--- title: "69 Percent of AI Users Admit Shipping Work They Cannot…" canonical: "https://www.metaintro.com/blog/workers-shipping-unverified-ai-work-index" language: "en" author: "drashtigarach" published: "2026-09-07T10:28:43.000Z" modified: "2026-09-07T10:28:43.000Z" --- [Back to Blog](/blog) [AI](/blog/tag/ai)[Skills](/blog/tag/skills)[Career Tips](/blog/tag/career-tips) # 69 Percent of AI Users Admit Shipping Work They Cannot Stand Behind Glean's Work AI Index 2026 finds 69 percent of AI users ship work they have not verified, while only 13 percent say their company performs better. [![Drashti Garach](https://cdn.metaintro.com/rs:fill:40:40/q:72/plain/images/5719d740-e510-42bc-8017-e040d145f35f_1766029465094.png)Drashti Garach @DrashtiGarach](/blog/author/drashtigarach) [September 7, 2026](/blog/archive/2026/09)13 min read ![A wall of internal mail pigeonholes holding blank documents](https://cdn.metaintro.com/rs:fill:1200:675/q:78/plain/images/10169-pigeonhole-wall.png) [https://x.com/intent/tweet?text=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index](https://x.com/intent/tweet?text=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index&title=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index&title=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind)[mailto:?subject=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index](mailto:?subject=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index) The most uncomfortable finding in [Glean's Work AI Index 2026](https://www.glean.com/work-ai-institute/work-ai-index) is not about jobs disappearing. It is that 69 percent of AI users admit to shipping work they have not verified, do not fully understand, or cannot confidently stand behind. The same research puts adoption at 87 percent of digital workers, who report saving 11 hours a week, while only 13 percent say their organisation is performing significantly better because of it. At [Metaintro](https://www.metaintro.com), we read findings like this as career information rather than technology commentary, because the person whose name is on unverified work is the one who carries it. Here is what the numbers say and what to do about them. ## What Exactly Did the Research Find? Four figures that only make sense together. Adoption is close to universal among digital workers at 87 percent. The perceived time saving is large at 11 hours a week. The organisational payoff is small, with 13 percent saying their company performs significantly better. And the quality admission is enormous, with 69 percent of users acknowledging they have shipped work they cannot fully vouch for. Read in sequence, those describe a specific failure mode rather than a general anxiety about AI. People are using the tools, the tools are genuinely saving them time, and a large share of that saved time is being converted into output that nobody has checked. The gap between individual speed and organisational performance is exactly what you would expect if a meaningful proportion of the faster output needs rework later. It is worth being precise about what the 69 percent figure is and is not. It is a self-report, which means it understates rather than overstates, since people are generally reluctant to admit to shipping work they cannot defend. It also covers three different things at once: work that was not checked, work that was not understood, and work the person would not stand behind. Those are separate problems with separate fixes, and the survey groups them. ## Why Does Individual Time Saved Not Become Company Performance? Because the time is being saved at one step and spent again at another. If a task that took four hours now takes one, but the output then needs review, correction or rework by someone else, the organisation has not saved three hours. It has moved them. That is the most straightforward explanation for a picture where 87 percent of workers feel faster and 13 percent of organisations feel better off. This matches what other research has found from a different direction. [Workers lose nearly half their AI time to cleaning up its mistakes](https://www.metaintro.com/blog/bamboohr-workers-half-ai-time-fixing-output), which is the same arithmetic seen from the correction side, and the [AI productivity paradox in which most companies see no gains](https://www.metaintro.com/blog/ai-productivity-paradox-2026-companies-no-gains-economists-breakthrough) is the macro version of it. The pattern is consistent enough across sources that it is now the default expectation rather than a surprise. There is a second mechanism worth naming, because it affects your workload directly. When output speeds up but review capacity does not, the bottleneck moves to whoever checks the work. In most teams that is a senior person or a manager, which means AI adoption at the individual level can quietly increase the load on the people above you while making your own throughput look excellent. That dynamic is part of why [some workers report heavier workloads as employers raise expectations](https://www.metaintro.com/blog/ai-productivity-trap-workers-heavier-workloads-employers-raise-expectations-2026). ## Are Workers Now Asking AI Instead of Their Manager? Increasingly, and it is the behavioural change underneath the verification problem. [Inc. reported](https://www.inc.com/iza-montalvo/employees-used-to-ask-their-managers-now-theyre-asking-ai/91399406) on September 3 that a growing share of workers now turn to AI before reaching out to a manager, drawing on the same Work AI Index research and on earlier survey work about how people use AI at work. There is an obvious appeal to it. Asking a model costs nothing socially. It does not interrupt anyone, does not reveal what you do not know, and answers immediately at any hour. For questions that are genuinely factual, that is a straightforward improvement over waiting for a reply. The problem is that a large share of workplace questions are not factual. When someone asks a manager how to handle a situation, they are frequently asking about context that exists nowhere in writing: what this client is sensitive about, why the last version was rejected, which