AI for due diligence: review contracts 10× faster

AI for due diligence is no longer a promise: it’s the legal application with the strongest published data. Reviewing contracts 10 times faster is, in fact, a conservative figure compared with the available studies. Here’s the method for doing it without sacrificing rigor.

# What can AI do in due diligence today?

Today, AI can read an entire data room, extract key clauses, flag deviations from your checklist, and prepare a preliminary risk matrix. Specifically: changes of control, exclusivity, non-compete clauses, early termination, liquidated damages, and consents required to close the transaction.

What it cannot do: decide which risks are acceptable to your client, negotiate a price adjustment, or sign the report. AI produces the analysis draft; the lawyer provides the judgment. Clio estimates that up to 74% of hourly billable tasks, such as gathering and analyzing information, can be automated, according to its Legal Trends Report 2024. Document due diligence falls squarely into that category.

Consider the scale of the problem: in a mid-sized transaction, a data room can contain more than a thousand documents. With traditional review, the firm samples: it reviews the material contracts and assumes the risk for the rest. With AI, reviewing everything is no longer economically impossible, and sampling becomes a strategic decision, not a budgetary compromise.

# How fast and accurate is AI compared with a lawyer?

More accurate on average and far faster for narrowly defined review tasks. The LawGeex study, reported by the World Economic Forum, pitted 20 corporate lawyers against an AI trained to review NDAs:

Two fair takeaways. First: the best lawyer matched AI for accuracy; the decisive difference was time. Second: the study dates to 2018, covered a single type of contract, and used a specialized tool. Today’s models are more general and powerful, but every firm should validate them using its own documents.

# How do you build a due diligence workflow with Claude?

The workflow has five steps, and the fifth is not optional. Here’s how we implement it with law firms:

• 1. Custom checklist: turn your usual review list into explicit instructions for Claude, clause by clause.
• 2. Controlled upload: data room documents are handled in an enterprise environment configured for confidentiality, never through personal accounts.
• 3. Extraction: Claude returns relevant clauses with verbatim citations and their exact locations in the source document.
• 4. Risk matrix: findings are classified by severity against the transaction’s standards.
• 5. Verification: a lawyer confirms each critical finding against the original document before reporting it.

The golden rule: Claude flags; the lawyer confirms. Every finding that reaches the client has been reviewed by a person. Verification isn’t the cost of this method; it’s what lets you sign the report with the same professional responsibility as always, but with full coverage of the data room.

# What are its limitations, and where does the lawyer come in?

The main limitation is that the model can be confidently wrong: it may omit a relevant clause or paraphrase it incorrectly. That’s why the workflow requires verifiable verbatim citations, not source-free summaries. The lawyer comes in at three points: designing the checklist, validating critical findings, and translating risk into a business decision.

The industry is moving in that direction: in May 2026, Thomson Reuters and Anthropic announced the integration of Claude with CoCounsel Legal, which reasons over 1,900 million documents from Westlaw and Practical Law and is already used by one million professionals in 107 countries. The emerging standard is general AI connected to citable legal sources.

# What mistakes can ruin AI-assisted due diligence?

First: asking for conclusions without requiring citations, which turns the report into unverifiable opinion. Second: uploading confidential documents to free accounts without a confidentiality agreement with the provider. Third: skipping the baseline, so you’ll never know whether you were actually faster.

And fourth, more subtle: using speed to review just as little in less time, instead of reviewing 100% of the data room. The real competitive advantage is full coverage, not just speed. A report that says “we reviewed all the contracts, and these fifteen have a change-of-control clause” is worth more to the client than one that reviewed a sample in record time.

# Frequently asked questions

# How accurate is Claude at reviewing contracts?

Published data on AI contract review shows 94% accuracy on NDAs, compared with 85% for the average lawyer (LawGeex, 2018). For your firm, the relevant figure comes from validating the results against your own contracts and checklist.

# Can I upload the data room to Claude without breaching confidentiality?

It depends on the plan. By default, Anthropic does not train its models on data from commercial products such as Claude for Work. Even so, review your confidentiality agreement with the client and use business accounts, not personal ones.

# How much time does it save on a typical transaction?

For document review, the typical timeline shrinks from weeks to days. The exact savings depend on the data room’s volume and how much verification critical findings require.

# Does it work for contracts in Spanish and Mexican law?

Yes, for reading, extracting, and comparing against your checklist in Spanish. The lawyer remains responsible for validating the legal analysis under Mexican law: the model does not replace knowledge of local law or registry practice.

# Do I need a specialized legal tool, or is Claude enough?

For extraction and building a risk matrix using your own checklist, Claude is usually enough. Specialized tools add value when you need citable case-law research, as with Claude’s integration with CoCounsel.

# Conclusion

• LawGeex/WEF: AI with 94% accuracy in 26 seconds vs. 85% and 92 minutes for the average lawyer.
• The right workflow: your own checklist, controlled uploads, extraction with citations, a risk matrix, and human verification.
• The real advantage is reviewing 100% of the data room, not just finishing sooner.
• Every critical finding is checked against the original document before it reaches the client.

If your firm handles M&A or real estate transactions, due diligence is the best entry point for AI. Learn about the Claude Legal for Law Firms, visit the lawyers’ page or contact us to assess your first pilot.