Meet Marcus
Marcus is a paralegal at a twenty-person litigation firm that handles commercial disputes, employment law matters, and professional liability cases for corporate clients and high-net-worth individuals. He has been with the firm for four years, having joined directly after completing his legal studies, and has developed a strong working knowledge of civil procedure, discovery practice, and litigation document management. He works within a team structure where he supports two to three attorneys simultaneously, with the attorney-paralegal relationship in his firm being relatively stable: each paralegal is assigned to a consistent group of attorneys rather than being deployed across the practice as a whole, which means Marcus has developed deep familiarity with the working styles, preferences, and expectations of his supervising attorneys.
At any given point, Marcus is actively working across ten to fifteen open matters at different stages of the litigation lifecycle. Some are in the discovery phase, generating large volumes of documents requiring review, organisation, and analysis. Others are in the pleadings or motion practice phase, requiring the preparation and filing of court documents to precise procedural standards. Others are approaching trial preparation, requiring the assembly and organisation of exhibit materials, witness preparation documents, and trial binders. Managing the specific requirements of each matter in its current phase, while maintaining awareness of upcoming deadlines across all active matters, is the central operational challenge of Marcus's working week.
The nature of litigation paralegal work creates specific constraints that shape the AI practice described in this walkthrough. Attorneys bear ultimate professional responsibility for all work product and all client communications, and the firm's professional conduct rules and indemnity insurer requirements create clear boundaries around what paralegals may do without attorney review and approval. These constraints are not bureaucratic obstacles. They are the professional framework within which litigation support work must be conducted, and the AI practice described here is built within rather than around them.
Marcus's primary pain points before establishing an AI practice are three. First, the time cost of large discovery productions. Commercial litigation in his firm's practice areas regularly involves document productions from opposing counsel that run to thousands of individual documents, each of which must be reviewed, assessed for relevance and privilege, logged, and analysed for its bearing on the key legal issues in the matter. Manual review of a large production can consume days of continuous work before the attorneys can be briefed on what the production contains and how it affects the litigation strategy. Second, the time cost of legal research. Research tasks assigned by attorneys require identifying relevant authorities, reviewing case law, and synthesising findings into a usable form, a process that currently requires Marcus to conduct Boolean keyword searches in legal databases, read through large result sets to identify the genuinely relevant cases, and then synthesise his findings manually into a research memo. Third, the practical frustration of finding relevant precedents. The firm has accumulated significant work product from previous matters that would be directly useful in current matters, but the organisation of that work product is uneven and retrieval is unreliable without a clear picture of where relevant material is stored.
The Before State
Before establishing an AI practice, Marcus's case file organisation reflects a structure that was adopted when he joined the firm and that has accumulated inconsistencies over four years of active use. Files are organised at the top level by attorney rather than by case: each attorney has a folder, and within that folder cases are stored in a variety of arrangements that differ between attorneys and have evolved over time as each attorney's preferences changed. The consequence of attorney-level organisation is that cross-case search is unreliable: finding all documents in the firm's file system that relate to a specific legal issue, a specific opposing counsel, or a specific procedural motion type requires knowing which attorney handled the relevant matters and navigating to their folder, rather than being able to search by case or legal issue directly.
The deeper problem with attorney-level organisation is its fragility when assignments change. When an attorney leaves the firm, takes leave, or transfers responsibility for a matter to a colleague, the files for the relevant matters remain in the departing attorney's folder, associated by location with a professional who is no longer responsible for the matter. Marcus has experienced this problem directly on two occasions and has spent significant time relocating materials and reconstructing context when matter assignments changed unexpectedly.
Discovery document review in Marcus's current practice is almost entirely manual. When a production arrives from opposing counsel, Marcus receives a collection of documents, typically in PDF format, and begins reviewing them document by document to assess relevance, flag potentially privileged material, identify documents bearing on the key legal issues in the matter, and log each document in a master index. For a modest-sized production of a few hundred documents, this review takes several days. For large productions in the hundreds or thousands, it can consume a full working week or more, during which other matters receive less attention than they require.
Legal research tasks currently follow a standard database search workflow: Marcus constructs search queries in Boolean syntax, reviews the result sets, identifies the most relevant authorities, reads those authorities in full, and drafts a research memo from notes taken during reading. The process is reliable but slow, and the manual synthesis of multiple authorities into a coherent research memo is the most time-consuming component. Marcus estimates that a typical research assignment, from receiving the task to delivering a draft memo to the supervising attorney, currently takes between half a day and a full day depending on the complexity of the legal question and the volume of relevant authority.
