Earlier this year, The New York Times hired Oumi, an AI testing company, to answer a basic question: How often are Google's AI Overviews right? The AI overviews are the summaries that sit on top of the search results that billions of people see every day.

Oumi measured Google's Gemini 3 model as accurate 91% of the time across 4,326 searches, up from 85% with Gemini 2.  While that sounds reassuring, the article went on to point out that Google handles around five trillion searches a year, and 9% of inaccuracies translates to tens of millions of wrong answers every hour.

AI can make mistakes, but the bigger concern may be what happens when people trust those mistakes.

In this article, Kolmogorov Law examines new survey findings on how workers use AI, how often they verify its answers, and what happens when they don't.

The citations AI uses don't always hold up

In the New York Times article, Oumi also found that 56% of AI Overviews are "ungrounded," meaning the source the AI cites does not actually support what the AI portrays in its answer. A year earlier, that figure was 37%, showing that, while answer accuracy went up, the citations AI overviews use are becoming less accurate.

Studies show most people are inclined to trust AI

Researchers at the University of Pennsylvania's Wharton School conducted three experiments involving 1,372 participants to find out what people actually do when an AI hands them an answer.  When the AI produced a confident but incorrect answer, 73% of participants accepted it. The research also found that when participants used AI, it also increased their confidence in the incorrect answers.

The researchers called this phenomenon "cognitive surrender": the act of trusting AI's answer enough that you stop doing the thinking or checking yourself because the answer came back fast and read smoothly. This is different from “cognitive offloading,” which is handing a task to AI on purpose and then checking the output afterward.

While the University of Pennsylvania study had interesting findings, it was run under lab conditions, so it might not translate into real-world scenarios.  To analyze whether similar behavior occurs in workplaces, Kolmogorov Law commissioned a national survey, fielded through the Pollfish research platform, of 500 employed American adults ages 18 to 64 who use AI tools for work.

How 500 Workers Are Using AI at Work

Seventy-two percent of respondents said they use AI daily or several times a week at work. Nearly eight in ten use it for research, and about half use it to draft documents and email. Twenty-three percent said they have used it for legal, financial, or compliance questions.

But, when asked how often they independently verify an AI answer before acting on it or passing it along, only 35% of respondents said always. The remaining 65% said they verify AI answers less often: 33% said usually, 26% sometimes, and 6% rarely or never.

Of those respondents who don't always check AI answers, some said they used AI-generated answers even when they suspected the answers were wrong. Forty-two percent admitted they had gone ahead with an answer they suspected was wrong, either because it was quicker or because the AI sounded sure of itself. A quarter of respondents said they had done it more than once.

The findings also show how inaccurate AI answers can affect people's work. Thirty percent of respondents said a wrong AI answer has caused a problem in their work, such as a mistake in a deliverable, a bad decision, lost time or money or a client or legal issue. Four percent described the problem as serious, with another 24% saying they were unsure whether inaccurate AI data caused a problem.

The survey findings also show that verification isn't a cure-all. Among the 173 respondents who said they always verify AI answers, 40% also said they had accepted an answer they suspected was wrong. That group reported work-related problems at a similar rate to the overall sample, 32% compared with 30%. These findings echo what the Wharton team found in the lab as well.

Workers Using AI for High-Stakes Questions Report More Problems

The workers using AI for legal, financial, or compliance questions reported problems at a higher rate. Of the 113 respondents who said they use AI for those matters, 43% said a wrong answer had already cost them something, yet 64% of that same group still do not always verify AI answers.

Workers Already Know Who Pays

In a legal setting, responsibility for an inaccurate AI-generated answer can ultimately fall on the person who relies on it.

OpenAI, Google, and Anthropic all disclaim responsibility for the accuracy of their output in their terms of service. Yet 52% of surveyed workers did not know that the AI vendors' terms of service say this in writing.

Despite the lack of knowledge, workers largely understand that they may bear the consequences for relying on inaccurate data. When asked, “If you acted on a wrong AI answer at work and it caused financial harm, who do you think would be legally responsible?” Fifty-one percent of respondents said, "me personally."   Eighteen percent said their employer, 15% were unsure, and only 12% named the company that built the AI.

Most employers aren't creating policies to help. Only 22% of respondents said their employer has a written policy requiring AI output to be verified before it goes into work product. Fifty-nine percent said there is no such policy. Nineteen percent were not sure. Among client-serving professionals in law, finance, healthcare, consulting, and real estate, the figure was about one in three.

How Workers Can Protect Themselves

AI can be useful for a first draft, a way to organize a problem, or a place to start research. The risk comes when the AI output gets treated as finished work without verifying if it's accurate.

For the individual professional, that means using AI as you would any other tool: trust but verify.

Employers have a role, too. With 65% of workers saying they do not always verify AI answers and only 22% reporting a written policy requiring verification, clear workplace policies could help provide guidance and protect workers.

As AI becomes a routine part of work, knowing when to question an AI answer could mean the difference between getting ahead and facing the consequences of a costly mistake.

Methodology

This article is based on a survey of 500 employed U.S. adults, ages 18 to 64, conducted September 1, 2026, through the Pollfish online research platform. Respondents were screened to full-time, part-time, and self-employed workers. All 500 reported using AI tools for work at least occasionally; 72% use them daily or several times a week. The questionnaire consisted of one occupation question, one usage-frequency question, eight substantive questions, and one attention check, which all 500 respondents passed. Results are unweighted. The margin of error is approximately plus or minus 4.4 percentage points at the 95% confidence level for the full sample. Subgroup figures (client-serving professionals, n=79; users of AI for legal, financial, or compliance questions, n=113; respondents who say they always verify, n=173) carry wider margins. Public findings are drawn from the New York Times-commissioned Oumi analysis of Google AI Overviews and from University of Pennsylvania/Wharton research on AI verification behavior.