{"id":326842,"date":"2026-10-24T05:00:00","date_gmt":"2026-10-24T02:00:00","guid":{"rendered":"https:\/\/ceotudent.com\/ai-augmented-performance-review-measure-document-real-contribution"},"modified":"2026-10-24T05:00:00","modified_gmt":"2026-10-24T02:00:00","slug":"ai-augmented-performance-review-measure-document-real-contribution","status":"publish","type":"post","link":"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution","title":{"rendered":"The AI-Augmented Performance Review: How to Measure and Document Your Real Contribution When AI Does the Work"},"content":{"rendered":"<p><strong>TL;DR.<\/strong> Performance ratings were noisy long before AI. In Scullen, Mount and Goff&rsquo;s study of two large manager samples, the rater&rsquo;s idiosyncratic tendencies explained 62% and 53% of rating variance; the person&rsquo;s actual performance explained 21% and 25%. A meta-analysis put the mean interrater reliability of supervisory ratings at .52. In the latest governmentwide US federal survey, only 47% of employees said differences in performance are recognized in a meaningful way, the lowest-scoring item. AI adds a second distortion: in four preregistered experiments with 4,439 participants, people helped by AI were rated lazier and less competent than people given the same help by a person or a conventional tool, and in a four-country experiment a standard-chatbot user got 11.83 to 24.72 points less credit (0-100 scale) than someone who produced the same useful result without AI. The CEO move: keep a weekly contribution log that separates outcome, decisions owned, verification performed and leverage created, and bring evidence a rater can check. The student move: read the log monthly to see which parts of your work AI is absorbing and which judgment skills to build next.<\/p>\n<p>This piece is the review-documentation layer on top of the CEOtudent guide to <a href=\"\/en\/how-to-measure-knowledge-work-output\">how to measure knowledge work output<\/a>. Whether to mention AI at all is covered in <a href=\"\/en\/should-you-tell-your-boss-you-use-ai-workplace-disclosure\">should you tell your boss you use AI<\/a>; turning the evidence into a pay case is covered in <a href=\"\/en\/how-to-negotiate-a-raise-when-ai-multiplied-your-output-benchmarks-scripts\">how to negotiate a raise when AI multiplied your output<\/a>.<\/p>\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_84 counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">Table of Contents<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#Why-is-this-review-season-different\" >Why is this review season different?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#How-reliable-is-a-performance-rating-in-the-first-place\" >How reliable is a performance rating in the first place?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#What-does-rater-noise-mean-for-your-evidence\" >What does rater noise mean for your evidence?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#How-do-raters-judge-work-that-AI-helped-with\" >How do raters judge work that AI helped with?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#Who-gets-the-credit-when-AI-helped\" >Who gets the credit when AI helped?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#Where-does-AI-make-your-work-worse-without-you-noticing\" >Where does AI make your work worse without you noticing?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#What-counts-as-your-contribution-when-AI-did-part-of-the-work\" >What counts as your contribution when AI did part of the work?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#The-weekly-contribution-log\" >The weekly contribution log<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-9\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#How-do-you-disclose-AI-use-without-paying-the-penalty\" >How do you disclose AI use without paying the penalty?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-10\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#The-CEO-move-and-the-student-move\" >The CEO move and the student move<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-11\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#What-this-does-not-mean\" >What this does not mean<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-12\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#Frequently-asked-questions\" >Frequently asked questions<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-13\" href=\"https:\/\/ceotudent.com\/en\/ai-augmented-performance-review-measure-document-real-contribution\/#Sources\" >Sources<\/a><\/li><\/ul><\/nav><\/div>\n<h2 id=\"why-is-this-review-season-different\"><span class=\"ez-toc-section\" id=\"Why-is-this-review-season-different\"><\/span>Why is this review season different?