Knowledge Base Setup
Marcus's knowledge base is built around a case-primary folder structure that replaces the attorney-level organisation of his previous approach. The structure is:
Cases / Year / CaseNumber-ClientName / Pleadings, Discovery, Correspondence, Research, Archive
The decision to organise by case rather than by attorney reflects the principle from Module 4.1 that the top level of a folder structure should represent the most stable unit of organisation. Cases have a defined identity, a unique identifier, and a durable existence as organisational units that persists regardless of which attorney is responsible for them at any given time. Attorneys are assigned to cases, not the reverse, and an organisation built around the case remains coherent when assignments change.
The Year folder at the second level separates active cases from closed matters without mixing them indiscriminately and aligns with how the firm's conflict checking and matter management systems organise cases administratively. Within the year, each case folder is identified by the case number, which is the firm's unique matter identifier and the reference used in all official correspondence and court filings, combined with the client name in a format that allows identification without opening the folder.
Within each case folder, four subfolders and one archive folder address the fundamental categories of litigation document work.
Pleadings holds all court filings: complaints, answers, motions, briefs, and associated exhibits. Within Pleadings, filed documents are distinguished from draft documents through a consistent naming convention that includes the filing date for filed documents and the designation Draft for documents in preparation. This separation is critical in litigation practice, where the distinction between what has been filed with the court and what is still in preparation has direct procedural significance.
Discovery holds all materials relating to the discovery process: documents received from opposing counsel in productions, deposition transcripts, interrogatory responses, and the master index and review notes that track the discovery record. A subfolder structure within Discovery separates incoming productions, which are numbered sequentially as they arrive, from the analysis and index documents produced from review of those productions.
Correspondence holds all external written communication relating to the matter: letters and emails exchanged with opposing counsel, communications with the court, and correspondence with clients that has been approved by the supervising attorney for retention in the case file. Internal communications, such as emails between Marcus and the supervising attorney about litigation strategy, are held separately and are not retained in the case correspondence folder.
Research holds legal research materials: case law downloads, statutory text, regulatory materials, secondary sources, and the research memos produced from analysis of these materials. Research is organised within the folder by legal issue rather than by date or by source, reflecting the way research is used: a paralegal returning to the matter's research needs to find all materials relating to a specific legal issue, not all materials produced in a specific month.
Archive holds superseded drafts, resolved interim motions, and documents from closed phases of the litigation that should be retained for the matter record but are no longer active working material. Moving documents to Archive when their phase of the matter closes keeps the active folders navigable and prevents AI tools searching within the case folder from surfacing irrelevant historical material alongside current documents.
Marcus maintains four context documents for each active matter, each stored at the top level of the case folder.
The Case Summary Document is the primary orientation document for the matter. It records the parties and their roles, the nature of the claims and defences, the key facts of the underlying dispute, the current procedural status of the matter, and the litigation strategy as Marcus understands it from discussions with the supervising attorney. It also records the key people involved: not only the parties but the key witnesses, expert witnesses retained or anticipated, opposing counsel and their firm, and the relevant court personnel. This document serves as the README document described in Module 4.1: it orients anyone, whether human or AI tool, who encounters the case folder for the first time, providing a complete context for the documents it contains.
The Discovery Master Index is a running log of all documents received in discovery productions, with a record of the production in which each document was received, the document type, a brief description of the document's content, an assessment of relevance to the key legal issues, and a flag for any document that has been identified as potentially significant for the litigation strategy. This index is updated throughout the discovery review process and becomes an increasingly valuable reference as the matter's discovery record grows.
The Legal Issues Memo records the specific legal questions that are active in the matter: the legal standards that must be met to establish or defeat each claim, the jurisdictional authority applicable to those standards, and the current state of the legal analysis. This document ensures that AI-assisted discovery review and legal research is consistently directed toward the legal issues that are actually in dispute rather than toward tangentially related material.
The Deadlines and Status Tracker is a structured record of all upcoming procedural deadlines, the current status of each active workstream, and any open questions requiring resolution by the supervising attorney. In litigation, procedural deadlines are strictly enforced, and the consequences of missing a court-imposed deadline can be severe. This tracker provides a single reference for the matter's procedural obligations that is more reliable than relying on memory or on searching through correspondence to reconstruct the deadline schedule.
