--- title: "1,222 Adults Tested | Metaintro" canonical: "https://www.metaintro.com/blog/ai-10-minutes-impairs-brain-cognition" language: "en" author: "drashtigarach" published: "2026-05-12T11:39:16.000Z" modified: "2026-05-12T11:39:16.000Z" --- [Back to Blog](/blog) [AI](/blog/tag/ai)[ChatGPT](/blog/tag/chatgpt)[Technology](/blog/tag/technology)[News](/blog/tag/news)[Information](/blog/tag/information) # 1,222 Adults Tested: Even 10 Minutes of AI Costs You IQ at Work A new preprint of 1222 people finds 10 minutes of AI use measurably weakens problem-solving once the tool is taken away. What it means for workers. [![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) [May 12, 2026](/blog/archive/2026/05)12 min read ![1,222 Adults Tested: Even 10 Minutes of AI Costs You IQ at Work](https://cdn.metaintro.com/rs:fill:1200:675/q:78/plain/images/splitscreen.png) [https://x.com/intent/tweet?text=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition](https://x.com/intent/tweet?text=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition&title=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition&title=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work)[mailto:?subject=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition](mailto:?subject=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition) A new working paper making rounds among cognitive scientists has a finding that should make every knowledge worker pause before they open a chatbot tab. After just 10 to 15 minutes of using a generative AI assistant, participants performed measurably worse on problem-solving tasks once the tool was taken away. The decline was not subtle, and it was not limited to lazy users. The researchers, drawn from four major universities, are framing the effect as a [warning sign for anyone who treats AI as a permanent crutch](https://www.metaintro.com) rather than a temporary aid. Here is what the paper actually says, what it does not say, and what the careers implications look like. ## What did the study actually find? The [preprint](https://arxiv.org/abs/2604.04721), posted in late April 2026 and covered by [Fast Company](https://www.fastcompany.com/91539907/cognitive-science-scientists-found-using-generative-ai-chatgpt-impairs-brain-performance-thinking-problem-solving-skills) and [Digital Trends](https://www.digitaltrends.com/cool-tech/even-brief-ai-use-could-hurt-your-ability-to-think-a-new-study-finds/), pooled three randomized controlled trials run by researchers at Carnegie Mellon University, the University of Oxford, MIT, and UCLA. In total 1222 adults completed tasks built around two staples of cognitive science: a graded set of fraction-based math problems and a reading-comprehension battery. Half the participants worked alone, while the other half had access to an AI assistant powered by OpenAI's GPT-5. For the final three problems in the math trial, the AI was removed without warning. The headline result is straightforward. While the AI was available, the assisted group outperformed the controls, exactly as you would expect. The moment the assistant disappeared, the assisted group's solve rate dropped to 71%, against 77% in the control group. The participants who had used AI most aggressively, asking it to produce full solutions rather than nudges, showed a 10-point decline from their own baseline and skipped roughly twice as many questions as the control. In other words, the people who let the model do their thinking lost the most ground when the model went away. ## Why does brief AI reliance damage cognitive performance? The mechanism the authors describe is cognitive offloading: when you delegate the hard parts of a task, you stop practicing them. Educational psychologists have studied this for decades in the context of calculators, GPS, and spell-check, and the broad finding has been [well documented in the Journal of Experimental Psychology](https://www.apa.org/pubs/journals/xap) and replicated repeatedly. The novelty of generative AI is the speed and breadth of offloading. Where a calculator only takes the arithmetic, a chatbot can take the framing, the strategy, the synthesis, and the wording, all in a single prompt. There is also a dopamine angle. A [2025 paper in Nature Human Behaviour](https://www.nature.com/nathumbehav/) on instant-answer tools found that the speed of frictionless responses can suppress the desire to wrestle with a problem, the same brain circuitry implicated in short-form video consumption. The new preprint does not measure neural activity directly, but its authors note that the persistence drop, that increase in skipped questions, looks more like motivation collapse than skill loss. People did not forget how to do fractions in 10 minutes. They lost the appetite to try. ## How does this fit earlier research on AI and the brain? The preprint is not landing in a vacuum. In February 2025, [Microsoft Research and Carnegie Mellon published a survey](https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/) of 319 knowledge workers and 936 self-reported AI-assisted tasks. That study found that higher confidence in AI was associated with less critical thinking, and that workers using AI produced a less diverse set of outputs than peers who worked unaided. The authors wrote that overreliance left judgment "atrophied and unprepared" for the exceptions where human reasoning is most needed, a finding [Fortune covered in