Harvey AI Legal Research Tool: Features, Capabilities, and Documented Limitations

What Harvey Is — and What It Is Not

Harvey is a generative AI platform built specifically for legal professionals, developed by Harvey AI (formerly Harvey Technologies). It is not a general-purpose chatbot repurposed for law. The product is built on fine-tuned large language models — including versions of OpenAI's GPT-4-class models — adapted for legal language, and is primarily sold to law firms and in-house legal teams on an enterprise basis.

The distinction matters when evaluating it. Harvey does not index a primary-law database the way Westlaw or Lexis does. It does not retrieve cases from a curated legal corpus by citation. What it does is assist with the drafting, analysis, and summarization tasks that sit around legal research — memo drafting, document review, contract analysis, regulatory mapping — and it can be connected to document sets via retrieval-augmented generation (RAG) pipelines.

Core Feature Set

Harvey's documented capabilities cluster into four functional areas. Each has a distinct use case and a distinct risk profile.

Legal Drafting and Memo Generation

This is the feature most consistently reported in firm deployments. Harvey can generate first-draft legal memos, correspondence, contract clauses, and regulatory summaries from prompts or uploaded documents. The output quality varies by practice area — transactional drafting tends to perform more consistently than litigation-style analysis, partly because transactional language is more formulaic and the training data is denser.

Several Am Law 100 firms have reported using Harvey for first-draft associate-level work on standard commercial agreements. The workflow typically involves a senior attorney reviewing and editing the output rather than treating it as final. That human-in-the-loop step is not optional — it is the only viable deployment model given current hallucination rates on legal specifics.

Document Review and Contract Analysis

Harvey supports uploading documents — contracts, filings, regulatory submissions — and querying them in natural language. A user can ask "What are the termination provisions in this agreement?" or "Does this contract contain a most-favored-nation clause?" and receive a structured answer with cited excerpts.

This is where Harvey's RAG architecture is most visible. The model retrieves relevant passages from the uploaded document before generating a response, which reduces (but does not eliminate) the risk of fabricated content. The limitation is context window size — very long contracts or multi-document sets can exceed what the model handles reliably in a single session.

Regulatory and Compliance Mapping

Harvey can be used to map regulatory requirements across jurisdictions — for example, identifying which provisions of a proposed transaction trigger reporting obligations under different national competition regimes. This use case is popular with in-house compliance teams dealing with multi-jurisdictional filings.

The accuracy here depends heavily on how current the model's training data is and whether the deployment includes a live regulatory corpus via retrieval. Without a connected, regularly updated regulatory database, Harvey's knowledge of recent rule changes will lag. Firms using it for compliance work should confirm what data sources underlie the specific deployment they have contracted.

Deposition and Discovery Support

Harvey has been marketed for litigation support tasks including deposition preparation (generating potential questions from transcripts or case materials) and summarizing discovery documents. This is a narrower use case than the drafting and contract review functions, and the documented adoption is correspondingly thinner. Litigators who have reported using it tend to describe it as useful for volume-reduction tasks — getting a 500-page deposition transcript down to a structured summary — rather than for substantive legal strategy.

Deployment Model and Data Handling

Supported Jurisdictions and Practice Areas

Harvey is marketed as a multi-jurisdictional tool, and its firm client list spans the US, UK, EU, and Australia. However, jurisdiction coverage is uneven in practice. The model's training data is weighted toward English-language common law jurisdictions, and performance on civil law systems (France, Germany, Spain) is reported as less consistent by practitioners in those markets.

Practice area coverage is similarly uneven. Corporate M&A, private equity, and commercial contracts are well-represented. IP litigation, family law, and criminal defense are not documented use cases in Harvey's published materials, and the tool is not positioned for those areas.

Documented Limitations and Risk Areas

• Hallucination on case citations: Harvey does not retrieve from a live primary-law database. When asked to cite supporting cases, the model can generate plausible-sounding but nonexistent citations. This is the highest-risk failure mode for litigation use. Every citation produced by Harvey must be verified against a primary-law database before use in any filing.
• Context window limitations: Long documents — multi-hundred-page contracts, large discovery sets — can exceed reliable processing capacity. The model may miss provisions in later sections of very long documents.
• Regulatory currency: Without a live-updated regulatory corpus, the model's knowledge of recent rule changes lags. Compliance teams should treat Harvey's regulatory output as a starting point requiring verification against current primary sources.
• Non-English and civil law jurisdictions: Performance degrades meaningfully outside English-language common law contexts. Practitioners in civil law jurisdictions have reported inconsistent output quality on jurisdiction-specific legal concepts.
• No independent accuracy benchmarks: As of this profile's verification date, no third-party published accuracy benchmarks for Harvey specifically exist in the public domain. Vendor-reported performance figures should be treated as unverified claims.

Pricing and Access Model

Harvey does not publish pricing. It sells on an enterprise basis, with contracts negotiated directly with law firms and in-house legal departments. Trade reporting has referenced annual contract values in the range of $50,000–$500,000+ depending on firm size and usage volume, but these figures are not confirmed by the vendor.

There is no self-serve or per-seat consumer tier. Individual attorneys cannot subscribe directly — access is through a firm-level enterprise agreement. This limits Harvey's accessibility for solo practitioners and small firms, and it means evaluation requires a formal sales process rather than a free trial.

Who This Tool Is and Is Not Suited For

Comparison Context: Where Harvey Sits in the Legal Research Tool Landscape

Harvey competes in a crowded field, but it is not always competing for the same use case. Westlaw CoCounsel and Lexis+ AI both include primary-law retrieval as a core component — they can search and cite from curated case law and statutory databases. Harvey does not offer this natively, which is a meaningful distinction for any attorney whose workflow depends on verified citations.

Where Harvey differentiates is in the breadth of its drafting and document-analysis capabilities and in its positioning as a platform that can be customized for a firm's own document corpus. Firms that want a tool that works across their internal precedent library — not just against a public legal database — have found Harvey's architecture more flexible for that use case than the database-native research tools.

Professional Responsibility Considerations

Using Harvey in client matters raises several professional responsibility questions that vary by jurisdiction. The most immediate is competence under ABA Model Rule 1.1: attorneys have an obligation to understand the tools they use and their limitations. Deploying Harvey for research tasks without understanding that it does not retrieve from a primary-law database — and without verifying its outputs — creates a material malpractice exposure.

Confidentiality under Rule 1.6 is the second concern. Uploading client documents to a cloud-hosted AI system requires the attorney to have taken reasonable steps to confirm the vendor's data handling. Harvey's enterprise agreements include confidentiality terms, but attorneys should confirm those terms are in place before uploading any client-confidential material — and should check whether their jurisdiction's bar has issued specific guidance on cloud-based AI tools.

Editorial Assessment

Harvey is a well-built tool for what it actually does: AI-assisted drafting, document analysis, and workflow acceleration in transactional and corporate legal work. The risk of overestimating its capabilities is real — particularly among attorneys who assume "legal AI" means primary-law retrieval with verified citations. It does not, and that gap matters.

For firms that have clearly scoped their use case, implemented supervision workflows, and confirmed data handling terms, Harvey represents a defensible and productive deployment. For firms that have not done that groundwork, the risks — hallucinated citations, stale regulatory output, confidentiality exposure — are not theoretical.