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>AI use at work became routine for a sizeable minority before most employers said how they want it evaluated. Pew Research Center found that 21% of US workers said at least some of their work is done with AI in September 2025, up from 16% roughly a year earlier. In a February 2026 Pew survey, 38% of employed adults reported using AI chatbots for tasks at work. On rules, a Pew survey published in February 2025 found that half of workers who are not self-employed say their employer neither encourages nor discourages chatbot use at work; 12% said their employer encourages it and 8% said it discourages it. Among workers who had used chatbots for work, 40% found them extremely or very helpful for doing things more quickly, and 29% found them highly helpful for improving quality.<\/p>\n<p>So a growing share of the work a manager will rate this year was produced with some AI help, often under unwritten rules, and mostly faster rather than better. The rater must decide how much of the result belongs to the person. Without evidence, that decision falls back on impression, which is where reviews are weakest.<\/p>\n<h2 id=\"how-reliable-is-a-performance-rating-in-the-first-place\"><span class=\"ez-toc-section\" id=\"How-reliable-is-a-performance-rating-in-the-first-place\"><\/span>How reliable is a performance rating in the first place?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p><strong>Table 1. What research says about the reliability of performance ratings (figures as printed)<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Study<\/th>\n<th>Data<\/th>\n<th>Finding as printed<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Scullen, Mount and Goff, Journal of Applied Psychology, 2000<\/td>\n<td>Two manager samples (n = 2,350 and n = 2,142), each rated on 3 dimensions by 7 raters: 2 bosses, 2 peers, 2 subordinates and self<\/td>\n<td>Idiosyncratic rater effects 62% and 53% of variance; general and dimensional ratee performance 21% and 25%; random error 11% and 18%<\/td>\n<\/tr>\n<tr>\n<td>Viswesvaran, Ones and Schmidt, Journal of Applied Psychology, 1996<\/td>\n<td>Meta-analysis of rating reliability across 10 dimensions and overall performance<\/td>\n<td>Mean interrater reliability of supervisory ratings of overall job performance: .52; supervisory ratings appear more reliable than peer ratings<\/td>\n<\/tr>\n<tr>\n<td>Jackson et al., Human Performance, 2022<\/td>\n<td>Two reanalysed samples (392 ratees, 244 raters; 342 ratees, 397 raters)<\/td>\n<td>Structure mainly general (more than 27% of variance) and rater-related (more than 49%); dimension effects 1% to 11%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Sources: Scullen et al. and Viswesvaran et al. (abstracts), Jackson et al. (accepted manuscript). Scullen&rsquo;s data were developmental multisource ratings, not pay-linked appraisals.<\/p>\n<p>The .52 figure is debated: Jackson and co-authors note it assumes all rater-related variance is rater bias, an assumption other researchers have challenged, since raters may see different real parts of a job. What travels is the direction: a single rating carries a large share of the rater.<\/p>\n<p>Employees notice. Gallup, a consulting and research company, reported in 2019 that only 14% of employees strongly agree their performance reviews inspire them to improve, and in 2018 that only 21% strongly agree they have performance metrics within their control.<\/p>\n<p>The largest public survey with item-level results is the US Office of Personnel Management&rsquo;s Federal Employee Viewpoint Survey. The 2024 edition (revised April 2025) is the latest governmentwide management report listed on OPM&rsquo;s reports page as of October 2026. Over 674,000 employees responded, a 41% response rate.<\/p>\n<p><strong>Table 2. Accountability versus recognition, 2024 Federal Employee Viewpoint Survey (governmentwide percent positive, as printed)<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Item<\/th>\n<th>Percent positive<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>&ldquo;My supervisor holds me accountable for achieving results.&rdquo; (Q. 53)<\/td>\n<td>88%<\/td>\n<\/tr>\n<tr>\n<td>&ldquo;I am held accountable for the quality of work I produce.