depth](https://fortune.com/2025/02/11/ai-impact-brain-critical-thinking-microsoft-study/) earlier this year. A separate [MIT Media Lab preprint, "Your Brain on ChatGPT"](https://www.media.mit.edu/publications/your-brain-on-chatgpt/), used EEG on a smaller sample of 54 participants writing essays under three conditions: AI, search engine, or unaided. Brain-connectivity networks were strongest in the unaided group, weakest in the AI group, and the AI-to-unaided group later showed lingering deficits, what the authors call cognitive debt. That study has drawn methodological [critiques](https://arxiv.org/abs/2601.00856) over sample size and EEG analysis, and it is worth weighing carefully, but its directional finding lines up with the new four-university preprint. Different methods, different tasks, similar conclusion. ## What should workers and students do about it? The mitigation in the data is specific, and it is good news. The new preprint splits AI users into two groups: those who asked for direct solutions, and those who asked for hints, definitions, or partial explanations. The hint-seekers held onto their unaided performance. The answer-seekers did not. That distinction maps cleanly onto an emerging best practice that learning scientists call active AI use, where the user keeps the cognitive driver's seat and treats the model as a study partner rather than a substitute brain. A simple working rule: draft first, then prompt. Write your rough answer, your messy outline, your first-pass code, and only then bring the AI in to review, challenge, or extend. Researchers at [Stanford's Digital Economy Lab](https://digitaleconomy.stanford.edu/) have argued for the same pattern in workplace settings, calling it human-in-the-lead rather than human-in-the-loop. The order matters. Use the model after your thinking, not before it. The other mitigation is a boundary on offloading. Some tasks should stay unaided on purpose, the way long-distance runners build base mileage without headphones. Reserve a portion of your workweek, perhaps the first half-hour of deep work, for analysis you do entirely on your own. The point is not to be a purist. The point is to keep the muscle warm so that when AI is unavailable, ambiguous, or simply wrong, you can still produce. ## Which roles are most exposed to cognitive offloading? Not every job carries the same risk. The cognitive scientists behind the new preprint single out three categories of work where the offloading effect compounds fastest: analytical writing, multi-step problem solving, and creative synthesis. Those are precisely the tasks that anchor the highest-paid white-collar roles, from legal research to financial analysis to product strategy. The [Bureau of Labor Statistics](https://www.bls.gov/) projects above-average growth through 2033 for management analysts and operations researchers, both occupations defined by exactly the kind of reasoning the preprint says is most vulnerable to AI dependence. The risk is not theoretical. A [2025 BCG study of consultants](https://www.bcg.com/publications/2023/how-people-create-and-destroy-value-with-gen-ai) using generative AI found that workers leaning heavily on the model produced higher-quality output on tasks inside its competence frontier, and noticeably worse output on tasks just outside it. The model gave them false confidence, and the workers stopped checking. Replace consulting with paralegal research, journalism, or product management, and the pattern transfers. The jobs most attractive to AI augmentation are also the jobs where unmonitored reliance does the most damage to underlying skill. Early-career workers face a sharper version of the same problem. Senior analysts and editors built their judgment over years of unaided practice, and they can spot when the model is wrong. Recent graduates who learned the job alongside a chatbot may never build that baseline at all, a concern the preprint's authors flag directly. For anyone within five years of starting a career, the implication is uncomfortable but clear: the skills that make you promotable are the same skills the model is quietly eroding. The exposure is uneven across industries too. In law, finance, and software engineering, where output is reviewed line by line by senior staff, the offloading risk is partially absorbed by the review layer. In marketing, journalism, and consulting, where the work product is the polished narrative itself, there is no second pair of eyes catching the moment the model stopped helping and started replacing. Workers in those fields carry the cognitive debt directly, and the new preprint suggests it accumulates faster than their managers can spot. The implication is not to avoid AI in these fields, but to track honestly which parts of your output are still yours. ## How are employers updating their AI policies in response? The corporate response is moving faster than most workers realize. According to a [PwC pulse survey from March 2026](https://www.pwc.com/us/en/library/pulse-survey.html), 62% of large employers have either issued or are drafting AI usage guidelines that explicitly require unaided work for certain task categories. The most common carve-outs are first-draft analysis, client-facing reasoning, and any output that will be defended in front of a regulator. The pattern reflects a growing recognition that the company, not the individual, absorbs the legal and reputational risk when AI-assisted work goes wrong. Hiring practices are shifting in