{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T06:42:17Z","timestamp":1792478537822,"version":"4.5.0"},"reference-count":42,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T00:00:00Z","timestamp":1792454400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:p>Legal AI research often evaluates large language models (LLMs) on legal reasoning tasks under the assumption that relevant facts are already structured and available. This assumption overlooks a prior and practically consequential stage of legal work: fact construction, namely the transformation of incomplete, fragmented, and interactionally disclosed information into coherent factual narratives. This study conceptualizes fact construction as a distinct evaluation layer in legal AI and introduces a practice-grounded framework for assessing it in AI-assisted legal consultation. Rather than proposing a legal advice system or doctrinal rule-matching architecture, we use a role-separated multi-agent workflow consisting of a fact-constrained Litigant Twin, a structured-questioning LegalAider, a NarrativeAgent for factual synthesis, and an Evaluator for factual-alignment assessment. Using 100 de-identified legal clinic consultation records, we evaluate multiple LLMs under rubric-based assessment, comparing generated factual narratives against original human-authored consultation records in terms of factual accuracy, completeness, and consistency. The results show that LLM-based agents can generate usable factual narratives under constrained conditions, particularly when cases are internally consistent, limited to a single dispute, and supported by clear actor-role relationships. However, cross-model evaluator agreement remains moderate, and outlier analysis reveals recurring structural sources of failure, including anonymization-induced referential ambiguity, third-party reporting, multi-episode case complexity, and numerically dense facts. These findings suggest that reliability in AI-assisted legal consultation depends not only on model capability, but also on interaction design, data representation, institutional record-keeping, and human oversight. By treating fact construction as a distinct analytical layer, this article contributes a practice-grounded framework for evaluating responsible LLM integration in legal settings and highlights the need for factual grounding, contestability, and human-supervised review before downstream legal analysis begins.<\/jats:p>","DOI":"10.3389\/frai.2026.1896166","type":"journal-article","created":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T06:23:28Z","timestamp":1792477408000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Fact construction as a missing layer in legal AI: a practice-grounded framework for evaluating multi-agent legal consultation"],"prefix":"10.3389","volume":"9","author":[{"given":"Hsuan-Lei","family":"Shao","sequence":"first","affiliation":[{"name":"Graduate Institute of Health and Biotechnology Law, Taipei Medical University","place":["Taipei, Taiwan"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ying-Chu","family":"Yu","sequence":"additional","affiliation":[{"name":"College of Law, National Taiwan University","place":["Taipei, Taiwan"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sieh-Chuen","family":"Huang","sequence":"additional","affiliation":[{"name":"College of Law, National Taiwan University","place":["Taipei, Taiwan"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,10,20]]},"reference":[{"key":"B1","volume-title":"Guideline b-2 on Initial Exploration of the Client's Legal Problem. Standards for the Provision of Civil Legal Aid","year":"2021"},{"key":"B2","doi-asserted-by":"publisher","DOI":"10.1017\/CBO9780511610585","author":"Anderson","year":"2005","journal-title":"Analysis of Evidence, 2nd Edn"},{"key":"B3","doi-asserted-by":"publisher","first-page":"71","DOI":"10.1007\/s10588-006-9537-7","article-title":"Introduction to normative multiagent systems","volume":"12","author":"Boella","year":"2006","journal-title":"Comput. Math. Organ. Theory"},{"key":"B4","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2025.naacl-short.39","article-title":"\u201cDebate-feedback: a multi-agent framework for efficient legal judgment prediction,\u201d","volume-title":"Proceedings of the 2025 conference of the nations of the Americas chapter of the association for computational linguistics: human language technologies, Vol. 2 (short papers)","author":"Chen","year":"2025"},{"key":"B5","doi-asserted-by":"publisher","first-page":"177","DOI":"10.1145\/1165485.1165513","article-title":"\u201cEffective document clustering for large heterogeneous law firm collections,\u201d","author":"Conrad","year":"2005"},{"key":"B6","doi-asserted-by":"publisher","first-page":"64","DOI":"10.1093\/jla\/laae003","article-title":"Large legal fictions: profiling legal hallucinations in large language models","volume":"16","author":"Dahl","year":"2024","journal-title":"J. Legal Anal"},{"key":"B7","doi-asserted-by":"publisher","DOI":"10.17159\/2225-7160\/2018\/v51n1a9","article-title":"Clinical legal education: interviewing skills","author":"Du Plessis","year":"2018","journal-title":"De Jure Law J"},{"key":"B8","article-title":"BERT goes to law school: quantifying the competitive advantage of access to large legal corpora in contract understanding","author":"Elwany","year":"2019","journal-title":"arXiv"},{"key":"B9","doi-asserted-by":"publisher","first-page":"157","DOI":"10.3233\/FAIA230961","article-title":"\u201cCan gpt alleviate the burden of annotation?