&rdquo; (Q. 11)<\/td>\n<td>87%<\/td>\n<\/tr>\n<tr>\n<td>&ldquo;How satisfied are you with the recognition you receive for doing a good job?&rdquo; (Q. 69)<\/td>\n<td>57%<\/td>\n<\/tr>\n<tr>\n<td>&ldquo;In my work unit, differences in performance are recognized in a meaningful way.&rdquo; (Q. 17)<\/td>\n<td>47%, the lowest-scoring item<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Source: US Office of Personnel Management, 2024 FEVS Governmentwide Management Report and item workbook (Q. 17: 26.3% neutral, 27.2% negative).<\/p>\n<p>Nearly nine in ten say they are held accountable for results; fewer than half say performance differences are recognized meaningfully. That 40 to 41 point gap is what contribution looks like when a system can see that work was done but not how well, or whose it was.<\/p>\n<h2 id=\"what-does-rater-noise-mean-for-your-evidence\"><span class=\"ez-toc-section\" id=\"What-does-rater-noise-mean-for-your-evidence\"><\/span>What does rater noise mean for your evidence?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>If one impression is weak evidence, the remedy is more independent evidence, not a more confident impression.<\/p>\n<p><strong>Table 3. Signal, rater noise and the value of independent evidence (CEOtudent analysis of Scullen et al. 2000 and Viswesvaran et al. 1996)<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Measure<\/th>\n<th>Value<\/th>\n<th>Derivation<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Rater idiosyncrasy relative to ratee performance<\/td>\n<td>2.95 times (data set 1); 2.12 times (data set 2)<\/td>\n<td>62\/21; 53\/25<\/td>\n<\/tr>\n<tr>\n<td>Rater idiosyncrasy plus random error<\/td>\n<td>73% and 71% of variance<\/td>\n<td>62 + 11; 53 + 18<\/td>\n<\/tr>\n<tr>\n<td>Reliability of one supervisor&rsquo;s overall rating<\/td>\n<td>.52<\/td>\n<td>As printed<\/td>\n<\/tr>\n<tr>\n<td>Expected reliability, average of 2 \/ 3 \/ 4 independent raters<\/td>\n<td>.68 \/ .77 \/ .81<\/td>\n<td>Spearman-Brown formula<\/td>\n<\/tr>\n<tr>\n<td>Independent raters needed to reach .70 \/ .80<\/td>\n<td>3 \/ 4<\/td>\n<td>Spearman-Brown solved for k, rounded up<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Spearman-Brown assumes parallel, independent raters, which real raters rarely are; treat these as an upper-bound illustration.<\/p>\n<p>You cannot choose how many people rate you. You can choose how many independent, checkable facts stand behind each claim: a metric someone else owns, a decision note written at the time, a confirmed caught defect. The table is an analogy, not a law, but it explains why three modest verifiable facts persuade more reliably than one impressive adjective.<\/p>\n<h2 id=\"how-do-raters-judge-work-that-ai-helped-with\"><span class=\"ez-toc-section\" id=\"How-do-raters-judge-work-that-AI-helped-with\"><\/span>How do raters judge work that AI helped with?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>The strongest evidence comes from Reif, Larrick and Soll of Duke University&rsquo;s Fuqua School of Business (PNAS, 2025).<\/p>\n<p><strong>Table 4. Evidence of a social evaluation penalty for using AI (Reif, Larrick and Soll, 2025; figures as printed)<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Study<\/th>\n<th>Design and sample<\/th>\n<th>Key result<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>1<\/td>\n<td>500 participants imagined using a generative AI tool or a dashboard tool at work (N = 497 analysed)<\/td>\n<td>Expected to be seen as lazier (3.25 vs 2.72) and less competent (4.72 vs 5.45) with AI; less likely to disclose it to managers (4.91 vs 5.25)<\/td>\n<\/tr>\n<tr>\n<td>2<\/td>\n<td>1,215 participants rated an employee from 384 vignettes (N = 1,203 analysed)<\/td>\n<td>AI-helped employee rated lazier (2.50) than one helped by a non-AI source (2.16) or no help (2.02); also less competent, diligent, independent and self-assured; no moderation by age, gender or occupation<\/td>\n<\/tr>\n<tr>\n<td>3<\/td>\n<td>801 real candidates and 1,718 manager participants, incentive-compatible hiring (N = 1,667 analysed)<\/td>\n<td>No average task-fit difference, but managers who use AI less favoured the candidate who never uses AI, while frequent users favoured the daily user<\/td>\n<\/tr>\n<tr>\n<td>4<\/td>\n<td>1,006 participants hiring a gig worker (N = 1,002 analysed)<\/td>\n<td>AI users seen as lazier, lowering task fit. Manual task: 4.75 (traditional tools) vs 4.18 (AI). Digital task where AI was useful: 5.58 vs 5.75, not significant (P = 0.112)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>The 4,439 total counts Study 3&rsquo;s managers, not its candidates.