internal politics are in play. A model cannot know any of that, and it will answer anyway. Substituting a confident generic answer for a colleague's specific one is a reasonable description of how work gets shipped that its author cannot fully stand behind. ## Does This Make You More Exposed at Work? Yes, and in a way that is worth understanding before it becomes a problem. Accountability has not moved. If your name is on a document, an analysis or a piece of code, you own it regardless of what produced the first draft. No employer has adopted a policy in which a model shares responsibility for an error, and none is likely to. That matters more now because employers have started measuring AI use directly. Some [companies now track employee AI usage in performance reviews](https://www.metaintro.com/blog/companies-track-employees-ai-usage-performance-reviews), which creates an awkward incentive: you may be assessed on how much you use the tools and separately held responsible when unverified output goes wrong. Those two pressures push in opposite directions, and the resolution has to come from you rather than from a policy. The practical exposure is concentrated in a specific moment, which is the handover. Work you produce and check yourself carries normal risk. Work you produce quickly and pass on without checking carries the risk that the error surfaces later, in front of someone senior, attached to your name and with no record of what you verified. The 69 percent figure suggests that moment is now extremely common rather than rare. ## What Should You Actually Do Differently? Separate the tasks where speed is the point from the tasks where being right is the point, and use the tools accordingly. Drafting, summarising, reformatting and first-pass research are places where an unchecked output costs little because you will read it anyway. Numbers, names, dates, citations, client-facing commitments and anything that will be acted on by someone else are places where verification is the job rather than an optional extra. Build one habit rather than a system: check the things a model is known to get wrong. That means every figure, every proper noun, every date and every source. Most AI errors in professional work are not subtle reasoning failures, they are confident specifics that are simply incorrect, and a two-minute check catches the large majority of them. If you cannot verify something, say so explicitly when you hand it over rather than letting the recipient assume it was checked. Be able to describe your process out loud. The single most valuable professional skill in this environment is being able to answer what you used the tool for, what you verified, and what you would not delegate to it. That answer is now a standard interview question in some sectors, it is what a manager is really asking when they query a piece of work, and it is the difference between someone who uses AI well and someone who is in the 69 percent. Employers pushing an [AI-first strategy the team actually trusts](https://www.metaintro.com/blog/how-to-tell-team-going-ai-first-2026) are looking for exactly that capability. ## Is This a Reason to Use AI Less? No, and the data does not support that reading. Adoption at 87 percent with 11 hours a week saved is a real productivity gain that individuals are experiencing, and stepping back from the tools while colleagues use them is not a serious career strategy. The finding is about how the tools are used, not whether they should be. The more useful conclusion is that the differentiator has shifted. When almost everyone has access to the same capability, using it stops being an advantage and using it well becomes one. In a workforce where 69 percent ship work they cannot fully stand behind, being reliably in the other 31 percent is a genuine professional distinction, and it is one that shows up in the quality of what you hand over rather than in a line on your résumé. There is a related point about learning that early-career workers should take seriously. If you use a model to produce work you do not understand, you have delivered the task without acquiring the skill. Over months, that compounds into a career where your output looks competent and your underlying capability has not moved. The people most exposed to that are the ones with the least experience to check against, which is why the [entry-level productivity bar rising](https://www.metaintro.com/blog/2026-entry-level-squeeze-ai-raised-productivity-bar-new-hires) is a more complicated development than it first appears. ## How Does This Compare With Other Research on AI Output? Consistently, which is what makes it credible. Independent work from different sources keeps landing in the same place: adoption is high, individual time savings are real, and organisational returns are much harder to find. The [US Bureau of Labor Statistics](https://www.bls.gov/) has not yet published occupational productivity measures that isolate AI use, so the evidence base is still survey-driven rather than statistical, and that is worth stating plainly rather than glossing over. Two comparisons are useful, and both come from bodies that track this independently of any vendor. The [US Census Bureau](https://www.census.gov/hfp/btos/) runs a recurring business survey that asks firms directly whether they use AI to produce goods and services, which measures adoption at the organisation level rather than the individual one. The [Stanford Institute for Human-Centered AI](https://hai.stanford.edu/ai-index) publishes an annual index compiling adoption, investment and capability data across sources, and the [National Bureau of Economic Research](https://www.nber.org/) publishes the working papers where economists test whether any of it shows up in measured productivity. The practical implication is that you should be sceptical of both the most enthusiastic and the most dismissive claims you encounter. AI is measurably changing how individual work gets done and has not yet measurably changed how most organisations perform. Both of those can be true simultaneously, and holding both is the accurate position rather than a fence-sitting