\u201d","author":"Gray","year":"2023","journal-title":"Legal knowledge and information systems-JURIX 2023: the thirty-sixth annual conference, volume 379 of Frontiers in Artificial Intelligence and Applications"},{"key":"B10","article-title":"A survey on LLM-as-a-judge","author":"Gu","year":"2024","journal-title":"arXiv"},{"key":"B11","doi-asserted-by":"crossref","DOI":"10.52202\/075280-1915","article-title":"\u201cLEGALBENCH: a collaboratively built benchmark for measuring legal reasoning in large language models,\u201d","volume-title":"Advances in neural information processing systems 36, Vol. 36","author":"Guha","year":"2023"},{"key":"B12","doi-asserted-by":"publisher","first-page":"86","DOI":"10.3233\/FAIA251579","article-title":"\u201cA two-dimensional evaluation framework for factual and reasoning assessment of LLMs in legal question answering,\u201d","author":"Gultekin","year":"2025","journal-title":"Legal knowledge and information systems-JURIX 2025: the thirty-eighth annual conference, volume 416 of Frontiers in Artificial Intelligence and Applications"},{"key":"B13","article-title":"Good AI legal help, bad AI legal help: establishing quality standards for responses to people's legal problem stories","author":"Hagan","year":"2023","journal-title":"SSRN Scholarly Paper"},{"key":"B14","article-title":"The FACTS grounding leaderboard: benchmarking LLMs' ability to ground responses to long-form input","author":"Jacovi","year":"2025","journal-title":"arXiv"},{"key":"B15","first-page":"589","article-title":"What can information technology do for law?","volume":"21","author":"Jenkins","year":"2008","journal-title":"Harv. J. Law Technol"},{"key":"B16","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2026.acl-long.471","article-title":"Ready jurist one: benchmarking language agents for legal intelligence in dynamic environments","author":"Jia","year":"2025","journal-title":"arXiv"},{"key":"B17","doi-asserted-by":"publisher","DOI":"10.1017\/9781316529683","author":"Katz","year":"2021","journal-title":"Legal Informatics"},{"key":"B18","doi-asserted-by":"publisher","DOI":"10.2139\/ssrn.4976189","article-title":"Large language models (LLMs) for legal advice: a scoping review","author":"Krook","year":"2024","journal-title":"SSRN Scholarly Paper"},{"key":"B19","article-title":"LLMs and legal aid: understanding legal needs exhibited through user queries","author":"Kuk","year":"2025","journal-title":"arXiv"},{"key":"B20","doi-asserted-by":"publisher","first-page":"3","DOI":"10.3233\/FAIA210310","article-title":"Visualization of legal informatics","volume":"3","author":"Lachmayer","year":"2021","journal-title":"J. Vis. Law"},{"key":"B21","doi-asserted-by":"publisher","first-page":"2322","DOI":"10.18653\/v1\/2025.acl-long.116","article-title":"\u201cLegalAgentBench: evaluating LLM agents in legal domain,\u201d","author":"Li","year":"2025","journal-title":"Proceedings of the 63rd annual meeting of the association for computational linguistics, Vol. 1 (long Papers)"},{"key":"B22","doi-asserted-by":"crossref","DOI":"10.1609\/aaai.v38i20.30232","article-title":"\u201cInterpretable long-form legal question answering with retrieval-augmented large language models,\u201d","volume-title":"Proceedings of the AAAI conference on artificial intelligence","author":"Louis","year":"2024"},{"key":"B23","doi-asserted-by":"publisher","first-page":"216","DOI":"10.1111\/jels.12413","article-title":"Hallucination-free? Assessing the reliability of leading ai legal research tools","volume":"22","author":"Magesh","year":"2025","journal-title":"J. Empir. Leg. Stud"},{"key":"B24","doi-asserted-by":"publisher","first-page":"156","DOI":"10.3233\/FAIA251585","article-title":"\u201cPersonaMatrix: a recipe for persona-aware evaluation of legal summarization,\u201d","author":"Pang","year":"2025","journal-title":"Legal knowledge and information systems-JURIX 2025: the thirty-eighth annual conference, volume 416 of Frontiers in Artificial Intelligence and Applications"},{"key":"B25","first-page":"463","article-title":"Introduction to clinical teaching for the new clinical law professor: a view from the first