<\/p>\n<p>Three details matter for a review. The penalty concerned attributed effort and ability (laziness, competence, diligence), not ambition or dominance. It depended on the evaluator: in Study 4 the laziness gap was significant among participants who had never used AI, had tried it once or twice or used it occasionally, and not among weekly or daily users. And it depended on fit: where AI was clearly useful for the task, the task-fit penalty disappeared. A rater who sees only &ldquo;used AI&rdquo; fills the gap with an attribution about effort; a rater who sees the decision, the check and why the tool suited the task has something else to credit.<\/p>\n<h2 id=\"who-gets-the-credit-when-ai-helped\"><span class=\"ez-toc-section\" id=\"Who-gets-the-credit-when-AI-helped\"><\/span>Who gets the credit when AI helped?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>Porsdam Mann and colleagues argued in Nature Machine Intelligence (2023) that generative AI creates a credit-blame asymmetry; a follow-up paper summarises it as AI use that &ldquo;elevates the bar for earning credit, but standards for assigning blame remain the same.&rdquo;<\/p>\n<p>Earp, Porsdam Mann and co-authors tested it in a preregistered experiment with representative sampling (N = 1,802) in the UK, US, China and Singapore. A fictional writer published a blog post made with a standard large language model, with a model personalized on the writer&rsquo;s own past work, or without AI, after quickly skimming it. Participants rated credit for a useful post, or blame for a harmful one, from 0 to 100.<\/p>\n<table>\n<thead>\n<tr>\n<th>Country<\/th>\n<th>Credit gap: no AI minus standard LLM<\/th>\n<th>Credit gap: personalized minus standard LLM<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>United Kingdom<\/td>\n<td>18.51 points<\/td>\n<td>14.31 points<\/td>\n<\/tr>\n<tr>\n<td>United States<\/td>\n<td>24.72 points<\/td>\n<td>17.56 points<\/td>\n<\/tr>\n<tr>\n<td>China<\/td>\n<td>11.83 points<\/td>\n<td>5.13 points (not significant)<\/td>\n<\/tr>\n<tr>\n<td>Singapore<\/td>\n<td>Not available (control excluded after a programming error)<\/td>\n<td>17.60 points<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Source: Earp et al., Annals of the New York Academy of Sciences, 2024, Table 1 (Bonferroni-adjusted; standard-minus-personalized contrasts restated as positive gaps).<\/p>\n<p>Credit for a useful post made with a model trained on the writer&rsquo;s own past work was not significantly below credit for the no-AI post in any country with a control group; participants pointed to provenance. Blame barely depended on AI type, except in China, and UK participants blamed the writer more for using any model.<\/p>\n<p>Draxler and colleagues (two studies, n = 30 and n = 96) add the AI Ghostwriter Effect: users did not consider themselves owners of AI-generated text yet did not publicly declare AI authorship. More influence over the text increased their sense of ownership.<\/p>\n<p><strong>Table 5. What moves the judgment of AI-assisted work, and what to document (CEOtudent analysis of Reif et al. 2025, Earp et al. 2024 and Draxler et al. 2023)<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>What the studies varied<\/th>\n<th>Effect on evaluation<\/th>\n<th>Documentation that addresses it<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Source of identical help<\/td>\n<td>Laziness 0.34 higher than for non-AI help, 0.48 higher than no help (Reif Study 2)<\/td>\n<td>Name the effort that stayed yours: framing, decisions, checks<\/td>\n<\/tr>\n<tr>\n<td>Evaluator&rsquo;s own AI use<\/td>\n<td>Penalty significant for non-users and occasional users, not weekly or daily users (Reif Study 4)<\/td>\n<td>Write for the least AI-familiar reader of the review<\/td>\n<\/tr>\n<tr>\n<td>Whether AI suited the task<\/td>\n<td>Task-fit penalty of 0.57 on a manual task; none significant on a digital task (Reif Study 4)<\/td>\n<td>State why the tool fitted and where it was not used<\/td>\n<\/tr>\n<tr>\n<td>Provenance of content<\/td>\n<td>Standard-LLM credit 11.83 to 24.72 points below no-AI control, mean 18.35; personalized LLM not significantly below (Earp)<\/td>\n<td>Show what came from you: data, prior analysis, structure<\/td>\n<\/tr>\n<tr>\n<td>Influence over the text<\/td>\n<td>More influence, more felt ownership (Draxler)<\/td>\n<td>Keep before and after versions of what you changed<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>These are online or laboratory findings, not measured effects on real appraisals; the right-hand column is CEOtudent&rsquo;s inference.