[{"data":1,"prerenderedAt":324},["ShallowReactive",2],{"\u002Fresearch\u002Finsights\u002Fmeasuring-trust-in-ai-tools-at-work":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"author":11,"readTime":12,"category":13,"area":14,"body":15,"_type":318,"_id":319,"_source":320,"_file":321,"_stem":322,"_extension":323},"\u002Fresearch\u002Finsights\u002Fmeasuring-trust-in-ai-tools-at-work","insights",false,"","Methodology spotlight: how knowledge workers actually trust AI tools","A series of peer-reviewed studies measured pharmacists using AI-assisted decision tools — and what they found generalises directly to enterprise software UX.","2026-03-30","Pagegazer team","8 min","Methodology spotlight","Behavioral & Custom",{"type":16,"children":17,"toc":308},"root",[18,26,31,38,51,56,62,74,86,92,118,124,143,149,162,168,173,203,208,214,295],{"type":19,"tag":20,"props":21,"children":22},"element","p",{},[23],{"type":24,"value":25},"text","Trust in AI is the question of the moment for enterprise software. Every B2B SaaS roadmap in 2026 includes an AI feature, a copilot, an agentic step. The question every vendor is asking — and very few are measuring well — is whether their users actually trust the AI's output enough to act on it, or are quietly working around it.",{"type":19,"tag":20,"props":27,"children":28},{},[29],{"type":24,"value":30},"A connected series of peer-reviewed studies — run on the platform that powers Pagegazer — measured exactly this. The studied population was pharmacists doing medication verification with AI assistance. It is a high-stakes, real-task domain, but the methods and findings transfer directly to general enterprise UX research on AI features.",{"type":19,"tag":32,"props":33,"children":35},"h2",{"id":34},"trust-is-not-uniform-and-depends-on-demographics-and-context",[36],{"type":24,"value":37},"Trust is not uniform — and depends on demographics and context",{"type":19,"tag":20,"props":39,"children":40},{},[41,43,49],{"type":24,"value":42},"Whitaker, Rowell, Kim, Al Kontar, Yang, and Lester (2026), in the ",{"type":19,"tag":44,"props":45,"children":46},"em",{},[47],{"type":24,"value":48},"Journal of the American Pharmacists Association",{"type":24,"value":50},", measured pharmacists' propensity to trust automated technologies as a function of demographic and contextual factors. The headline result is that trust is not a single value users carry into an interaction — it varies systematically by experience level, prior exposure to automation, and the kind of task. Two pharmacists looking at the same AI suggestion bring different priors and behave differently.",{"type":19,"tag":20,"props":52,"children":53},{},[54],{"type":24,"value":55},"The implication for enterprise UX research is that population-level satisfaction scores hide the variance that matters. A 7.4 on a trust scale can mean \"everyone trusts it 7.4\" or \"half are 9, half are 5 and quietly ignoring it.\" The latter is a product problem; the former isn't.",{"type":19,"tag":32,"props":57,"children":59},{"id":58},"showing-ai-uncertainty-changes-what-knowledge-workers-do",[60],{"type":24,"value":61},"Showing AI uncertainty changes what knowledge workers do",{"type":19,"tag":20,"props":63,"children":64},{},[65,67,72],{"type":24,"value":66},"Lester, Rowell, Zheng, Co, Marshall, Kim, Chen, Kontar, and Yang (2025) ran a randomized controlled trial measuring how ",{"type":19,"tag":44,"props":68,"children":69},{},[70],{"type":24,"value":71},"uncertainty-aware AI models",{"type":24,"value":73}," — AI that displays its confidence — change pharmacists' reaction time and decision-making in a mock medication verification task. Showing uncertainty did not just shift trust ratings; it changed the behaviour itself: how long pharmacists spent looking at the AI's suggestion, when they overrode it, and how they paced themselves through verification.",{"type":19,"tag":20,"props":75,"children":76},{},[77,79,84],{"type":24,"value":78},"This was published in ",{"type":19,"tag":44,"props":80,"children":81},{},[82],{"type":24,"value":83},"JMIR Medical Informatics",{"type":24,"value":85}," — a journal where claims have to clear methodological bars not unlike a clinical trial. The study was randomised, controlled, and large enough to distinguish behavioural change from noise. The implication is that displaying confidence