[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273668-105":59,"doc-detail-273668-en":135},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":128,"head_meta":130,"extra_data":132,"updated_unix":134},105,"en","personalbench-measuring-the-authorship-gap-in-llm-personalization","PERSONALBENCH - Measuring the Authorship Gap in LLM Personalization","","Personalized text generation aims to make LLM outputs resemble a specific individual’s writing style, but existing benchmarks mostly test task performance or preference alignment rather than stylistic authorship similarity. PERSONALBENCH evaluates inference-time personalization using three complementary lenses: LUAR authorship verification, an LLM-as-judge, and automated stylometrics. Results over 50 authors, 1,000 generations, and two model families show author-differentiated outputs (LUAR gen↔gen AUC=0.918) that still remain confined within the LLM style space and do not reach human author resemblance (gen→real LUAR below cross-author human floor).",{"@graph":69,"@context":127},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/personalbench-measuring-the-authorship-gap-in-llm-personalization/273668/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/personalbench-measuring-the-authorship-gap-in-llm-personalization/273668.png","ImageObject",300,407,{"name":92,"@type":93},"Rowan","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-05","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",13,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119,123],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does PERSONALBENCH address in existing LLM personalization benchmarks?","Question",{"text":113,"@type":114},"Existing benchmarks focus on task accuracy or preference alignment, not whether generated text actually resembles the target author’s writing style. PERSONALBENCH evaluates authorship similarity directly.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How does PERSONALBENCH measure whether personalization changes writing style?",{"text":118,"@type":114},"It uses three independent lenses: LUAR authorship verification, an LLM-as-judge, and automated stylometrics. This combination targets the shift toward the target author’s authorship fingerprint.",{"name":120,"@type":111,"acceptedAnswer":121},"What key finding is reported about the authorship gap between humans and personalized LLMs?",{"text":122,"@type":114},"Personalization creates author-differentiated outputs within the LLM regime (LUAR gen↔gen AUC=0.918), but it does not bridge to human authorship similarity (gen→real LUAR remains below the human cross-author floor).",{"name":124,"@type":111,"acceptedAnswer":125},"Why do LLM-judge results appear to differ from LUAR results?",{"text":126,"@type":114},"The document traces the discrepancy to circularity between trait extraction and profile extraction used by the judge. LUAR does not show the corresponding advantage, indicating the judge’s signal may not reflect genuine target-author style resemblance.","https://schema.org",{"og:url":83,"og:type":129,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":131,"canonical":83},"index,follow",{"doc_id":133,"site_id":62},273668,1790644615,{"code":4,"msg":5,"data":136},{"doc_id":133,"user_id":137,"nickname":92,"user_avatar":138,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":139,"file_id":140,"file_url":141,"file_type":142,"file_size":143,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":144,"language":145,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":146,"faqs":147,"seo_title":148,"seo_description":67,"update_tm":149,"read_time":150,"preview_page_count":4},1099514067415,"https://ap-avatar.wpscdn.com/avatar/100002539d78ffe74a7?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779092875211072502","arXiv : 2608 . 19746v 1 [cs .CL] 20 Aug 2026  \nPERSONALBENCH: Measuring the Authorship Gap in  \nLLM Personalization  \nYash Ganpat Sawant  \nIndependent AI Researcher  \n[sawantyash13@gmail.com](sawantyash13@gmail.com)  \nAbstract  \nPersonalized text generation aims to make LLMs write in a specific individual’s style, yet existing benchmarks measure task accuracy or preference alignment rather than whether the model’s output actually resembles the target author’s writing. We introduce PERSONALBENCH, a benchmark that evaluates inferencetime personalization methods through three independent lenses: LUAR (a trained authorship verification model), an LLM-as-judge, and automated stylometrics.  \nAcross 50 authors, 1,000 generations, and two model families (Qwen 3, GLM-4), we find that personalization methods do produce author-differentiated output—LUAR discriminates target authors within generated text at AUC=0.918—but this differentiation never crosses the human–LLM boundary. All methods achieve LUAR similarity to real authors in the range 0.484–0.508, below the cross-author human floor of 0.626 (ceiling 0.756) . The LLM’s own authorship fingerprint dominates: generated text is more distant from any human author than random humans are from each other. Methods are statistically indistinguishable from each other on LUAR (spread 0.024) despite appearing differentiated on the LLMjudge—a discrepancy we trace to circularity between trait extraction and profile extraction. We validate that LUAR reliably measures authorship in our corpus (AUC=0.76 single-post, 0.96 multi-post) . We release PERSONALBENCH as a calibrated measuring stick: inference-time personalization modulates the LLM’s style but does not bridge the gap to human authorship.1  \n1 Introduction  \nLarge language models are increasingly deployed where personalization matters: writing assistants matching a user’s voice, customer agents maintaining brand tone, and content tools adapting to individual preferences. These applications share a core requirement: the model must write like a specific person. The standard approach is inference-time personalization—conditioning the model via few-shot examples, style profiles, or contrastive prompts at generation time, without modifying model weights.  \nBut does this actually work? Current personalization benchmarks cannot answer this question because they measure the wrong thing. LaMP [Salemi et al., 2024] evaluates task accuracy (headline generation, product rating) . PersonalLLM [Zollo et al., 2025] measures preference alignment. PRISM [Kirk et al., 2024] captures stated value preferences. None evaluate whether the generated text sounds like the target author—whether the model’s underlying authorship fingerprint actually shifts toward the target.  \nWe ask a direct question: do inference-time personalization methods change how the model writes, or only what it writes about? To answer this, we need evaluation that goes beyond surface-  \n1 Code and data: [https://github.com/yashsawant22/personalbench](https://github.com/yashsawant22/personalbench).  \nPreprint.  \nlevel text similarity. We need metrics grounded in authorship science—the decades-old discipline of identifying authors from their writing style.  \nWe introduce PERSONALBENCH, a benchmark that evaluates personalization through three independent lenses:  \n1. LUAR [Rivera-Soto et al., 2021]: a neural authorship verification model trained on millions of Reddit posts, providing calibrated similarity scores. Our primary metric.  \n2. LLM-as-judge: decoupled binary trait matching and same-author assessment, providing interpretable style diagnostics.  \n3. Automated stylometrics: function word distributions, punctuation patterns, and lexical overlap. Our central finding is that inference-time personalization produces author-differentiated output that remains trapped in the LLM’s own style space. Within generated text, LUAR discriminates target authors at AUC=0.918—personalization is doing someth","cbCaiiFqxe1IRaoU","https://ap.wps.com/l/cbCaiiFqxe1IRaoU","pdf",409202,17,"English","# Abstract\n# Introduction\n# Related Work","[{\"question\":\"What problem does PERSONALBENCH address in existing LLM personalization benchmarks?\",\"answer\":\"Existing benchmarks focus on task accuracy or preference alignment, not whether generated text actually resembles the target author’s writing style. PERSONALBENCH evaluates authorship similarity directly.\"},{\"question\":\"How does PERSONALBENCH measure whether personalization changes writing style?\",\"answer\":\"It uses three independent lenses: LUAR authorship verification, an LLM-as-judge, and automated stylometrics. This combination targets the shift toward the target author’s authorship fingerprint.\"},{\"question\":\"What key finding is reported about the authorship gap between humans and personalized LLMs?\",\"answer\":\"Personalization creates author-differentiated outputs within the LLM regime (LUAR gen↔gen AUC=0.918), but it does not bridge to human authorship similarity (gen→real LUAR remains below the human cross-author floor).\"},{\"question\":\"Why do LLM-judge results appear to differ from LUAR results?\",\"answer\":\"The document traces the discrepancy to circularity between trait extraction and profile extraction used by the judge. LUAR does not show the corresponding advantage, indicating the judge’s signal may not reflect genuine target-author style resemblance.\"}]","PERSONALBENCH - Measuring the Authorship Gap in LLM Personalization | PDF",1789472424,43]