How to Use AI to Find Inconsistencies Between Documents

When documents disagree

The contract says the project starts on March 1. The statement of work says March 15. The proposal that both parties signed says "within 30 days of contract execution," which was February 10 -- making the start date March 12. Three documents, three different dates, and nobody noticed until the vendor started billing for work that hadn't begun.

Document inconsistencies are common in any organization that produces related documents. Contracts reference proposals. Amendments modify contracts. SLAs define performance standards that may not align with the master agreement. When these documents disagree -- and they often do, because they are written by different people at different times -- the inconsistencies create real problems: legal disputes, operational confusion, audit findings, and financial exposure.

Finding these inconsistencies manually requires reading every document, holding all details in working memory, and mentally cross-referencing every claim against every other claim. For five documents totaling 80 pages, this is a full day's work. For 20 documents, it is impractical without a systematic approach.

docrew provides that systematic approach. The agent reads every document in the set, cross-references claims across the entire corpus, and produces a detailed inconsistency report identifying exactly what conflicts, where it appears, and the specific values from each document.

Why inconsistencies matter

Document inconsistencies create three categories of real-world risk.

Legal risk. Conflicting terms in a contract and its exhibits become a matter of contract interpretation -- meaning litigation. A master agreement might require arbitration in New York while an exhibit specifies litigation in Delaware.

Operational and financial risk. When an SLA promises 99.9% uptime but the contract caps liability at one month's fees, resources get misallocated. When a proposal quotes $200 per hour, the contract says $225, and the purchase order authorizes $200, billing disputes consume time out of proportion to the difference.

Audit and compliance risk. Regulations require that documented policies match actual procedures. If a privacy policy says 12-month data retention but the retention schedule says 24 months, an auditor will flag the gap regardless of which timeframe is actually followed.

The cost of finding inconsistencies after they cause a problem is always higher than finding them before.

Prepare the document set

Collect all related documents into a single folder -- documents that should be consistent because they govern the same relationship, project, or process.

Common document sets for inconsistency checking:

Contract sets. Master agreement, amendments, statement of work, SLA, data processing agreements, order forms, and original proposal. These should align on dates, pricing, scope, obligations, and termination rights.

Policy and procedure sets. Corporate policy, implementing procedure, training materials, and regulatory guidance. These should align on requirements, definitions, and thresholds.

Project documentation sets. Project charter, requirements document, design specification, test plan, and acceptance criteria. These should align on scope, deliverables, timelines, and success criteria.

Financial document sets. Budget, purchase orders, invoices, and contract pricing schedule. These should align on rates, quantities, and totals.

Include every version currently in effect. If Amendment 2 supersedes Amendment 1 on certain clauses, include both so the agent can verify the reference chain.

Ask the agent to cross-reference

Open docrew and start a conversation. Drop the document set in or point the agent at the folder, then give a specific instruction.

A general prompt: "Read all documents in this folder. These are related documents for the Meridian Services engagement. Cross-reference them for any inconsistencies in dates, amounts, pricing, terms, obligations, definitions, or scope. For each inconsistency found, tell me exactly what conflicts, the specific value or language in each document, and the section or page where it appears."

A targeted prompt: "Compare the master services agreement with the statement of work and Amendment 2. Check specifically for conflicts in: start and end dates, hourly rates and fee caps, deliverable descriptions, termination notice periods, and liability caps. Also check whether the amendment properly updates all references to changed terms."

The more specific the prompt, the more thorough the results. But even a general prompt produces useful findings because the agent is systematic -- it compares every claim in every document against every related claim in every other document.

What the agent does

The agent processes the document set in four phases.

Reading and indexing. It reads every document, extracting all factual claims -- dates, amounts, names, definitions, obligations, deadlines, and conditions -- and builds an internal index tagged by topic and source.

Cross-referencing. It compares claims across documents. For every date, it checks consistency in related documents. For every dollar amount, it verifies the same figure is used wherever referenced. For every defined term and obligation, it checks alignment.

Conflict identification. When two documents state different values for the same item, it records the conflict with full context: which documents, the exact language from each, section references, and a classification of the inconsistency type.

Report generation. It produces an inconsistency report organized by severity. Critical inconsistencies (pricing conflicts, date mismatches affecting obligations) appear first. Minor inconsistencies (formatting differences, defined term variations that don't change meaning) appear later.