is not a UI detail. It is part of the decision system the user is operating in.",{"type":19,"tag":32,"props":87,"children":89},{"id":88},"helpfulness-and-uncertainty-interact",[90],{"type":24,"value":91},"Helpfulness and uncertainty interact",{"type":19,"tag":20,"props":93,"children":94},{},[95,97,102,104,109,111,116],{"type":24,"value":96},"In a related study, Tsai et al. (2025) extended the question to how ",{"type":19,"tag":44,"props":98,"children":99},{},[100],{"type":24,"value":101},"AI helpfulness",{"type":24,"value":103}," and ",{"type":19,"tag":44,"props":105,"children":106},{},[107],{"type":24,"value":108},"AI uncertainty",{"type":24,"value":110}," interact in shaping pharmacist cognition, published in the ",{"type":19,"tag":44,"props":112,"children":113},{},[114],{"type":24,"value":115},"Journal of Medical Internet Research",{"type":24,"value":117},". Showing high helpfulness without uncertainty produced different patterns of cognitive engagement than showing helpfulness with uncertainty cues. The combination of the two is the design surface — not either independently.",{"type":19,"tag":32,"props":119,"children":121},{"id":120},"earlier-stage-exploration-ai-with-different-uncertainty-representations",[122],{"type":24,"value":123},"Earlier-stage exploration: AI with different uncertainty representations",{"type":19,"tag":20,"props":125,"children":126},{},[127,129,134,136,141],{"type":24,"value":128},"A related earlier study by Kim and colleagues (2025), in ",{"type":19,"tag":44,"props":130,"children":131},{},[132],{"type":24,"value":133},"JMIR Human Factors",{"type":24,"value":135},", presented different forms of AI uncertainty information (probabilistic, categorical, qualitative) and measured pharmacists' trust response across them. The finding was specific: ",{"type":19,"tag":44,"props":137,"children":138},{},[139],{"type":24,"value":140},"how",{"type":24,"value":142}," uncertainty is shown changes the trust response measurably, and the optimal representation depends on the user's prior experience with automation. This is actionable for any enterprise team designing how their AI feature surfaces confidence.",{"type":19,"tag":32,"props":144,"children":146},{"id":145},"self-reported-trust-diverges-from-observed-behaviour",[147],{"type":24,"value":148},"Self-reported trust diverges from observed behaviour",{"type":19,"tag":20,"props":150,"children":151},{},[152,154,160],{"type":24,"value":153},"The methodological lesson across all four studies is that ",{"type":19,"tag":155,"props":156,"children":157},"strong",{},[158],{"type":24,"value":159},"self-reported trust and observed trust behaviour are not the same construct.",{"type":24,"value":161}," Saying \"I trust this\" in a survey is informative; pausing on the AI suggestion for 1.4× longer than baseline before accepting it is a different kind of informative. Both should be measured. When they diverge, the observed behaviour is usually the better predictor of what will happen at scale.",{"type":19,"tag":32,"props":163,"children":165},{"id":164},"what-this-means-for-enterprise-ux-research",[166],{"type":24,"value":167},"What this means for enterprise UX research",{"type":19,"tag":20,"props":169,"children":170},{},[171],{"type":24,"value":172},"The same methods — eye tracking on the AI suggestion field, reaction-time measurement, decision tracking, in the user's own browser — answer the questions enterprise software teams actually have:",{"type":19,"tag":174,"props":175,"children":176},"ul",{},[177,183,188,193,198],{"type":19,"tag":178,"props":179,"children":180},"li",{},[181],{"type":24,"value":182},"Where does the user's eye go first when the AI's suggestion appears? Do they read it, or skip past?",{"type":19,"tag":178,"props":184,"children":185},{},[186],{"type":24,"value":187},"How long do they spend evaluating it before accepting or overriding?",{"type":19,"tag":178,"props":189,"children":190},{},[191],{"type":24,"value":192},"Does adding a confidence indicator change accept rates — and does it change them in the right direction (more rejection of low-confidence outputs, sustained acceptance of high-confidence ones)?",{"type":19,"tag":178,"props":194,"children":195},{},[196],{"type":24,"value":197},"Which segments of users adopt the AI feature and which work around it, and is that segmentation visible in the