Does GenAI actually save lawyers time on fact verification?

Based on semi-structured interviews with 18 non-litigation lawyers (7 male, 11 female; corporate transactions, cross-border compliance, contract review), the paper reports that lawyers currently use GenAI "for low-risk tasks like drafting and language optimization" while "concerns over accuracy, confidentiality, and liability are currently limiting its adoption for fact verification." Its discussion argues this limitation has a specific cause: "efficiency collapses not only due to inherent GenAI inaccuracy, but more fundamentally because its opacity obscures the source of error." AI-generated summaries "often appear convincing but provide little visibility into the selection or interpretation of information," so that "verifying it requires retracing the system's steps, often consuming more time than manual work."

The paper's account turns on where lawyers' cognitive burden actually sits. It locates the heaviest load not in legal reasoning itself but in "the upstream work of transforming unstructured information into verifiable elements" — extracting shareholder changes from interview notes, reconciling multiple versions of corporate records, building chronological transaction tables. Because lawyers "carry full accountability for every factual claim they endorse," they cannot delegate this extraction and organization work without inheriting liability for errors they can no longer trace. The paper's proposed resolution is that GenAI becomes useful only "as an infrastructural intermediary" producing "structured intermediate artifacts with explicit provenance" — aligning data across sources, flagging timestamp discrepancies, generating audit trails — so lawyers can "shift their cognitive effort from reconstructing factual coherence to evaluating legal implications." Automation's value, on this account, is "an epistemic design problem": stabilizing the evidentiary substrate, not replacing judgment.

This sharpens the asymmetry Which workplace cues survive AI mediation and which disappear? finds across 1,250 interviews outside law, where provenance is one of the cues workers still protect because it "attaches an artifact to a recognizable source." Here, provenance is not just a social cue colleagues police after the fact — lawyers describe it as the design requirement that decides whether GenAI output is usable at all, because without it the chain of evidentiary interpretation they are personally accountable for breaks. It also complicates How often do legal AI tools actually hallucinate citations?: that paper measures hallucination rates in existing legal-research products, while this one argues that opacity itself, not just the error rate, is what forces costly re-verification — distinct from the task-speed-and-quality measure in Does GPT-4 actually improve the quality of legal analysis?.

The excerpt's claims about lawyers' practices are self-reported from interviews, not observed; the authors' own limitations note that "participants' descriptions of their verification routines and GenAI use may diverge from actual practices." No accuracy or time figures are given for any specific GenAI tool, and the 18-lawyer sample was recruited by snowball sampling from one researcher's network, concentrated in corporate and compliance practice areas. The finding should be read as a design hypothesis distilled from practitioner accounts, not a measured efficiency effect — the implication that follows is that provenance-preserving intermediate artifacts are a testable design target for legal AI tools, not yet a demonstrated gain.