York Times, AI startup Oumi, commissioned by the publication, used the industry-standard test SimpleQA, developed by OpenAI, to evaluate the accuracy of Google's AI Overviews feature. The test covered 4,326 search queries, conducted in two rounds: one in October last year (powered by Gemini 2) and another in February this year (upgraded to Gemini 3). The results showed that Gemini 2's accuracy was about 85%, which improved to 91% with Gemini 3.

91% sounds good, but it's a different story when considering Google's massive scale. Google processes approximately 5 trillion search queries annually. With a 9% error rate, AI Overviews generates over 57 million inaccurate answers per hour, nearly 1 million per minute.

Correct Answers, Wrong Sources

More alarming than the accuracy rate is the issue of "unsubstantiated citations."

Oumi's data shows that in the Gemini 2 era, 37% of correct answers had the problem of "unsubstantiated citations," meaning the links attached to the AI summary did not support the information provided. After upgrading to Gemini 3, this proportion increased instead of decreasing, jumping to 56%. In other words, while the model gives correct answers, it is increasingly failing to "show its work."

Oumi CEO Manos Koukoumidis pointedly questioned: "Even if the answer is correct, how do you know it's correct? How do you verify it?"

The heavy reliance on low-quality sources by AI Overviews exacerbates this problem. Oumi found that Facebook and Reddit are the second and fourth most cited sources for AI Overviews, respectively. In inaccurate answers, Facebook was cited 7% of the time, higher than the 5% rate in accurate answers.

BBC Journalist's Fake Article "Poisons" Results Within 24 Hours

Another serious flaw of AI Overviews is its susceptibility to manipulation.

A BBC journalist tested the system with a deliberately fabricated false article. In less than 24 hours, Google's AI Overview presented the false information from the article as fact to users.

This means anyone who understands how the system works could potentially "poison" AI search results by publishing false content and boosting its traffic. Google spokesperson Ned Adriance responded by stating that the search AI feature is built on the same ranking and security mechanisms used to block spam, and claimed that "most examples in the test are unrealistic queries that people wouldn't actually search for."

Google's Rebuttal: The Test Itself Is Flawed

Google raised several concerns about Oumi's study. A Google spokesperson called the research "seriously flawed," citing reasons including: the SimpleQA benchmark itself contains inaccurate information; Oumi used its own AI model, HallOumi, to judge another AI's performance, potentially introducing additional errors; and the test content does not reflect real user search behavior.

Google's internal tests also showed that when Gemini 3 operates independently outside the Google Search framework, it produces false outputs at a rate as high as 28%. However, Google emphasized that AI Overviews, leveraging the search ranking system, performs better in accuracy than the model alone.

Nevertheless, as PCMag pointed out in a logical paradox: If your defense is that "the report pointing out our AI's inaccuracies itself uses potentially inaccurate AI," this likely does not enhance user confidence in your product's accuracy.

Пов'язані питання

QWhat was the accuracy rate of Google's AI Overviews feature as tested by Oumi, and how many errors does this translate to per hour given Google's search volume?

AThe accuracy rate of Google's AI Overviews was found to be 91% in the test. Given Google's annual volume of 5 trillion searches, this 9% error rate translates to over 57 million inaccurate answers generated every hour.

QAccording to the Oumi study, what was the trend in 'unsubstantiated citations' between the Gemini 2 and Gemini 3 versions of the AI Overviews?

AThe problem of 'unsubstantiated citations' (where the provided links did not support the AI's answer) increased from 37% with Gemini 2 to 56% with the upgraded Gemini 3.

QWhich low-quality websites were identified as major sources frequently cited by Google's AI Overviews?

AFacebook and Reddit were identified as the second and fourth most frequently cited sources by the AI Overviews feature.

QHow did a BBC journalist demonstrate the vulnerability of Google's AI Overviews to manipulation?

AA BBC journalist tested the system by publishing a deliberately fabricated article. Within 24 hours, Google's AI Overviews began presenting the false information from that article as a factual answer to user queries.

QWhat were Google's main criticisms of the Oumi study's methodology?

AGoogle criticized the study for having 'serious flaws,' stating that the SimpleQA benchmark itself contains inaccuracies, that using Oumi's own AI model to judge another AI could introduce errors, and that the test queries did not reflect real user search behavior.

Пов'язані матеріали

From Gold to Bitcoin: Fixed Supply + Institutional Frenzy, Might It Repeat the 'Explosive' Price Trend?

"From Gold to Bitcoin: Fixed Supply and Institutional Frenzy May Lead to 'Explosive' Price Rally Analysts suggest Bitcoin's price action could mirror gold's over the past two decades, following the launch of spot Bitcoin ETFs. Gold ETFs, introduced in 2004, drove gold's price surge to a current market cap near $28 trillion. Both gold and Bitcoin are non-yielding stores of value, with prices driven purely by investor sentiment rather than cash flows or credit. Gold ETFs experienced dramatic cycles: explosive growth, painful drawdowns, and slow recoveries, with each cycle reaching higher peaks. Bitcoin ETFs, approved in early 2024, saw rapid institutional adoption but are now facing similar volatility. Recent warnings highlight the risk of significant ETF outflows disrupting the current rebound. BlackRock's IBIT, a leading Bitcoin ETF, has sold nearly 100,000 BTC to meet redemptions while still holding over 733,000. The core parallel is fixed supply: when demand surges, prices explode, but demand is often volatile and wave-like, not steady. Institutional interest, through ETFs and corporate adoption, remains a key support pillar, helping to cushion sell-offs. If Bitcoin captures even a fraction of gold's role as a store of value, its upside potential is immense, though the path will be marked by high volatility. For investors, focusing on long-term trends and managing risk is crucial as this 'price explosion' narrative unfolds."