floor","volume":"28","author":"Quigley","year":"1995","journal-title":"Akron Law Rev"},{"key":"B26","volume-title":"Argumentation in Artificial Intelligence","author":"Rahwan","year":"2009"},{"key":"B27","doi-asserted-by":"publisher","first-page":"1279794","DOI":"10.3389\/frai.2023.1279794","article-title":"The unreasonable effectiveness of large language models in zero-shot semantic annotation of legal texts","volume":"6","author":"\u0160avelka","year":"2023","journal-title":"Front. Artif. Intell"},{"key":"B28","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.scichat-1.3","article-title":"Reliable LLM-based user simulator for task-oriented dialogue systems","author":"Sekuli\u0107","year":"2024","journal-title":"arXiv"},{"key":"B29","doi-asserted-by":"publisher","first-page":"2139","DOI":"10.1109\/BigData50022.2020.9378201","article-title":"\u201cCustomizing contextualized language models for legal document reviews,\u201d","author":"Shaghaghian","year":"2020","journal-title":"2020 IEEE international conference on big data (big data)"},{"key":"B30","doi-asserted-by":"publisher","first-page":"218","DOI":"10.1017\/S1472669621000384","article-title":"Emerging legal informatics towards legal innovation: current status and future challenges and opportunities","volume":"21","author":"Sharma","year":"2021","journal-title":"Legal Inf. Manage"},{"key":"B31","doi-asserted-by":"publisher","first-page":"6371","DOI":"10.18653\/v1\/2026.findings-eacl.335","article-title":"\u201cLearning to judge: LLMs designing and applying evaluation rubrics,\u201d","author":"Siro","year":"2026","journal-title":"Findings of the association for computational linguistics: EACL 2026"},{"key":"B32","doi-asserted-by":"publisher","first-page":"345","DOI":"10.1023\/A:1008942012299","article-title":"Multiagent systems: a survey from a machine learning perspective","volume":"8","author":"Stone","year":"2000","journal-title":"Auton. Robots"},{"key":"B33","article-title":"A short survey of viewing large language models in legal aspect","author":"Sun","year":"2023","journal-title":"arXiv"},{"key":"B34","first-page":"193","article-title":"\u201cA data science environment for legal texts,\u201d","volume-title":"Proceedings of the international conference on artificial intelligence and law","author":"Waltl","year":"2015"},{"key":"B35","article-title":"Self-preference bias in LLM-as-a-judge","author":"Wataoka","year":"2024","journal-title":"arXiv"},{"key":"B36","doi-asserted-by":"publisher","first-page":"115","DOI":"10.1007\/3-540-58855-8","article-title":"Intelligent agents: theory and practice","volume":"10","author":"Wooldridge","year":"1995","journal-title":"Knowl. Eng. Rev"},{"key":"B37","doi-asserted-by":"publisher","first-page":"4","DOI":"10.20517\/aiagent.2025.06","article-title":"From single-agent to multi-agent: a comprehensive review of LLM-based legal agents","volume":"1","author":"Yang","year":"2025","journal-title":"AI Agents"},{"key":"B38","article-title":"Justice or prejudice? Quantifying biases in LLM-as-a-judge","author":"Ye","year":"2024","journal-title":"arXiv"},{"key":"B39","doi-asserted-by":"publisher","first-page":"228","DOI":"10.3233\/FAIA251591","article-title":"\u201cA multi-agent ai framework for legal aid: real-world interactive dialogue in case consultations,\u201d","author":"Yu","year":"2025","journal-title":"Legal knowledge and information systems-the thirty-eighth annual conference, JURIX 2025, volume 416 of Frontiers in Artificial Intelligence and Applications"},{"key":"B40","doi-asserted-by":"crossref","DOI":"10.1145\/3583780.3615993","article-title":"\u201cUnleashing the power of large language models for legal applications,\u201d","volume-title":"Proceedings of the 32nd ACM international conference on information and knowledge management","author":"Zhang","year":"2023"},{"key":"B41","article-title":"\u201cModeling future conversation turns to teach LLMs to ask clarifying questions,\u201d","volume-title":"The thirteenth international conference on learning representations","author":"Zhang","year":"2025"},{"key":"B42","doi-asserted-by":"crossref","DOI":"10.52202\/075280-2020","article-title":"\u201cJudging LLM-as-a-judge with MT-Bench and Chatbot Arena,\u201d","volume-title":"Advances in neural information processing systems, Vol. 36","author":"Zheng","year":"2023"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1896166\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,20]],"date-time":"2026-10-20T06:23:31Z","timestamp":1792477411000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1896166\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,10,20]]},"references-count":42,"alternative-id":["10.3389\/frai.2026.1896166"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1896166","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,10,20]]},"article-number":"1896166"}}