How AI Is Changing Knowledge Work for Lawyers

The Problem AI Is Solving (and Creating) in Legal Practice

A first-year associate is asked to research whether a specific contract clause is enforceable in five states. Traditionally: 8-10 hours of Westlaw research, jurisdiction by jurisdiction. With AI legal research tools in 2025: a structured query produces a state-by-state analysis in 15 minutes — which the attorney then verifies and refines.

AI knowledge work for lawyers is the use of AI tools — particularly large language models trained on legal data — to accelerate the research, drafting, review, and synthesis tasks that have traditionally consumed significant attorney time. The tools are real and in wide use; so are the risks. The 2023 Mata v. Avianca case, where attorneys submitted AI-hallucinated case citations to a federal court and faced sanctions, demonstrated the consequences of using AI without adequate verification.

This article covers what's actually useful, what actually works, what the professional responsibility obligations are, and how to build an AI-augmented legal workflow that captures the efficiency gains without the professional exposure.

---

The AI Tasks That Are Genuinely Useful for Legal Work

Legal Research Assistance

What AI does well:

• Initial case law summaries: AI can summarize the holdings and key facts of large numbers of cases in seconds, helping attorneys quickly understand the landscape
• Cross-jurisdictional comparisons: state-by-state analysis of how a doctrine applies across jurisdictions, which traditionally required parallel research in each jurisdiction
• Finding analogous cases: surfacing cases with similar fact patterns the attorney hadn't thought to search for
• Secondary source summaries: summarizing treatise chapters and law review articles

What AI does poorly:

• Accuracy: AI tools hallucinate cases that don't exist. Mata v. Avianca (S.D.N.Y. 2023) involved ChatGPT fabricating case citations, then fabricating summaries of the fabricated cases when asked to verify them. Every AI-generated citation must be independently verified in Westlaw or Lexis before use.
• Currency: AI training data has cutoff dates; it may not know about a statute amendment or case decided last month
• Jurisdiction-specific nuance: AI often generalizes across jurisdictions; verify jurisdiction-specific rules independently

Best tools:

• Westlaw AI Precision / Lexis AI: AI integrated with actual legal databases; citations are drawn from real, available cases. Significantly more reliable than general-purpose LLMs for legal citation.
• Harvey: Law-firm-specific AI platform built on GPT-4, accessed through firm contracts; strong on research and drafting
• CoCounsel (Thomson Reuters): AI integrated into Clio and Westlaw workflows

---

Contract Review and Analysis

What AI does well:

• Flagging non-standard clauses against a template or market standard
• Identifying missing provisions that are typically present
• Summarizing key commercial terms (payment, termination, limitation of liability)
• First-pass comparison of contract drafts (what changed between version 1 and version 2)

What AI does poorly:

• Legal judgment: whether a non-standard clause is acceptable for a specific client situation requires attorney judgment
• Novel clause types: AI trained on market-standard contracts may miss issues in unusual transaction structures

Best tools:

• Luminance, Kira Systems, Ironclad: contract AI tools specifically trained for legal review
• GPT-4/Claude for drafting prompts: useful for first drafts of standard clauses when the attorney knows exactly what they want and reviews carefully

---

Document Drafting Assistance

What AI does well:

• First drafts of standard documents: engagement letters, form contracts, standard motions (motion to dismiss, summary judgment motion in standard formats)
• Filling in templates based on matter-specific information
• Suggesting alternative language for clauses

What AI does poorly:

• Jurisdiction-specific requirements: AI doesn't always know local rules for formatting, word limits, or required provisions
• Strategy: AI doesn't know your litigation strategy; its drafting follows the prompt, not the broader litigation plan
• Accuracy of facts: AI may introduce factual errors if given incomplete or complex fact patterns

Best practice: Use AI for first drafts of standard documents; attorney must review and substantially edit before filing or sending.