Foresight News7 хв тому

From Gold to Bitcoin: Fixed Supply + Institutional Frenzy, Might It Repeat the 'Explosive' Price Trend?

Foresight News7 хв тому

Why Is AI Agent Shopping Hard to Popularize?

The article argues that the popular narrative of "AI agent shopping" – equipping AI with a wallet to autonomously handle purchases – is fundamentally flawed and oversimplifies the complexity of shopping. It deconstructs shopping into two core actions: **information retrieval** (standardized, easily automated) and **value judgment** (deeply subjective and human-centric). The narrative mistakenly assumes AI can fully handle both. Value judgment itself has two layers: **evaluation** (assessing options against criteria) and **demand definition** (setting the criteria, weights, and values). The latter is inherently human and dynamic, as preferences are not fixed but constructed during the decision-making process ("constructive preferences"). The real dividing line for automation is not product standardization, but whether the **act of choosing** itself holds experiential value. For mundane purchases (e.g., printer paper), full AI delegation works. For experiential goods (e.g., wine, furniture), the joy of selection is core to consumption, so AI should act as an assistant that narrows options, leaving the final choice to humans. The "AI wallet" concept confuses three separate elements: decision-making, execution, and fund custody. Current payment industry solutions (e.g., from Stripe, Mastercard, Google, Visa) show that limited, scoped payment authorization tokens are sufficient for most consumer scenarios, not full fund custody. The true use case for autonomous AI wallets is in **B2B procurement** and **machine-to-machine (M2M) settlements** for standardized, high-frequency, low-value transactions. The real bottlenecks for AI shopping are not payment technology, but **1) the lack of trusted data sources** (e.g., fake reviews, counterfeit goods) and **2) the impossibility of automating human demand definition**. The conclusion is that the focus should be on safely automating the assessment and filtering process while reserving for humans the rights to define their criteria and enjoy the final act of choice. For experiential goods, the platform's competitive advantage shifts to providing a superior selection experience.

Foresight News1 год тому

Why Is AI Agent Shopping Hard to Popularize?

Foresight News1 год тому

After Nine Months of Shorting, a Full Turn to Long: Renowned Trader Opens Bitcoin Positions Around 64K, Crypto Market Long-Short Divergence Intensifies

After nine months of being short, prominent crypto trader Doctor Profit has closed all his bearish positions and started buying Bitcoin near $64,000, signaling a complete bullish reversal. He argues that structural market changes—such as impending U.S. regulation (CLARITY Act) and institutional adoption via securities tokenization—are rewriting the traditional four-year cycle script, potentially bringing the market bottom forward from the widely expected September/October timeframe. This view finds some technical support from on-chain analyst gumsays, who notes a bullish divergence on Bitcoin's weekly chart has persisted for 147 days, nearing the 161-day duration seen before the 2022 cycle low. However, cycle researcher Jake Pahor presents a counter-argument based on historical data. Analyzing patterns since 2014, he identifies three common features of past bear market bottoms: a ~12-month duration from peak to trough, a sustained period of extreme fear (with a proprietary risk score below 20), and the price falling below Bitcoin's realized price (~$53,000 currently). The current cycle, only nine months from its October 2025 peak, meets none of these conditions. The debate highlights a market torn between "front-running" a potential early bottom driven by new fundamentals and waiting for confirmation through traditional on-chain and sentiment metrics. While Doctor Profit opts for aggressive buying, Pahor maintains a disciplined, tiered accumulation strategy, continuing weekly buys at current risk levels but reserving larger orders for if more extreme fear emerges.

marsbit1 год тому

After Nine Months of Shorting, a Full Turn to Long: Renowned Trader Opens Bitcoin Positions Around 64K, Crypto Market Long-Short Divergence Intensifies

marsbit1 год тому

Senior Trader's Confession: How to Trade Market's False Expectations?

Veteran trader's case study: trading the market's "wrong expectations". This trade centered on a textbook "expectation error" after a weak CPI report. While the market initially priced in broad monetary easing (sending Nasdaq to 30,060), the crucial 30-year real yield hit a 20-year high. This signaled a fractured transmission mechanism: short-term rates eased, but long-term funding costs (vital for tech valuations) refused to fall. The trader executed five short positions on the Nasdaq (NQ) as it fell from 30,060 to 28,768. The core methodology: don't just trade the data, but analyze the market's implied causal chain and identify where it breaks. In this case, the chain was: Weak CPI → Policy Easing → Lower Long-Term Funding Costs → NQ Valuation Expansion. The break occurred between policy easing and long-term rates. The "veto variable" – long-term real yields – refused to confirm the bullish narrative. Trades were structured around "fast variables" (price) temporarily repairing while "slow variables" (funding conditions) remained broken. The article outlines a repeatable framework: 1) Map the market's implied causal chain. 2) Identify the veto variable. 3) Observe if it rejects the narrative. 4) Enter when price still follows the old script. 5) Choose the cleanest asset expression (e.g., short NQ, not broad S&P). 6) Define both invalidation and fulfillment exit conditions. The key insight: Alpha often comes not from an information edge, but from a "reaction function edge" – recognizing when the market is applying an outdated causal logic to new data. The critical question: What causal chain is the market's first reaction relying on, and is that chain still valid today?

marsbit1 год тому

Senior Trader's Confession: How to Trade Market's False Expectations?

marsbit1 год тому

Торгівля

Спот
活动图片