Model and Tool Selection
Marcus's primary AI tool is Gemini, selected on the basis of the task type matching framework from Module 4.2. His most demanding AI-assisted tasks involve processing very large volumes of documents, many of which are individually lengthy. Deposition transcripts can run to hundreds of pages, and a large discovery production may include extensive contracts, financial records, and correspondence that each require careful reading and analysis in the context of the specific legal issues in the matter. Gemini's exceptionally large context window allows Marcus to process massive volumes of material in a single session without the accuracy degradation that affects models working at the edge of their context capacity. Furthermore, Gemini's strong cross-document synthesis and retrieval capabilities align perfectly with the professional demands of legal discovery, where identifying subtle factual connections across hundreds of disparate files is critical to building an accurate case timeline.
For firm-wide document search and retrieval, Marcus uses the native AI search features of the firm's document management system. This native integration provides a specific advantage for legal practice. Documents retrieved through the firm's document management system remain entirely within the system's access control and audit logging framework. AI-assisted retrieval of matter documents therefore bypasses the need to transmit those documents to an external system. For a law firm with strict professional confidentiality obligations governing all matter-related documents, this provides a significant governance advantage over connecting an external AI tool directly to the firm's document repository.
Marcus has made two deliberate integration decisions that carry as much importance as the connections he has built. First, he maintains zero email integration in his AI practice. The firm's professional conduct policies require attorney review and approval of all client-facing communications. The data handling implications of connecting an external AI tool to a system containing privileged attorney-client communications are highly significant from a privilege perspective, as addressed in Module 4.1. Marcus drafts client-facing communications manually using AI assistance strictly for the initial drafting step. He submits each draft to the supervising attorney for review and approval before sending. He deliberately avoids maintaining a connected integration that would give an external AI tool standing access to the firm's email system.
Second, Marcus avoids AI assistance for the final preparation and submission of court filings. The formatting requirements of court filings include pagination, margin specifications, font requirements, word count limits, and caption formatting. These rules are jurisdiction and court-specific and subject to frequent change. An error in the formatting of a court filing can result in rejection by the court clerk, which carries severe procedural consequences including missed deadlines. The risk that AI-assisted formatting might introduce or overlook a structural error provides sufficient justification to maintain a fully manual workflow for the final preparation of court documents. AI assistance remains limited entirely to the drafting of the substantive content. Marcus then formats the document manually and checks it meticulously against the applicable court rules before filing.
Workflow One: Discovery Document Review
Discovery document review is the workflow where AI assistance delivers the most significant time reduction in Marcus's practice, and it is also the workflow where the verification requirements are most stringent. The combination of volume and legal significance means that the efficiency gains of AI-assisted review must be achieved without any reduction in the accuracy of the review output.
Step one: Receive and organise the production. When a discovery production arrives from opposing counsel, Marcus creates a new subfolder within the Discovery folder of the relevant case, numbered sequentially and dated. He renames individual documents within the production according to the firm's file naming convention, giving each document a name that indicates the production number, a sequential document number within the production, and a brief document type identifier. Where documents arrive with their own identifiers, such as Bates numbers in US-style productions, those identifiers are incorporated into the file name. This naming step, while taking time upfront, ensures that every document in the production can be located and referenced precisely for the remainder of the matter.
Step two: Review the case summary and legal issues memo. Before beginning AI-assisted document review, Marcus reads the case summary and legal issues memo to refresh his understanding of the key legal issues guiding the review. The legal issues memo is submitted to Gemini as context for all subsequent document review prompts. This step directs the AI tool's summaries and relevance assessments specifically toward the legal questions in dispute.
Step three: Generate AI summaries of each document. For each document in the production, Marcus submits the document content to Gemini with a prompt structured to focus the summary on the legal issues relevant to the matter. A representative prompt structure includes: "Summarise this document. Key legal issues in this matter: [paste relevant section of legal issues memo]. In your summary, identify the document type and its relationship to the parties. Identify any facts relevant to the key legal issues listed. Note any dates, figures, or specific representations with potential significance. Highlight any inconsistencies with other documents or facts described. Flag any aspects of the document considered potentially significant for the litigation." The structured prompt focuses the AI summary directly on the specific dimensions of the document relevant to the legal work.