<\/p>\n<h2 id=\"where-does-ai-make-your-work-worse-without-you-noticing\"><span class=\"ez-toc-section\" id=\"Where-does-AI-make-your-work-worse-without-you-noticing\"><\/span>Where does AI make your work worse without you noticing?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>In the preregistered experiment by Dell&rsquo;Acqua and colleagues with Boston Consulting Group (758 knowledge workers; Organization Science, 2026), participants using GPT-4 completed 12.2% more tasks and worked 25.1% faster on 18 tasks inside the frontier of AI capability, but on a managerial task outside it they were 19% less likely to produce correct solutions. The authors call this a jagged technological frontier.<\/p>\n<p>That is the strongest argument for recording overrides. Your most valuable act on a project may be refusing a confidently wrong model answer, and it leaves no trace because the error never shipped. The time cost of checking is covered in <a href=\"\/en\/verification-tax-how-much-time-checking-ai-output-costs\">the verification tax<\/a>, the skill in <a href=\"\/en\/the-evaluation-skill-judging-ai-output\">the evaluation skill<\/a>, and the controlled-study gains in <a href=\"\/en\/ai-productivity-paradox-individual-gains-company-results\">the AI productivity paradox<\/a>.<\/p>\n<h2 id=\"what-counts-as-your-contribution-when-ai-did-part-of-the-work\"><span class=\"ez-toc-section\" id=\"What-counts-as-your-contribution-when-AI-did-part-of-the-work\"><\/span>What counts as your contribution when AI did part of the work?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>The Contribution Ladder in the cluster hub says what to measure. This model says what to document.<\/p>\n<p><strong>Table 6. The AI-Era Contribution Model (CEOtudent editorial framework)<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Layer<\/th>\n<th>Evidence that counts<\/th>\n<th>Evidence that does not<\/th>\n<th>How AI changes it<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>A. Outcome delivered<\/td>\n<td>A metric someone else owns moved; a decision was taken on your work; a risk closed<\/td>\n<td>Drafts, decks, tickets, hours, tool statistics<\/td>\n<td>Volume is cheap now, so outcomes carry the weight<\/td>\n<\/tr>\n<tr>\n<td>B. Problem framing and decision rights<\/td>\n<td>Dated note of options, choice, reason and sign-off<\/td>\n<td>A polished document with no record of what was rejected<\/td>\n<td>AI generates options in seconds; choosing and owning one is scarce<\/td>\n<\/tr>\n<tr>\n<td>C. Verification and error-catching<\/td>\n<td>Log of corrected outputs, what each error would have cost, how it was found<\/td>\n<td>&ldquo;Reviewed&rdquo; without saying against what<\/td>\n<td>The jagged frontier makes this layer larger and less visible<\/td>\n<\/tr>\n<tr>\n<td>D. Leverage created for others<\/td>\n<td>Templates, prompts, checklists or data others adopted; reuse counts<\/td>\n<td>Personal speed nobody else can use<\/td>\n<td>AI workflows are easy to share, so leverage is newly measurable<\/td>\n<\/tr>\n<tr>\n<td>E. Learning velocity<\/td>\n<td>Dated new responsibilities; a skill shown in work that was absent last cycle<\/td>\n<td>Certificates with no applied use<\/td>\n<td>The tasks AI absorbs keep changing; standing still is slow decline<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Layers B and C are where AI-assisted work is most under-credited, because they are what the Reif and Earp evaluators could not see. The <a href=\"\/en\/decision-journal-template-protocol-improving-judgment\">decision journal protocol<\/a> is the natural home for layer B, and the <a href=\"\/en\/manager-of-ai-playbook-direct-evaluate-improve\">manager-of-AI playbook<\/a> produces layer C evidence as a side effect.<\/p>\n<h2 id=\"the-weekly-contribution-log\"><span class=\"ez-toc-section\" id=\"The-weekly-contribution-log\"><\/span>The weekly contribution log<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>A short block at the end of each <a href=\"\/en\/weekly-review-protocol-high-output-2026\">weekly review<\/a> turns review season from a memory exercise into an editing exercise.