Workers Spend 20 Days a Year Fixing AI Errors, Survey Finds

Workers dedicate significant time to correcting AI mistakes

Employees now spend an average of 90 minutes per day interacting with artificial intelligence tools at work, totaling approximately 47 days annually, according to a recent BambooHR survey of more than 1,600 full-time U.S. workers. But the productivity picture is more complicated than it appears: nearly half of that time—roughly 20 days per year—goes toward troubleshooting AI errors and iterating on prompts to get usable results.

The phenomenon of low-quality AI output has become common enough to earn its own term: "workslop." While 65% of employees report feeling confident and enthusiastic about using AI at work, and 58% cite time savings as a motivating factor, the reality involves substantial overhead in quality control and correction.

Executives are the heaviest AI users, with vice presidents and C-suite leaders spending nearly twice as much time with AI tools compared to individual contributors, the survey found.

AI replacing human interaction in knowledge transfer

The rise of AI usage is reshaping workplace dynamics in ways that extend beyond simple task automation. Survey respondents reported spending 41% of their time interacting with AI versus 59% with people—a split that raises concerns about declining human collaboration.

Mentorship appears particularly vulnerable. More than one-third of workers said knowledge transfers within their organizations now happen primarily through AI tools rather than human colleagues. BambooHR noted that 27% of respondents would rather ask AI than admit to a coworker they need help, citing "a mix of self-sufficiency and self-consciousness."

This trend aligns with earlier research from Workday showing that 33% of employees rarely or never have non-task-related conversations with colleagues during a given week, while 76% have turned to AI for advice.

Why it matters

These findings reveal a critical gap between AI adoption and effective deployment. Organizations are investing heavily in AI tools while simultaneously creating new categories of work—error correction, prompt engineering, quality assurance—that weren't factored into initial productivity calculations. The shift away from human knowledge transfer could have long-term implications for organizational learning, employee development, and workplace culture that many companies haven't yet addressed in their AI strategies.

The tolerance problem

Not all workers are committed to fixing AI mistakes. Research from Zety published in February found that only 39% of workers consider workslop completely unacceptable and correct it, while 31% find it unacceptable but tolerate it anyway. Another 21% deem it somewhat acceptable if deadlines are met. Zety estimated workers spend up to six hours or more weekly correcting AI output, closely matching BambooHR's findings.

Talent management firm Talogy noted in an August report that AI has advanced faster than organizations' ability to assess its optimal use cases—a lag that shows up in daily work patterns.

In HR specifically, 80% of professionals surveyed by Paylocity said they were actively managing at least one issue with their AI tools, even as 43% reported AI saved their recruiting teams the equivalent of one full working day each week.

The details were first reported by HR Dive, based on BambooHR's survey data.