behaviour before it shows up in feature-adoption analytics?",{"type":19,"tag":178,"props":199,"children":200},{},[201],{"type":24,"value":202},"Which uncertainty representation works best for which segment of your user base?",{"type":19,"tag":20,"props":204,"children":205},{},[206],{"type":24,"value":207},"The pharmacist studies are the clearest demonstration in the published literature that these questions can be measured precisely, with methods that scale to a representative sample and don't require lab visits. The same instrumentation runs for any knowledge worker using any software in any browser.",{"type":19,"tag":32,"props":209,"children":211},{"id":210},"citations",[212],{"type":24,"value":213},"Citations",{"type":19,"tag":174,"props":215,"children":216},{},[217,238,257,276],{"type":19,"tag":178,"props":218,"children":219},{},[220,222,227,229],{"type":24,"value":221},"Whitaker, M., Rowell, B., Kim, J. Y., Al Kontar, R., Yang, X. J., & Lester, C. A. (2026). ",{"type":19,"tag":44,"props":223,"children":224},{},[225],{"type":24,"value":226},"Pharmacists' propensity to trust automated technologies: A demographic analysis.",{"type":24,"value":228}," Journal of the American Pharmacists Association. ",{"type":19,"tag":230,"props":231,"children":235},"a",{"href":232,"rel":233},"https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.japh.2025.103011",[234],"nofollow",[236],{"type":24,"value":237},"doi.org\u002F10.1016\u002Fj.japh.2025.103011",{"type":19,"tag":178,"props":239,"children":240},{},[241,243,248,250],{"type":24,"value":242},"Lester, C., Rowell, B., Zheng, Y., Co, Z., Marshall, V., Kim, J. Y., Chen, Q., Kontar, R., & Yang, X. J. (2025). ",{"type":19,"tag":44,"props":244,"children":245},{},[246],{"type":24,"value":247},"Effect of uncertainty-aware AI models on pharmacists' reaction time and decision-making in a web-based mock medication verification task.",{"type":24,"value":249}," JMIR Medical Informatics. ",{"type":19,"tag":230,"props":251,"children":254},{"href":252,"rel":253},"https:\u002F\u002Fdoi.org\u002F10.2196\u002F64902",[234],[255],{"type":24,"value":256},"doi.org\u002F10.2196\u002F64902",{"type":19,"tag":178,"props":258,"children":259},{},[260,262,267,269],{"type":24,"value":261},"Tsai, C. C., Kim, J. Y., Chen, Q., Rowell, B., Yang, X. J., Kontar, R., et al. (2025). ",{"type":19,"tag":44,"props":263,"children":264},{},[265],{"type":24,"value":266},"Effect of Artificial Intelligence Helpfulness and Uncertainty on Cognitive Interactions with Pharmacists.",{"type":24,"value":268}," Journal of Medical Internet Research. ",{"type":19,"tag":230,"props":270,"children":273},{"href":271,"rel":272},"https:\u002F\u002Fdoi.org\u002F10.2196\u002F59946",[234],[274],{"type":24,"value":275},"doi.org\u002F10.2196\u002F59946",{"type":19,"tag":178,"props":277,"children":278},{},[279,281,286,288],{"type":24,"value":280},"Kim, J. Y., Marshall, V. D., Rowell, B., Chen, Q., Zheng, Y., Lee, J. D., Al Kontar, R., Lester, C., & Yang, X. J. (2025). ",{"type":19,"tag":44,"props":282,"children":283},{},[284],{"type":24,"value":285},"The Effects of Presenting AI Uncertainty Information on Pharmacists' Trust in Automated Pill Recognition Technology.",{"type":24,"value":287}," JMIR Human Factors. ",{"type":19,"tag":230,"props":289,"children":292},{"href":290,"rel":291},"https:\u002F\u002Fdoi.org\u002F10.2196\u002F60273",[234],[293],{"type":24,"value":294},"doi.org\u002F10.2196\u002F60273",{"type":19,"tag":20,"props":296,"children":297},{},[298,300,306],{"type":24,"value":299},"For the full list of peer-reviewed work using the Pagegazer measurement platform, see ",{"type":19,"tag":230,"props":301,"children":303},{"href":302},"\u002Fresearch\u002Fpublished",[304],{"type":24,"value":305},"published research",{"type":24,"value":307},".",{"title":7,"searchDepth":309,"depth":309,"links":310},2,[311,312,313,314,315,316,317],{"id":34,"depth":309,"text":37},{"id":58,"depth":309,"text":61},{"id":88,"depth":309,"text":91},{"id":120,"depth":309,"text":123},{"id":145,"depth":309,"text":148},{"id":164,"depth":309,"text":167},{"id":210,"depth":309,"text":213},"markdown","content:research:insights:measuring-trust-in-ai-tools-at-work.md","content","research\u002Finsights\u002Fmeasuring-trust-in-ai-tools-at-work.md","research\u002Finsights\u002Fmeasuring-trust-in-ai-tools-at-work","md",1780554430168]