<\/p>\n<p><strong>Table 7. Weekly contribution log fields (CEOtudent editorial framework)<\/strong><\/p>\n<table>\n<thead>\n<tr>\n<th>Field<\/th>\n<th>What to write<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Date and project<\/td>\n<td>Week, project, stakeholder<\/td>\n<\/tr>\n<tr>\n<td>Outcome<\/td>\n<td>The change the work is meant to cause and who owns that metric<\/td>\n<\/tr>\n<tr>\n<td>Decision owned<\/td>\n<td>The choice, the options rejected, the reason<\/td>\n<\/tr>\n<tr>\n<td>AI role<\/td>\n<td>Which parts the tool drafted, summarised or computed, and which it did not touch<\/td>\n<\/tr>\n<tr>\n<td>Verification<\/td>\n<td>What was checked, against which source, what was found<\/td>\n<\/tr>\n<tr>\n<td>Error caught or override<\/td>\n<td>What the model got wrong and what it would have cost<\/td>\n<\/tr>\n<tr>\n<td>Leverage<\/td>\n<td>What others can reuse, and who used it<\/td>\n<\/tr>\n<tr>\n<td>Learning note<\/td>\n<td>What was new, hard or handed to AI for the first time<\/td>\n<\/tr>\n<tr>\n<td>Proof pointer<\/td>\n<td>Where the evidence lives: file, ticket, message, dashboard<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><strong>Illustrative example entry (every value is an illustrative assumption, not data).<\/strong> Week 14, quarterly vendor report. Outcome: report cycle time cut from 5 days to 2 days, a 60% reduction, metric owned by the finance lead. Decision: dropped two unused legacy metrics for one cash-timing measure the finance lead requested. AI role: drafted the narrative and first-pass variance table; did not touch the data extract. Verification: reconciled every total against the ledger export; found two double-counted credit notes. Override: rejected a variance explanation citing a contract clause that does not exist. Leverage: reconciliation checklist now used by two colleagues. Learning: first cash-timing measure; next skill is forecasting it. Proof: version history and checklist file.<\/p>\n<h3 id=\"the-self-review-paragraph-template\">The self-review paragraph template<\/h3>\n<p>Replace each bracket and delete any sentence you cannot support.<\/p>\n<blockquote>\n<p>In [period], [project] delivered [outcome], measured by [metric] owned by [stakeholder], moving from [baseline] to [result]. The key decision was [choice] over [alternative] because [reason]. AI tools were used for [specific parts]; [named parts] were done without them. Every [figure or claim] was checked against [source], which caught [number] errors, including [the one that mattered], which would have [consequence]. The [template or workflow] built during this work is now used by [who]. The next step is [harder responsibility], which requires building [skill].<\/p>\n<\/blockquote>\n<p>The AI sentence is short, factual and tied to parts of the work. It answers the question a rater would otherwise answer with an assumption, and it attaches the tool to a task where it fitted, the condition under which the Reif penalty disappeared.<\/p>\n<h3 id=\"the-six-question-test-for-each-accomplishment\">The six-question test for each accomplishment<\/h3>\n<p>Keep accomplishments that pass at least five.<\/p>\n<ol>\n<li><strong>Owner test.<\/strong> Is the outcome measured by a number or decision someone else owns?<\/li>\n<li><strong>Counterfactual test.<\/strong> Would the result have been worse, later or wrong had the tool&rsquo;s first output been accepted?<\/li>\n<li><strong>Decision test.<\/strong> Can the rejected option and the reason be named?<\/li>\n<li><strong>Verification test.<\/strong> Is it clear what was checked, against what, and what was found?<\/li>\n<li><strong>Proof test.<\/strong> Is there an artifact a sceptical rater could open in two minutes?<\/li>\n<li><strong>Disclosure test.<\/strong> Is the AI involvement described accurately, at the level your employer&rsquo;s policy and client agreements require?<\/li>\n<\/ol>\n<p>Question 6 cuts both ways. Overstating your share of AI-produced work is a credibility risk and often a policy breach; understating your judgment because a tool was involved is a measurement error against yourself.<\/p>\n<h2 id=\"how-do-you-disclose-ai-use-without-paying-the-penalty\"><span class=\"ez-toc-section\" id=\"How-do-you-disclose-AI-use-without-paying-the-penalty\"><\/span>How do you disclose AI use without paying the penalty?