---

Case Strategy and Issue Spotting

What AI does well:

• Issue spotting in fact patterns: "Given these facts, what legal claims might a plaintiff assert?"
• Summarizing analogous cases for use in brief writing
• Preparing deposition outlines from case documents

Caution: AI legal strategy advice has no professional judgment behind it. The strategy suggestions need attorney evaluation — AI doesn't know the client's risk tolerance, budget, or business goals.

---

The Professional Responsibility Framework

Competence (ABA Model Rule 1.1): The ABA's formal opinion on technology competence (Formal Opinion 477R, 2017; Formal Opinion 498, 2021) establishes that competence includes understanding technology used in legal practice. Using AI tools without understanding their limitations — particularly hallucination risk — may constitute incompetent representation.

Supervision (ABA Model Rule 5.1, 5.3): Attorneys are responsible for work produced by subordinates and non-attorney staff, including AI tools. AI output must be supervised and verified, not treated as final work product.

Candor to tribunal (ABA Model Rule 3.3): Filing false statements of law or fact — including hallucinated case citations — violates Rule 3.3. Mata v. Avianca sanctions were grounded in counsel's failure to verify AI-generated citations before filing.

Confidentiality (ABA Model Rule 1.6): Client information input into AI tools may implicate confidentiality. Many general-purpose AI tools (ChatGPT, Claude) use conversation data for training. Law-firm-specific AI deployments (Harvey, CoCounsel in enterprise configurations) typically offer confidentiality agreements.

The verification obligation: Every case citation generated by AI must be independently verified in Westlaw or Lexis before use in any court filing or formal legal advice.

---

An AI-Augmented Legal Research Workflow

Step 1 — AI for orientation (5-15 minutes): Use Westlaw AI, Harvey, or CoCounsel to ask: "Summarize the current law on [issue] in [jurisdiction]." Use the output as a map — understanding the landscape of cases and doctrines, not as a verified citation list.

Step 2 — Verify every citation (30-60 minutes): Take every case the AI referenced and verify it in Westlaw or Lexis:

• Does the case exist?
• Does the AI's characterization of the holding match the actual case?
• Is the case still good law (KeyCite or Shepard's)?

This step cannot be skipped. It's where AI efficiency gains can be captured AND malpractice risk managed.

Step 3 — Deepen on verified cases: Once you have verified, good-law cases, use traditional research techniques to expand: KeyCite citing references, key number searches, secondary source follow-up.

Step 4 — AI for drafting (15-30 minutes): Use AI to draft the analysis section of the memo or the argument section of the brief, based on the verified research. Prompt with your verified cases and the argument structure you want.

Step 5 — Attorney review and revision: Review the AI draft carefully for:

• Factual accuracy (AI may mischaracterize your cases or facts)
• Jurisdiction-specific requirements
• Strategic alignment with the client's position
• Tone and formality

Revise substantially. The AI draft is a starting point, not a finished product.

---

A Worked Example

A solo employment attorney, Diego, is handling a wrongful termination case in California. He uses an AI-augmented workflow:

AI orientation: Diego asks Harvey: "Summarize California law on wrongful termination in violation of public policy and the current Tameny doctrine."

Harvey produces a 300-word summary with references to Tameny v. Atlantic Richfield Co. (Cal. 1980), Green v. Ralee Engineering (Cal. 1998), and several recent Court of Appeal decisions.

Verification: Diego verifies all four cases in Westlaw. All exist, all are good law, Harvey's characterizations are accurate. Diego also finds a 2024 California Court of Appeal decision on Tameny that Harvey didn't include — newer than Harvey's training data cutoff.

AI drafting: Diego prompts Harvey to draft an "Introduction and Background" section for his opposition to summary judgment, based on the verified cases and his fact pattern. Harvey produces a 600-word draft.

Attorney review: Diego finds two issues: Harvey overstated one case's holding; Harvey didn't address opposing counsel's specific argument. Diego revises both.

