How to Find Contradictions in Depositions Using AI

Why Do Depositions Contain Contradictions?

Depositions frequently contain contradictions because witnesses testify over extended periods, across multiple sessions, and often months or years after the events in question. A witness's account of a key meeting may shift between their first deposition and their second. Their sworn testimony may conflict with emails, internal memos, or prior interrogatory responses.

These inconsistencies are not always intentional -- memory fades, perspectives change, and complex facts create genuine confusion. But they are critical for impeachment, settlement leverage, and trial preparation.

Under the Federal Rules of Evidence, Rule 801(d)(1), prior inconsistent statements made under oath can be admitted as substantive evidence. That makes systematic contradiction detection a direct path to admissible impeachment material.

The challenge is finding these contradictions. In a case with 10 depositions averaging 200 pages each, plus thousands of supporting documents, the volume of cross-referencing is enormous.

Traditional methods require attorneys to read every transcript, manually flag key statements, and compare those statements against other testimony and documentary evidence. This process can take weeks and still miss inconsistencies buried in the volume.

The Sedona Conference has published cooperation guidance recommending proportional, technology-assisted approaches to exactly this kind of cross-document analysis.

How Do Lawyers Traditionally Find Deposition Contradictions?

Traditional contradiction analysis relies on manual cross-referencing by experienced attorneys and paralegals. The standard approach: read each deposition transcript, create a statement index organized by topic and witness, then compare entries across witnesses and against documentary evidence.

Some firms use keyword searches in litigation support databases to find matching topics across transcripts. But keyword search has a basic limitation: it only finds exact word matches, not conceptual contradictions.

A witness who says “I was not present at the board meeting” in one deposition and “I attended the quarterly governance session” in another is contradicting themselves. But a keyword search for “board meeting” will only find the first reference. The second statement uses entirely different language for the same event.

Finding this type of semantic contradiction requires understanding context, meaning, and relationships between statements -- exactly what AI-powered semantic analysis does.

The manual approach also suffers from scale limitations. An attorney can realistically cross-reference two or three depositions with close attention. But in multi-party litigation involving 15 to 20 deponents, the number of pairwise comparisons grows exponentially. Most firms simply lack the time and budget to conduct exhaustive cross-referencing, which means contradictions go undiscovered.

How Does AI Contradiction Detection Work?

AI contradiction detection performs semantic comparison across every statement in your document set at once. Unlike keyword search, the AI understands what each statement means and can identify conflicts even when witnesses use completely different language to describe the same events.

The system compares every deposition statement against every other deposition, plus the full body of documentary evidence -- emails, contracts, reports, and internal communications.

The AI processes contradiction detection in three stages. First, it extracts and indexes factual claims from each document -- statements about who did what, when, where, and why. Second, it compares these claims across the entire corpus, identifying pairs of statements that assert incompatible facts.

Third, it scores each contradiction by severity (direct conflict vs. tension vs. omission) and provides full citations to both source documents so attorneys can immediately verify the finding.

This approach catches categories of contradictions that manual review routinely misses:

• Cross-witness contradictions: Two different witnesses give conflicting accounts of the same event
• Self-contradictions across sessions: The same witness makes inconsistent statements in different depositions or at different points in the same deposition
• Testimony vs. documentary evidence: A witness's sworn statement conflicts with what emails, meeting minutes, or contracts actually show
• Timeline inconsistencies: Statements that are individually plausible but create impossible sequences when combined
• Omission-based contradictions: A witness claims no knowledge of a topic that their own emails demonstrate they discussed extensively

What Is the Workflow for Finding Contradictions with DiscoverLex?

DiscoverLex provides a streamlined workflow for deposition contradiction analysis that takes litigation teams from raw transcripts to actionable findings in hours rather than weeks. The process involves four steps.

Step 1: Upload Depositions and Supporting Documents

Upload deposition transcripts in any format -- PDF, DOCX, TXT, or scanned images. DiscoverLex's multi-engine OCR processes scanned transcripts and converts them to searchable text.

Upload supporting documentary evidence alongside the depositions so the AI can cross-reference testimony against emails, contracts, and other records. The platform handles 40+ file formats and processes documents in parallel.

Step 2: AI Analyzes and Indexes All Statements

The platform's 2-pass AI verification process extracts every factual claim, identifies the speaker and context, and indexes the full corpus. The first pass categorizes statements by topic, entity, and chronology. The second pass performs cross-document comparison, mapping relationships between witnesses, organizations, and events across the entire document set.

Step 3: Review Flagged Contradictions

DiscoverLex presents contradictions in a prioritized list, scored by severity and relevance. Each flagged contradiction includes the full text of both conflicting statements, direct links to the source documents and page numbers, the identity of each speaker, and a confidence score for the strength of the conflict.

Attorneys can filter contradictions by witness, topic, severity level, or document type to focus on the most trial-relevant inconsistencies first.

Step 4: Deep-Dive with Targeted Queries

After reviewing flagged contradictions, attorneys can run targeted queries to explore specific areas. For example: “What did [witness name] say about the March 2024 board meeting across all depositions and emails?”

The AI pulls together every relevant statement with full citations, making it straightforward to build an impeachment outline. The deep-dive analysis feature supports natural-language questions and returns multi-document synthesis with source attribution.

What Kinds of Contradictions Does AI Find in Practice?

In real litigation, AI contradiction detection uncovers inconsistencies that would be extremely difficult to find by hand. Consider a breach-of-contract dispute where a former CEO testifies that they “had no involvement in negotiating the partnership terms.”

The AI cross-references this against 50,000 emails and surfaces a message chain where the same CEO sent detailed markup comments on the partnership agreement and wrote “I want final approval on all material terms.” The contradiction is flagged with direct citations to the deposition transcript (page and line number) and the email (date, sender, recipient, and quoted text).

In a securities fraud investigation, an executive testifies that revenue projections were “developed by the finance team without my input.” AI analysis surfaces internal Slack messages where the same executive told the finance team to “adjust Q3 numbers upward to match what we promised investors.”

Without AI, finding this connection would require an attorney to read the deposition transcript, independently locate the Slack message in a database of thousands of communications, and recognize the semantic link between the two.

Timeline contradictions are another common finding. A witness claims they “first learned about the contamination issue in September 2024.” AI analysis of their email history shows they received a detailed environmental report about the same contamination in June 2024 and forwarded it to three colleagues with the note “we need to discuss this immediately.”

The three-month discrepancy is flagged automatically.

Why Does AI Outperform Manual Contradiction Analysis?

AI outperforms manual contradiction analysis for three reasons: scale, semantic understanding, and exhaustiveness.

A human attorney can hold perhaps 50 to 100 key facts in working memory while reviewing a transcript. An AI system compares every statement against every other statement in the entire corpus at once -- thousands or millions of pairwise comparisons that no human team could replicate.

The semantic advantage matters just as much. Manual cross-referencing depends on the attorney recognizing that two differently-worded statements refer to the same event. AI understands synonyms, paraphrases, and contextual references natively. When one witness says “the acquisition closed” and another says “the deal was finalized,” the AI recognizes these as references to the same event and compares the associated facts.

AI analysis is also exhaustive in a way that manual review cannot be. Budget and time constraints force manual reviewers to focus on the depositions they believe matter most, inevitably missing contradictions in lower-priority materials.

AI analyzes everything -- including documents a human reviewer would have deprioritized. Some of the most damaging contradictions turn up in peripheral communications that no one thought to cross-reference. Learn how AI review compares to manual review across cost, speed, and accuracy.