one. For a job seeker, that means describing concrete personal outcomes rather than repeating vendor claims about transformation, because interviewers in 2026 have heard the latter many times. ## What Should Employers Take From This? That the verification gap is a systems problem rather than a discipline problem. If 69 percent of users are shipping unchecked work, the explanation is not that 69 percent of people are careless. It is that the workflow rewards speed, provides no defined checkpoint, and offers no safe way to flag uncertainty. Any organisation seeing this pattern is looking at its own process design. The fix that works is unglamorous: define which categories of output require verification, make it acceptable to mark something as unchecked, and stop measuring AI adoption as though usage volume were the goal. Tracking how much staff use the tools without tracking whether the output holds up produces exactly the numbers in this research. Empowering managers has been identified as [the key to company-wide AI adoption](https://www.metaintro.com/blog/gartner-key-company-wide-ai-adoption-empowering-managers-2026), and this is a concrete instance of what that means in practice. There is also a training gap underneath the figures. Adoption running ahead of instruction is a recognised pattern, and [rolling out AI before training people](https://www.metaintro.com/blog/ai-adoption-gap-2026-employer-training-failure) reliably produces confident misuse rather than cautious competence. The organisations converting AI into performance rather than just into individual time savings are the ones that taught people where the tools fail, not just how to open them. ## Related Articles - [The AI Trust Gap and Why Managers and Workers Cannot Agree on Automation](https://www.metaintro.com/blog/ai-new-workplace-issue-dividing-managers-employees-2026) - [AI Job Fears Are Real But the Data Shows Adoption Is Still Too Immature](https://www.metaintro.com/blog/ai-job-fears-vs-immature-adoption-2026) - [Only a Fraction of Junior Staff Say They Have the AI Tools They Need](https://www.metaintro.com/blog/ai-access-gap-junior-staff-tools-2026) - [AI Anxiety Is Climbing at Work Even as Productivity Promises Grow](https://www.metaintro.com/blog/ai-anxiety-workplace-2026) - [Free and Low-Cost Ways to Learn AI Skills That Employers Trust](https://www.metaintro.com/blog/free-and-low-cost-ways-to-learn-ai-skills) - [The Work Task Everyone Hates Is the One AI Is Not Allowed to Do](https://www.metaintro.com/blog/admin-misery-index-ai-clerical-work-employers-block) - [Amazon Pushes Half-Baked AI Tools on Employees and Creates More Work](https://www.metaintro.com/blog/amazon-200-billion-ai-bet-backfiring-workforce-2026) - [How AI Hiring Tools Can Quietly Screen Out Qualified Candidates](https://www.metaintro.com/blog/how-ai-hiring-tools-screen-out-qualified-candidates) - [The AI Job Search Mistakes That Get You Screened Out](https://www.metaintro.com/blog/ai-job-search-mistakes-that-get-you-screened-out) - [AI Adoption Soars Among Desk Workers in India](https://www.metaintro.com/blog/desk-workers-ai-india) - [Gen Z's AI Revolution and What It Means for Careers and Productivity](https://www.metaintro.com/blog/gen-z-ai-tools-careers) - [AI Coding Tools May Soon Cost More Than Human Developers](https://www.metaintro.com/blog/ai-coding-tools-cost-more-than-developers-2028) ## People Also Asked ### Q: What is the Work AI Index? A: It is research published by the enterprise AI company [Glean](https://www.glean.com/) that surveys digital workers about how they use AI at work and what it changes. The 2026 edition reports 87 percent adoption among digital workers, 11 hours a week of reported time saved, 13 percent of workers saying their organisation performs significantly better, and 69 percent admitting to shipping work they have not verified or cannot confidently stand behind. ### Q: Am I responsible if AI produces an error in my work? A: In practice, yes. Accountability sits with the person whose name is on the output, and no mainstream employer policy treats a model as a party that shares responsibility for a mistake. That is why the verification step matters more than the drafting step, particularly for figures, names, dates and anything a colleague will act on without rechecking. ### Q: How do I show good AI judgment in an interview? A: Describe a specific task, which tool you used, what it got wrong, and how you checked the output before anyone relied on it. Interviewers are testing whether you know where the tools fail rather than whether you can operate them. Being able to name something you deliberately would not delegate to a model is usually the strongest part of that answer. Using AI well is quickly becoming a hiring criterion rather than a personal preference, and the difference shows up in whether you can stand behind what you hand over. [Metaintro](https://www.metaintro.com) tracks how employers are writing AI expectations into job descriptions across sectors. [Join free](https://www.metaintro.com/signup) to see which roles are asking for it and what they expect you to prove. ### Share this article [https://x.com/intent/tweet?text=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index](https://x.com/intent/tweet?text=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index&title=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index&title=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind)[mailto:?subject=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index](mailto:?subject=69%20Percent%20of%20AI%20Users%20Admit%20Shipping%20Work%20They%20Cannot%20Stand%20Behind&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fworkers-shipping-unverified-ai-work-index) For job seekers ## Skip the job boards. Keep the career. Tell Metaintro your story once. It reads every posting, every day, and brings you only what fits. [Get Started Free](/signup)[Search matching jobs](/jobs/search) - MatchCompare live roles against your current evidence. - PositionTurn proof projects into role-specific applications. - ImproveUse market feedback to keep the skill plan current. [Return to navigation](#main-navigation)