Outcome: Diego's research and first draft took 3 hours rather than his prior 8. He spent the saved time on client strategy.

---

Tools for AI-Augmented Legal Knowledge Work

Tool	Use case	Notes
Westlaw AI Precision	Legal research (citations from real database)	Most reliable for citations; integrated KeyCite
Lexis AI	Legal research (citations from real database)	Equivalent Lexis offering
Harvey	Research, drafting, document review	Enterprise contract; strong for law firms
CoCounsel	Westlaw/Clio integrated research and drafting	Good for solo and small firm
Kira / Luminance	Contract review and extraction	Specialized contract AI
ChatGPT / Claude	General drafting assistance	NEVER for citations without verification; good for general drafting
WebSnips	Regulatory web content capture	Clip agency guidance and legal news beyond AI's training cutoff

WebSnips and AI in legal practice: AI training data has cutoff dates — it doesn't know about regulatory guidance published last month, recent agency rule changes, or current enforcement priorities. WebSnips clips current web-published regulatory content (agency websites, government announcements) and organizes it by practice area. For attorneys doing AI-augmented research on regulatory matters, WebSnips fills the recency gap: the AI provides the doctrinal foundation; WebSnips provides current regulatory updates.

---

Common AI Mistakes in Legal Practice

Mistake 1: Treating AI citations as verified. They're not. AI hallucinates. Mata v. Avianca sanctions. Every citation gets independently verified. Non-negotiable.

Mistake 2: Inputting confidential client information into general-purpose AI. ChatGPT, Claude (consumer version), and similar general-purpose tools may use conversation data for training. Confidential client information should only go into AI systems with executed confidentiality agreements covering legal use.

Mistake 3: Using AI for jurisdiction-specific procedural rules. Local rules, filing requirements, and court-specific procedures change. AI may not know current rules. Always verify procedural requirements against current court rules.

Mistake 4: Billing AI time uncritically. Billing clients for AI time that wasn't actually spent is problematic. Billing for attorney time spent supervising and verifying AI output is appropriate; billing for the time that AI replaced (while pocketing the efficiency gain) raises billing ethics questions under ABA Formal Opinion 93-379.

---

Key Takeaways

1. AI knowledge work for lawyers is genuine and growing: research orientation, contract review, and first-draft generation are all meaningfully accelerated by AI tools in 2025.
2. Verification is non-negotiable: every AI-generated case citation must be independently verified in Westlaw or Lexis. Hallucinated citations in court filings result in sanctions.
3. Use legal-specific AI tools (Westlaw AI, Lexis AI, Harvey, CoCounsel) for citations rather than general-purpose LLMs — legal-specific tools draw from real databases.
4. Professional responsibility framework: competence (Rule 1.1), supervision (Rules 5.1, 5.3), candor to tribunal (Rule 3.3), and confidentiality (Rule 1.6) all apply to AI use.
5. Confidentiality caution: don't input client-specific confidential information into general-purpose AI tools without executed confidentiality agreements.
6. AI fills the orientation and drafting role — not the judgment role. The attorney's verification, strategy, and revision remain essential.

---

Conclusion

AI knowledge work for lawyers is no longer a future possibility — it's current practice, with real efficiency gains and real professional risks. The attorneys who will capture the most value from AI are those who integrate it into a structured workflow: use AI for orientation and first drafts, verify independently, revise substantially, and maintain the professional judgment that no AI tool can substitute. The risks — hallucination, confidentiality exposure, billing ethics — are manageable with appropriate process. The opportunity — time saved, scope expanded, quality improved with better research depth — is significant. The profession is adapting; attorneys who build sound AI workflows now will have a durable advantage over those who either avoid the tools entirely or use them carelessly.

Try WebSnips free — capture the current regulatory guidance, agency announcements, and legal news that AI tools miss due to training cutoffs, integrating current web content into your AI-augmented research workflow.