Step four: Review AI summaries and flag significant documents. Marcus reads each AI summary carefully, comparing the summary against the source document for a sample of documents in each production to verify that the summaries accurately represent the source material. Documents flagged by the AI as potentially significant, and documents that Marcus identifies as significant through his own reading of the summaries, are marked for full attorney review and their significance noted in the discovery master index. Documents assessed as clearly irrelevant are logged in the index with a brief relevance assessment but are not individually reviewed in full.
Step five: Update the discovery master index. Following the review of a production, Marcus updates the discovery master index with an entry for each document reviewed, recording the document's identifier, type, brief content description, relevance assessment, and any flags for attorney attention. The index is maintained as a living document throughout the discovery phase of the matter and becomes a critical reference for attorney briefing, deposition preparation, and the identification of exhibits.
Step six: Assist with privilege log preparation. Where documents in a production include communications between the client and counsel, or documents prepared in anticipation of litigation, these require identification and withholding from production as privileged. A privilege log entry must describe the withheld document without revealing the privileged content. Marcus uses Gemini to assist in drafting privilege log entries for potentially privileged documents. He provides the document metadata and a description of the document sufficient to draft the log entry while explicitly excluding the privileged content. Every privilege log entry produced with AI assistance undergoes review by the supervising attorney before production of the log. Marcus presents these entries exclusively as preliminary drafts for mandatory attorney review.
Workflow Two: Legal Research
Legal research tasks in Marcus's practice follow a structured workflow that uses AI assistance to accelerate the identification and initial synthesis of relevant authority, while preserving the independent verification of every citation and the attorney's review of the finished research memo before it is used in any filing or relied upon in any client advice.
Step one: Clarify the research task with the supervising attorney. Before beginning research, Marcus ensures that he has a precise and complete understanding of the legal question he is researching, the relevant jurisdiction, the procedural context in which the research will be used, and the format in which the attorney wants the research delivered. This clarity is essential for AI-assisted research because the precision of the research question directly determines the quality of the AI's initial analysis. A vague research question produces a broad and superficial AI output. A precisely stated legal question, specifying the jurisdiction, the procedural posture, and the specific issue in dispute, produces a more focused and more useful initial analysis.
Step two: Generate an initial analytical framework and search strategy. Marcus provides the research question and relevant context to Gemini. He requests an initial analysis of the applicable legal framework. This request covers the relevant areas of law, the key distinctions requiring attention, the leading cases in the applicable jurisdiction within Gemini's training data, and a structured set of search terms for use in legal databases. This step generates an analytical map to direct the subsequent database research. This mapping process precedes the retrieval of verified legal authority and structures the approach to keyword searching.
Step three: Conduct primary source research in legal databases. Using the analytical framework and search terms produced in step two, Marcus conducts structured searches in the firm's subscribed legal research databases. He identifies the five to ten cases most directly relevant to the specific legal question, reads those cases in full in the primary source database, and takes notes on the key holdings, the reasoning, and the factual circumstances that make each case relevant or distinguishable.
Step four: Use AI to assist with synthesis. Having read and taken notes on the relevant authorities, Marcus submits his research notes and key case extracts to Gemini. He prompts the model for a comparative synthesis detailing how the authorities address the relevant legal standard, identifying consistencies and tensions in the case law, and applying the legal principles to the current matter's facts. The synthesis produced through this step relies entirely on material Marcus has already verified and selected. This strict grounding anchors the AI's output directly to the provided primary source text.
Step five: Verify every citation independently. Every case cited in the AI-assisted synthesis is independently verified in the primary source database before it is included in the draft research memo. Verification means confirming that the case exists, that it is decided by the court the citation indicates, that it stands for the proposition the synthesis attributes to it, and that it has not been overruled or distinguished in subsequent decisions. This verification step is absolute. No case is included in a research memo on the basis of an AI citation without independent primary source confirmation. The hallucination risk in AI-assisted legal research, as addressed in Section 5 of Module 4.3, makes this verification a professional obligation rather than a precautionary preference.
Step six: Draft the research memo with AI assistance. UUsing the verified synthesis and his own research notes, Marcus produces a draft research memo presenting the applicable legal standard, reviewing the relevant authorities, and applying the legal principles to the facts of the matter. Gemini assists with the structure and drafting of this memo. Marcus independently generates the analytical conclusions, the assessment of how the law applies to the specific facts, and the identification of any strategic implications. These elements represent his own analytical work produced exclusively for attorney review.