parallel. Roughly a third of Fortune 500 companies have piloted in-person assessment days that include unaided writing or analytical tasks, according to recruitment data tracked by [SHRM](https://www.shrm.org/). The goal is to verify that a strong portfolio reflects a strong worker, not a strong prompt library. For candidates, the takeaway is that an AI-polished take-home assignment is no longer enough on its own. Recruiters now expect to see the underlying judgment in a setting where the chatbot is closed. The performance review cycle is the next frontier. HR analysts at [Gallup](https://www.gallup.com/workplace.aspx) have noted a growing employer interest in AI-attribution disclosures, where workers note which deliverables were AI-assisted and which were not. The intent is not to punish AI use, but to make the cognitive contribution visible. In a market where AI-fluent workers are abundant, the differentiator is the ability to demonstrate independent judgment when it counts. ## What does this mean for your career? If 10 minutes of passive AI use can dent measurable performance, the compounding effect over a five-day workweek is the part that should concern career-focused readers. Employers are already noticing. Hiring managers surveyed by [Resume Genius](https://resumegenius.com/) in early 2026 said they are increasingly screening candidates for what they call AI-independence, the ability to demonstrate reasoning, judgment, and writing without a chatbot at the keyboard. Live whiteboard interviews, in-office writing exercises, and case studies without internet access are coming back, and they are filtering out candidates whose unaided performance lags their AI-assisted portfolio. The labor-market implication is sharper than it looks. Surveys from [Gartner](https://www.gartner.com/en/newsroom) over the past year suggest that 85% of large employers are expanding AI agent roles, and at the same time most are reporting layoffs tied to AI investments that have not delivered the expected ROI. That combination means the workers who keep their jobs will be the ones who can both leverage AI for speed and operate cleanly without it when it matters. If you want to stay valuable, you cannot afford the cognitive debt the new preprint is measuring. You can stay current with hiring signals and labor-market shifts through [Metaintro's job market intelligence](https://www.metaintro.com/blog), which tracks how AI is reshaping demand role by role. The shortest version of all of this: AI is a power tool, and power tools build skill when you use them with care and erode skill when you use them as a substitute for thought. The cognitive scientists are giving us a stopwatch on the erosion, and 10 minutes is the new number to know. There is also a generational layer to the corporate shift. A survey from the Society for Human Resource Management found that managers under 40 are roughly twice as likely as their older peers to penalize an AI-only deliverable when the candidate cannot defend the underlying reasoning. The reasoning is pragmatic. Younger managers grew up with the tools and recognize the gap between a polished prompt and original analysis faster than executives who treat AI output as a black box. The same survey found that more than half of managers expect to add at least one unaided thinking task to their interview process within the next 12 months, and a third said they had already done it. The screening shift is not coming, it has already started. ## People Also Asked ### Q: Does using AI for 10 minutes really hurt your brain? A: The preprint shows measurable performance drops on unaided problem-solving after 10 to 15 minutes of AI use, but it does not show structural brain damage. The effect is functional and short-term, similar to how relying on GPS dulls your sense of direction in unfamiliar cities. ### Q: Is it safer to use AI for hints rather than full answers? A: Yes, according to the same preprint. Participants who prompted the AI for hints, definitions, or partial explanations retained near-baseline performance once the tool was removed, while those who asked for full solutions saw the largest declines and the highest skip rates. ### Q: How can knowledge workers protect their critical thinking while using AI daily? A: The pattern most learning scientists recommend is draft first, then prompt. Produce your own rough thinking before opening a chatbot, treat the model as a reviewer rather than an author, and protect a portion of your workweek for unaided deep work to keep the underlying skills warm. Future-proof your career by tracking how AI is reshaping the jobs you want and the skills employers are actually paying for. [Metaintro](https://www.metaintro.com) follows the labor-market shifts behind every AI rollout, so you know where to go next. ### Share this article [https://x.com/intent/tweet?text=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition](https://x.com/intent/tweet?text=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition&title=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition&title=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work)[mailto:?subject=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition](mailto:?subject=1%2C222%20Adults%20Tested%3A%20Even%2010%20Minutes%20of%20AI%20Costs%20You%20IQ%20at%20Work&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Fai-10-minutes-impairs-brain-cognition) 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)