AI is reshaping legal research faster than most legal ethics frameworks anticipated. In 2026, attorneys have access to purpose-built legal AI tools — Harvey, Westlaw AI, Lexis+ AI, and Casetext/CoCounsel — alongside general-purpose AI assistants. These tools differ substantially in accuracy, confidentiality terms, and how they handle the hallucination risk that has already resulted in sanctions in several high-profile cases. This guide compares the leading AI legal research tools, explains the confidentiality and work product considerations that apply to each, and identifies the workflow patterns that hold up under bar scrutiny.

The hallucination problem in legal AI

Before comparing tools, it is important to understand the core risk that distinguishes legal AI research from most AI use cases: citation hallucination. General-purpose AI models like ChatGPT and Claude generate plausible-sounding text, and in legal contexts they frequently generate plausible-sounding citations that do not exist.

The cases that gave this risk national attention:

Mata v. Avianca (SDNY, 2023): Attorneys submitted ChatGPT-generated citations to non-existent cases. Sanctions were imposed. The attorney had not verified the citations.
Park v. Kim (2d Cir., 2024): An attorney cited a non-existent case in a brief. The attorney claimed AI-generated citations were real. The court sanctioned the attorney and referred the matter to the grievance committee.
Multiple state court decisions in 2024-2025 where attorneys relied on AI-generated case summaries without verification.

The legal AI tools purpose-built for legal research (Harvey, Westlaw AI, Lexis+ AI, Casetext) are designed to ground their responses in actual legal databases and cite only real cases. This is a fundamental architectural difference from general-purpose AI models — and the reason law firms cannot treat "using ChatGPT for legal research" as equivalent to "using Harvey for legal research."

The major AI legal research tools in 2026

Harvey AI

Harvey is a legal AI company founded in 2022, backed by OpenAI and Allen & Overy, and now used by major law firms globally. Harvey uses large language models (including Claude and GPT-4 variants) fine-tuned on legal data and integrated with legal databases.

What it does: Legal research, contract drafting, due diligence analysis, regulatory analysis, deposition preparation, litigation support. Harvey can analyze large document sets (discovery corpora, contract repositories) and answer natural-language questions about their contents.

Confidentiality: Harvey is designed for law firm use and includes attorney-specific confidentiality terms. Client data submitted to Harvey is not used to train Harvey's models. Harvey has executed DPAs satisfying GDPR Article 28 and offers HIPAA BAA terms for healthcare law clients. Harvey's contracts are negotiated firm-by-firm and include confidentiality, data retention, and security provisions.

Hallucination risk: Lower than general-purpose AI for database-grounded research, but not zero. Harvey citations should be verified in the underlying database before use in filed documents.

Access: Enterprise contracts with law firms; not available as a consumer product. Pricing is per-seat annually.

Westlaw AI (Thomson Reuters)

Thomson Reuters has integrated AI features throughout Westlaw — the dominant legal research platform in the US. Westlaw AI is built on top of Westlaw's existing database (which includes case law, statutes, regulations, secondary sources) and Thomson Reuters' own LLM infrastructure.

What it does: Ask Westlaw natural-language questions about legal issues; receive answers with citations grounded in Westlaw's database. Summarize cases, identify key passages, compare cases on a specific legal point, generate research memos.

Confidentiality: Westlaw is a covered service under Thomson Reuters' enterprise data processing terms. Law firm and corporate legal department customers have DPAs in place. Research queries are not used to train Westlaw AI models. Thomson Reuters has longstanding confidentiality commitments in its enterprise legal research contracts.

Hallucination risk: Low for citation purposes — Westlaw AI is grounded in Westlaw's database and will not cite cases that do not exist in Westlaw. Still verify citations before filing; rare retrieval errors are possible.

Access: Included in current Westlaw enterprise subscriptions. Available to all Westlaw enterprise customers.

Lexis+ AI (LexisNexis)

LexisNexis has integrated AI throughout its Lexis+ platform, competing directly with Westlaw AI. Lexis+ AI offers natural-language legal research, document analysis, contract review, and brief generation.

What it does: Legal research grounded in LexisNexis' case law and statute database; contract analysis; brief drafting and editing; deposition preparation.

Confidentiality: Lexis+ AI operates under LexisNexis' existing enterprise customer terms. Research queries are not used to train LexisNexis' AI models for enterprise customers. GDPR DPA available; HIPAA terms available for healthcare law customers.

Hallucination risk: Low for citation purposes — grounded in the LexisNexis database. Same verification caveat as Westlaw AI applies.

Access: Included in enterprise Lexis+ subscriptions. Available to all Lexis+ enterprise customers.

Casetext / CoCounsel (Thomson Reuters)

Casetext was acquired by Thomson Reuters in 2023 and rebranded its AI legal assistant as CoCounsel. CoCounsel is now integrated into Thomson Reuters' legal AI product family alongside Westlaw AI.

What it does: Similar capabilities to Westlaw AI — legal research, document review, contract analysis, deposition preparation, legal memo drafting. CoCounsel is particularly known for its deposition preparation and contract analysis features.

Confidentiality: Post-acquisition, CoCounsel operates under Thomson Reuters' enterprise data processing terms, consistent with Westlaw AI.

Hallucination risk: Similar to Westlaw AI — grounded in legal databases, lower risk than general-purpose AI, verification still required.

Access: Available as a standalone subscription or bundled with Westlaw.