<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>The disclosure decision by policy regime is in the CEOtudent disclosure guide. For the review document, the experiments suggest three rules. <strong>Disclose method, not a confession:<\/strong> the Study 2 penalty attached to a bare mention of AI help, so say what the tool did and what you did. <strong>Anchor the tool to task fit:<\/strong> the Study 4 penalty vanished where AI was clearly useful, so say why it fitted. <strong>Show provenance:<\/strong> credit recovered in the Earp experiment when output rested on the author&rsquo;s own work, so point to your data, analysis, prompts and checklists. None of this guarantees a fair rating; it narrows the space in which an impression substitutes for evidence.<\/p>\n<h2 id=\"the-ceo-move-and-the-student-move\"><span class=\"ez-toc-section\" id=\"The-CEO-move-and-the-student-move\"><\/span>The CEO move and the student move<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p><strong>The CEO move.<\/strong> Treat the review like a board pack: outcomes, the decisions behind them, risks caught, capabilities built. Keep the weekly log, choose three to five accomplishments that pass the six-question test, attach a proof pointer to each, and ask beforehand which metric your manager weighs most.<\/p>\n<p><strong>The student move.<\/strong> Monthly, read the &ldquo;AI role&rdquo; and &ldquo;learning note&rdquo; fields together. Tasks that moved to &ldquo;drafted by AI&rdquo; are being absorbed; decide whether that frees time for harder work or erodes a skill you still need. Repeated catches of model errors mark a judgment skill worth deepening. Over a year this becomes your map of <a href=\"\/en\/career-capital-ai-era-what-compounds-what-decays\">what compounds and what decays<\/a>.<\/p>\n<h2 id=\"what-this-does-not-mean\"><span class=\"ez-toc-section\" id=\"What-this-does-not-mean\"><\/span>What this does not mean<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<ul>\n<li><strong>Ratings are not random.<\/strong> Reliability around .52 still carries information; one impression is simply weaker than several independent facts.<\/li>\n<li><strong>The AI penalty will not necessarily happen to you.<\/strong> The experiments used online samples, vignettes and short tasks, not real appraisals with history and context.<\/li>\n<li><strong>The federal figures do not describe your employer,<\/strong> and Gallup&rsquo;s figures come from a company that sells performance-management services.<\/li>\n<li><strong>Hiding AI use is not the answer.<\/strong> Concealment carries policy, contractual and credibility risks the experiments did not measure. Follow your employer&rsquo;s policy.<\/li>\n<li><strong>This is general information about published research, not legal or HR advice.<\/strong> Rules on monitoring, disclosure and appraisal differ by country, sector and contract.<\/li>\n<\/ul>\n<h2 id=\"frequently-asked-questions\"><span class=\"ez-toc-section\" id=\"Frequently-asked-questions\"><\/span>Frequently asked questions<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p><strong>Can someone claim an outcome if AI drafted most of the deliverable?<\/strong><br \/>\nYes, for the parts that were theirs: framing, decisions, verification and the accountable outcome, with the AI role described accurately. The Earp experiment suggests evaluators give full credit when the result clearly rests on the person&rsquo;s own prior work.<\/p>\n<p><strong>Should a self-review mention AI at all?<\/strong><br \/>\nFollow employer policy first. Where none exists, one specific sentence about what the tool did and what the person did is usually safer than silence or a vague confession.<\/p>\n<p><strong>How many accomplishments belong in a review?<\/strong><br \/>\nNo evidence-based number exists. Several checkable facts per claim matter more than a long list; three to five accomplishments that pass the six-question test is a practical CEOtudent rule of thumb, not a research finding.<\/p>\n<p><strong>What if the manager is sceptical of AI?<\/strong><br \/>\nThat is the condition with the strongest penalty in the Reif experiments. Explain task fit, show checks and lead with outcomes owned by other people.<\/p>\n<p><strong>Does catching AI errors count as performance?<\/strong><br \/>\nIn the Dell&rsquo;Acqua experiment, AI users were 19% less likely to reach correct solutions on a task outside the tool&rsquo;s capability. Preventing that is real value, but only if it is recorded with what was found and what it would have cost.<\/p>\n<p><strong>How is this different from the Contribution Ladder?<\/strong><br \/>\nThe ladder classifies metrics by how much they reveal about value. This model classifies the evidence to document so a rater can see that value, with a separate verification layer.