Step seven: Submit for attorney review. The draft research memo is submitted to the supervising attorney explicitly as a paralegal draft requiring attorney review and approval before use. Marcus flags any aspects of the research where he identified uncertainty or tension in the case law, and any questions that he believes the attorney should assess before relying on the research in a filing or client communication.
Workflow Three: Preparing a Routine Motion
Routine motions, including motions for extension of time, motions for adjournment of hearings, and motions addressing procedural matters, follow a relatively predictable structure and are well suited to AI-assisted drafting from a template and case-specific facts. The workflow described here addresses the preparation of these motions, not the preparation of substantive motions arguing contested legal questions, which require a level of legal judgment and strategic assessment that places them outside the scope of AI-assisted paralegal work.
Step one: Receive the instruction from the supervising attorney. The attorney requests the motion, specifying the type of motion, the relief sought, the grounds for the motion, and the filing deadline. Marcus confirms his understanding of the instruction before beginning work, particularly confirming the specific relief to be requested and any procedural requirements the motion must address.
Step two: Locate precedent motions in the firm's document library. Using the firm's document management system's AI search features, Marcus locates previous motions of the same type filed in matters within the firm. He identifies the most recent and most relevant precedents, paying attention to the jurisdiction and court in which they were filed, the structure and content of the supporting argument, and any local rules compliance features that the precedent motions address. Two or three well-chosen precedents provide a sound structural template for the new motion.
Step three: Draft the motion with AI assistance. Marcus submits the relevant precedent motions, the case summary document, and a brief description of the specific facts supporting the current motion to Gemini. His prompt requests a draft of the motion integrating the case-specific facts into the established structure of the precedent motions. The prompt explicitly specifies the court of filing, the parties' names, their procedural designations, the specific relief requested, and the grounds for the motion. The Gemini-generated draft adopts the structural pattern of the precedents while incorporating the case-specific material.
Step four: Verify all case-specific details. Every case-specific element of the draft motion is verified against the source documents before the motion is presented to the supervising attorney. This includes the case number and caption, the parties' names and their correct designations, all dates and deadlines referenced in the motion, any factual assertions about the case history, and all procedural references. A single error in the case number, a party's name, or a date in a filed court document can require a corrective filing and potentially affects the procedural record of the matter.
Step five: Check applicable court rules. The draft motion is reviewed against the applicable procedural rules and local rules of the court in which it will be filed. Page limits, font specifications, margin requirements, certificate of service requirements, and any specific content requirements for the type of motion being filed are all verified before the motion is presented for attorney review. Local court rules change, and the version of the rules checked should be the current version rather than one inherited from a previous matter's research.
Step six: Submit for attorney review, approval, and filing. The verified draft is submitted to the supervising attorney with a cover note identifying the motion, the filing deadline, and any aspects of the draft that Marcus has flagged for attorney attention. The attorney reviews, edits, approves, and affects the e-filing. Marcus does not file court documents independently.
Quality Control Checklist
Marcus applies a quality control checklist to all AI-assisted work before it is presented to the supervising attorney. The checklist reflects the specific accuracy and professional standards requirements of litigation paralegal work.
Has every case citation been independently verified in a primary source legal database? This check is absolute and without exception. No AI-generated or AI-assisted case citation is included in any work product without independent primary source verification. Verification means confirming the case's existence, its citation, its holding, and its current status in the database. A citation that cannot be verified in a primary source is removed from the work product, regardless of how plausible it appears.
Do AI summary facts match the source document exactly? Discovery review summaries and document analyses are compared against the source documents for accuracy. Where a summary characterises a document as saying something it does not actually say, or attributes a statement to a party or witness who did not make it, the summary is corrected before it enters the discovery master index. Errors in the documentary record of a litigation matter have direct consequences for case strategy and court submissions.
Does any AI-produced work product contain potentially privileged information in a context where privilege has not been assessed? Any document or communication that might be subject to attorney-client privilege or work product protection requires privilege assessment before it is produced to opposing counsel or shared outside the attorney-client relationship. AI tools processing case documents do not perform privilege assessment. That assessment is Marcus's responsibility, subject to attorney review for any borderline determination.