General-purpose AI for legal research: the caveat

ChatGPT, Claude, and Gemini can assist with legal tasks — drafting, summarization, outlining arguments — but they are not legal research databases. They do not have real-time access to case law, and they hallucinate citations. The appropriate use of general-purpose AI in legal practice is for tasks that do not require citation accuracy:

Drafting argument outlines.
Summarizing a case the attorney has already read.
Explaining a legal concept to a client in plain language.
Editing and proofreading briefs.
Generating initial draft language for contracts.

For any task that requires finding real cases, statutes, or regulations, use a legal database — Westlaw, Lexis, or a purpose-built legal AI tool. Never cite a case without verifying it in a reliable legal database.

Comparing the tools: a decision matrix

| Tool | Citation accuracy | Hallucination risk | DPA available | HIPAA available | Confidentiality terms | | --- | --- | --- | --- | --- | --- | | Harvey | High (database-grounded) | Low (verify before filing) | Yes | Yes | Law-firm-specific enterprise | | Westlaw AI | High (Westlaw-grounded) | Very low | Yes | Yes | Thomson Reuters enterprise | | Lexis+ AI | High (LexisNexis-grounded) | Very low | Yes | Yes | LexisNexis enterprise | | CoCounsel | High (TR database-grounded) | Very low | Yes | Yes | Thomson Reuters enterprise | | ChatGPT Enterprise | Not applicable | High for citations | Yes | Yes | OpenAI enterprise (no legal database) | | ChatGPT Free/Plus | Not applicable | Very high for citations | No | No | Consumer terms (not appropriate for client data) |

Ethics and professional responsibility

Duty of competence (Rule 1.1)

Comment 8 to ABA Model Rule 1.1 requires lawyers to keep abreast of the benefits and risks of relevant technology. Using AI for legal research without understanding its hallucination risk is an argument for incompetence. Competent use of AI legal research tools means:

Understanding which tools are grounded in legal databases and which are not.
Verifying all citations before filing.
Understanding the confidentiality terms of the tools you use.

Duty of supervision (Rules 5.1 and 5.3)

Partners and supervising attorneys are responsible for ensuring that associates and non-lawyer staff who use AI tools do so competently and in compliance with ethical rules. Firm-wide training on AI legal research tools — what they can and cannot do, when to verify, which tools are approved — is a supervision obligation.

Duty of candor (Rule 3.3)

Filing a brief with AI-generated citations that the attorney did not verify is potentially a Rule 3.3 violation (making false statements of fact or law to the tribunal). The cases already decided make clear that courts will impose sanctions for AI hallucinations presented as real citations. Every AI-generated citation must be verified before it appears in a filed document.

Practical workflow for AI-assisted legal research

Step 1: Use purpose-built legal AI for citation research. For finding cases, statutes, and regulations relevant to a legal issue, use Westlaw AI, Lexis+ AI, Harvey, or CoCounsel. These tools are grounded in legal databases and provide citations you can verify.

Step 2: Verify every citation in the source database. Before including any citation in a filed document, verify it exists in Westlaw or Lexis, review the cited case for accurate characterization, and confirm the case is still good law (Shepardize or KeyCite).

Step 3: Use general-purpose AI for non-citation tasks. Use ChatGPT Enterprise or Claude for Work for drafting, outlining, summarizing cases you have already verified, and plain-language explanation. Do not use general-purpose AI to find new cases or verify citations.

Step 4: Apply appropriate confidentiality controls. For work involving client data, use enterprise tiers of all AI tools. Consumer accounts of any AI tool are inappropriate for client information. See AI and the Work Product Doctrine for the privilege and confidentiality analysis.

Step 5: Document your AI use. Some courts and jurisdictions are beginning to require disclosure of AI use in filed documents. Even where not required, documenting your AI research workflow creates a record of competent, supervised AI use — useful if AI-related conduct is ever questioned.

Frequently asked questions

Can I use ChatGPT to research what cases exist on a legal topic?

Not reliably. ChatGPT will generate plausible-sounding case citations that may not exist. Even when the cases it cites do exist, the characterization may be inaccurate. Use Westlaw AI or Lexis+ AI for finding cases; use ChatGPT only for tasks that do not require citation accuracy.

Harvey says it has access to legal databases. Does that mean no hallucination?

Harvey significantly reduces citation hallucination by grounding responses in legal databases. But "significantly reduced" is not zero — complex queries may still produce errors, and Harvey may retrieve and mischaracterize real cases. Verify citations before filing, regardless of which legal AI tool you use.

Does using AI for legal research require client disclosure or consent?

Most ethics opinions to date do not require client disclosure simply because AI was used in legal research, provided the attorney is competent and supervises the AI's output. Some clients, particularly sophisticated institutional clients, have AI use provisions in their engagement letters that require disclosure. Check your engagement terms and local bar guidance.

Are there any states or courts that prohibit AI use in legal research?

No US state bar has categorically prohibited AI use in legal research. Several courts have implemented disclosure requirements for AI-generated content in filed documents (the Southern District of New York, among others). Check local rules before filing in any federal or state court.

The bottom line

The AI legal research landscape in 2026 has bifurcated: purpose-built legal AI tools (Harvey, Westlaw AI, Lexis+ AI, CoCounsel) that ground responses in legal databases and maintain law-firm-appropriate confidentiality terms; and general-purpose AI (ChatGPT, Claude) that can assist with drafting and non-citation tasks but will hallucinate if asked to find cases. The competent legal professional uses both categories for the tasks they are suited for — and verifies every citation in an authoritative legal database before it appears in a filed document. The lawyers sanctioned in the hallucination cases were not incompetent; they did not understand the tool's limitations. Understanding those limitations is now part of the duty of competence.

AI Tools for Legal Research in 2026: Harvey, Westlaw AI, Lexis+ AI, and Why Citation Hallucination Still Matters