<\/p>\n<h2 id=\"sources\"><span class=\"ez-toc-section\" id=\"Sources\"><\/span>Sources<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<ol>\n<li>Scullen, S. E., Mount, M. K., and Goff, M. (2000). Understanding the latent structure of job performance ratings. Journal of Applied Psychology, 85(6), 956-970 (abstract only, via Europe PMC).<\/li>\n<li>Viswesvaran, C., Ones, D. S., and Schmidt, F. L. (1996). Comparative analysis of the reliability of job performance ratings. Journal of Applied Psychology, 81(5), 557-574 (abstract only, as indexed by the Shanghai University of Finance and Economics Library).<\/li>\n<li>Jackson, D. J. R., et al. (2022). Uncertainty about rater variance and small dimension effects impact reliability in supervisor ratings. Human Performance, 35(3-4), 278-301 (accepted manuscript).<\/li>\n<li>Gallup (2019). More Harm Than Good: The Truth About Performance Reviews. Company-published article.<\/li>\n<li>Gallup (2018). Re-Engineering Performance Management. Company-published article.<\/li>\n<li>US Office of Personnel Management (2025). 2024 Federal Employee Viewpoint Survey Results: Governmentwide Management Report, revised April 2025.<\/li>\n<li>US Office of Personnel Management. 2024 FEVS governmentwide all levels, all index, all items report (xlsx).<\/li>\n<li>Reif, J. A., Larrick, R. P., and Soll, J. B. (2025). Evidence of a social evaluation penalty for using AI. Proceedings of the National Academy of Sciences, 122(19), e2426766122.<\/li>\n<li>Porsdam Mann, S., Earp, B. D., et al. (2023). Generative AI entails a credit-blame asymmetry. Nature Machine Intelligence, comment (standfirst only).<\/li>\n<li>Earp, B. D., Porsdam Mann, S., Liu, P., et al. (2024). Credit and blame for AI-generated content: Effects of personalization in four countries. Annals of the New York Academy of Sciences, 1542, 51-57.<\/li>\n<li>Draxler, F., et al. (2023). The AI Ghostwriter Effect: When Users Do Not Perceive Ownership of AI-Generated Text But Self-Declare as Authors. arXiv 2303.03283v2.<\/li>\n<li>Dell&rsquo;Acqua, F., et al. Navigating the Jagged Technological Frontier. Harvard Business School Working Paper 24-013; Organization Science, March-April 2026 (abstract only).<\/li>\n<li>Pew Research Center (2025). Workers&rsquo; experience with AI chatbots in their jobs. 25 February 2025.<\/li>\n<li>Pew Research Center (2025). 21% of US workers use AI on the job, up since 2024. 6 October 2025.<\/li>\n<li>Pew Research Center (2026). Americans and AI 2026: Chatbots, Smart Devices and Views on Impact. 17 June 2026.<\/li>\n<\/ol>\n<p><em>Tables 3 and 5 are CEOtudent analyses that recompute or combine the published figures named in each table; Tables 6 and 7, the self-review template and the six-question test are CEOtudent editorial frameworks, not findings of the cited studies. The example log entry uses invented values labelled as illustrative. Widely repeated figures deliberately not used: claims that a fixed share of companies has abandoned annual reviews, and vendor survey percentages on managers using AI to write reviews, because no primary text for them was opened for this piece.<\/em><\/p>\n<hr>\n<p><em>This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"<p>In a classic study of two large manager samples, rater idiosyncrasy explained 62% and 53% of rating variance, against 21% and 25% for actual performance. Four preregistered experiments (N = 4,439) now show that people known to use AI are judged lazier and less competent. This guide gives a contribution log, a self-review template and a six-question test so the rater has evidence to check instead of an impression to guess.<\/p>\n","protected":false},"author":1,"featured_media":326847,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5,18],"tags":[],"class_list":["post-326842","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-is","category-strateji"],"_links":{"self":[{"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/posts\/326842","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/comments?post=326842"}],"version-history":[{"count":0,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/posts\/326842\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/media\/326847"}],"wp:attachment":[{"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/media?parent=326842"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/categories?post=326842"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/tags?post=326842"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}