All processing happens locally on your machine. Documents are not uploaded anywhere.

Reading the inconsistency report

Each entry in the report typically includes:

The inconsistency. A clear description of what conflicts -- for example, "The project start date differs across three documents."

Document references. The exact language from each conflicting document with section and page numbers. For example: "Master Agreement, Section 3.1: 'Initial Term commences March 1, 2026.' Statement of Work, Section 1.2: 'Work begins March 15, 2026.'"

Classification. The type: date mismatch, pricing discrepancy, scope contradiction, definition conflict, obligation mismatch, or ambiguity.

Potential impact. A brief note on why it matters -- for example, "Billing disputes may arise if vendor begins work per the MSA but the SOW does not authorize work until two weeks later."

Not every inconsistency requires action. Some are intentional (an amendment deliberately changes a term) and some are immaterial. Focus first on critical items: pricing conflicts, date mismatches affecting obligations, scope contradictions, and liability discrepancies.

Common inconsistency patterns

Date mismatches. The most common type. Start dates, end dates, and notice periods frequently differ between master agreement, SOW, and amendments because documents are drafted at different times.

Pricing discrepancies. The proposal quotes one rate, the contract specifies another, the purchase order authorizes a third. Rate escalation clauses may not be reflected in order forms.

Scope contradictions. The proposal describes a comprehensive project, but the SOW scopes it to a subset. The client expects the full scope; the vendor intends to deliver only what the SOW describes.

Definition conflicts. The master agreement defines "Confidential Information" broadly. The data processing agreement defines "Personal Data" using GDPR definitions. The overlap creates ambiguity about which protections apply.

Obligation mismatches. The contract requires $5 million in professional liability insurance. The certificate on file shows $2 million. Nobody checked the certificate against the contract.

Termination conflicts. The contract allows 90 days' notice. Amendment 1 changes it to 60 days for the first renewal term. Amendment 2 extends the term but does not address which notice period applies.

Practical scenario: vendor contract audit

A procurement director needs to audit documentation for the company's top 10 vendors before the annual renewal cycle. Each vendor has four to eight files: master agreement, SOW, amendments, SLA, insurance certificates, and pricing schedules.

She starts with the largest vendor -- seven documents, 120 pages. She tells docrew: "Cross-reference the entire set for inconsistencies. Pay special attention to pricing terms, SLA metrics versus liability caps, insurance requirements versus certificates on file, and termination notice periods across all amendments."

The agent produces 14 findings, including:

• The master agreement specifies $150/hour. Amendment 2 changed this to $165 effective January 2026. The current purchase order still references $150.
• The insurance certificate shows a $3 million limit; the contract requires $5 million. The certificate expired two months ago.
• Amendment 1 changes the termination notice from 90 to 60 days. Amendment 3 extends the term but references the original "90-day notice period" -- unclear whether this was intentional or an error.

Each finding represents real exposure. The expired insurance alone is a significant risk gap. The director repeats the process for the remaining nine vendors. The full audit takes one day instead of two weeks.

Tips for effective inconsistency checking

Include all related documents. Missing one amendment can make an inconsistency invisible.

Be specific about what to check. A general cross-reference catches a lot, but targeted checks catch more. If pricing is sensitive, say so.

Check for omissions too. Sometimes the problem is not a conflict but a stale reference -- an amendment changes Section 5 but does not update the cross-reference in Section 12.

Run the check before signing. The most valuable time to find inconsistencies is before execution, when they can be fixed without negotiation.

Use the output as a negotiation tool. An inconsistency report is factual: here is what Document A states, here is what Document B states, they do not agree. This is more effective than a vague assertion that "the terms don't match."

From one-time check to standard practice

Most organizations run inconsistency checks reactively -- after a dispute, after an auditor flags a problem. A better approach is to build checking into standard workflows: before every signing, before every renewal cycle, after every major policy update.

With docrew, each check is a conversation. Drop the documents in, ask the agent to cross-reference, review the findings. For five documents, results come in minutes. For a portfolio of 50 vendor contracts, the process scales to a day of review instead of a month of manual work.

The inconsistencies are already in your documents. The only question is whether you find them before they cause a problem or after.