Are all dates, deadlines, case numbers, and party names accurate? Procedural accuracy in litigation is not merely a quality standard. It is a compliance requirement. Every date, every deadline, every case number, and every party name in a court filing or an external communication is verified against the authoritative source before presentation to the supervising attorney.
Would the supervising attorney be comfortable submitting this work product under their professional signature? This is a holistic quality test that applies the standard of attorney professional responsibility to paralegal work products. If Marcus would hesitate to present a piece of work to the supervising attorney because he is uncertain of its accuracy or quality, that hesitation is the signal that the work is not yet ready for attorney review. Work is not presented until Marcus is satisfied that it meets the standard his supervising attorneys expect.
The After State
After twelve weeks of consistent AI practice, the most significant change in Marcus's work is the time recovered from discovery document review. AI-assisted document review, applied to a large production, allows him to produce a complete initial review with a populated discovery master index in a fraction of the time that manual review of the same production would have required. The time recovered has been invested in more thorough attorney briefing on discovery findings and in the legal research tasks that were previously compressed by the time demands of manual document review.
Legal research has changed in character more than in speed. The time saving in the research workflow is real but less dramatic than in document review, because the primary source verification step, which is non-negotiable, still requires substantial time. The change is that the synthesis step, which previously required writing from scratch based on reading notes, now benefits from an AI-produced initial synthesis that Marcus verifies, corrects, and enhances rather than producing entirely from scratch. The quality of the finished research memos has improved, because the AI synthesis surfaces connections between authorities that Marcus might not have identified in a manual synthesis under time pressure.
The supervision and approval requirements of the firm remain unchanged and unchanged by design. All client communications remain subject to attorney review before sending. All court filings remain subject to attorney review before filing. All research memos are presented as paralegal drafts for attorney approval before use. The AI practice has improved the efficiency and quality of Marcus's contribution to these supervised outputs, not the nature of the supervision itself.
Common Mistakes for Paralegals
Using AI-generated citations without primary source verification. The hallucination risk in AI-assisted legal research is not theoretical and not marginal. AI tools trained on general text data can produce citations that look entirely plausible: a credible court name, a plausible volume and reporter, a realistic year, and a case name that sounds genuine. These citations, when searched in primary source databases, do not exist. In a legal filing, a fabricated citation is a professional and disciplinary risk for the supervising attorney and potentially for the firm. There is no threshold of plausibility at which an AI-generated citation may be treated as verified. Every citation requires primary source confirmation without exception.
Processing privileged communications through external AI tools without attorney guidance. The privilege implications of submitting attorney-client communications or work product to an external AI tool are a matter of ongoing legal development in most jurisdictions, as addressed in Section 5 of Module 4.3. In the absence of specific guidance from the supervising attorney or the firm's professional responsibility counsel on how AI tool use interacts with privilege, the conservative position is to treat all potentially privileged communications as excluded from external AI processing unless specific clearance has been obtained. The cost of unnecessary caution is a slightly less efficient AI practice. The cost of inadvertent privilege waiver in a litigation context can include adverse consequences for the client's case and professional consequences for the supervising attorney.
Allowing AI to formulate legal conclusions. Legal analysis, including the identification of how a legal standard applies to a set of facts and the strategic assessment of the strength of a legal position, is attorney work that falls outside the scope of paralegal practice in most jurisdictions. An AI tool will readily produce what reads as legal analysis, including assessments of the strength of a legal argument, recommendations about litigation strategy, and conclusions about how a court would rule on a specific issue. These outputs should not be included in research memos, client communications, or any other work product as though they were verified legal analysis. They are AI-generated texts that require attorney review and judgment before any reliance can be placed on it. The paralegal's role in AI-assisted legal work is to research, organise, and draft for attorney review, not to incorporate AI legal conclusions into professional work products.
Assuming AI understands jurisdiction-specific procedural rules. Litigation practice is governed by a complex web of procedural rules that vary between jurisdictions, between court levels within a jurisdiction, and between courts at the same level with different local rules. AI tools trained on general legal text may have a general understanding of procedural concepts but will not reliably know the specific current local rules of a particular court, the procedural preferences of a particular judge, or the recent amendments to a jurisdiction's procedural code. Any procedural assertion in AI-assisted work product, including the applicable deadlines, the format requirements for a filing, and the procedural prerequisites for a specific type of motion, must be independently verified against the current primary sources rather than accepted on the basis of the AI's output.