Deposition transcripts are long, they arrive at inconvenient times, and the critical testimony is usually buried somewhere in a hundred pages of preliminary questions and objections. This is exactly the kind of work where AI earns its keep — not by replacing the attorney's judgment, but by compressing the time between "transcript received" and "here are the passages that matter."
This guide covers the practical workflow: what to prepare before using AI on a transcript, how to run the analysis, what outputs are actually useful versus what looks impressive but needs heavy checking, and the ethics obligations that apply throughout.
What AI actually does well on depositions
The tasks where AI creates genuine time savings in deposition work are well-defined:
First-pass summarization. A general AI assistant or a legal-specific tool can produce a structured summary of a long transcript — what was discussed, in what order, the witness's core positions — in minutes. This is not the output you rely on alone, but it is a starting point that replaces a slow first read. Review the summary against the transcript, not instead of it.
Topic and issue extraction. Ask AI to identify every passage where the witness discussed a particular topic — damages, the contract, a specific date or event — and it will surface those sections across a long transcript. This is faster than ctrl+F with a keyword and broader than a single search string.
Inconsistency flagging. If you have prior testimony — an earlier deposition, an interrogatory response, a declaration — AI can compare it to the current transcript and flag passages where the witness's account diverges from prior statements. The flag is a pointer; you verify whether the inconsistency is real, material, and admissible.
Follow-up question drafting. Based on what the transcript says, AI can draft a list of questions you might have asked or follow-up questions for the next deposition. These are prompts and drafts, not a final outline — but they can surface angles the attorney might not reach under time pressure.
Deposition prep outlines. Ahead of a deposition, AI can review a witness's prior statements, declarations, or documents and draft a topical outline for examination. The attorney shapes the outline; AI compresses the raw-material review.
The workflow
Before you start
Decide what you are actually asking for. Vague instructions produce vague output. The difference between "summarize this deposition" and "identify every passage where the witness discussed the sequence of events on [date], and flag any statement that appears to contradict Exhibit 12" is the difference between a generic summary and a useful analysis.
Resolve the confidentiality question first. Deposition transcripts contain client information, witness information, and potentially privileged matter detail. Under Model Rule 1.6 and ABA Formal Opinion 512, client-identifying information should go only into a tool with terms that do not train on your inputs and that offers a data processing agreement. Consumer tiers of general AI tools — the free or personal plans — typically do not provide those protections. Before the transcript goes anywhere, confirm you are on a business or enterprise tier with appropriate data handling terms. Do not assume; verify the current terms on the tool you are using.
Warning
Running the analysis
Format matters. Most AI tools handle plain-text transcripts better than PDFs. If you have a court-reporter text file (.txt or .rtf), use that. If you have a PDF, convert it or paste the relevant sections. Many AI tools have context-window limits; a full-day transcript may need to be chunked.
Be explicit about output format. "Summarize in 3 paragraphs" produces different output than "produce a bullet-point index organized by topic, with transcript page and line references for each point." Tell the tool what you want the output to look like and how it will be used.
Run multiple targeted queries, not one omnibus request. One large "analyze everything" request produces a broad summary you will have to sift. Targeted questions — "list every time the witness was asked about the meeting on [date] and gave a substantive answer" — produce actionable output.
Ask for page and line references. A summary without citations is just prose. Instruct the AI to cite the page and line number for every specific statement it references, so you can verify against the actual transcript.
Checking the output
This is the step that many AI workflows skip and that the professional obligation requires. AI summarization is useful precisely because it is fast; it is fallible for the same reason.
Verify every cited passage. Pull up the transcript and read the full context around every page and line the AI cited. Do not read a summary of a summary — read the actual testimony. This is not optional under the duty of competence.
Check for omissions. Summaries leave things out by definition. Ask a follow-up: "What, if anything, did the witness say about [X] that was not in the summary above?" The absence of a topic from the summary does not mean the witness did not address it.
Run a targeted inconsistency check yourself. AI flags are starting points. Before relying on a claimed inconsistency in a motion or at trial, locate the passages in both documents and confirm that the apparent conflict is real, in context, and not explained by a question you did not see.
| Task | AI handles well | Verify carefully |
|---|---|---|
| First-pass summary | Yes — compresses reading time | Completeness — read the underlying transcript |
| Topic/keyword extraction | Yes — faster than manual search | Confirm omissions — ask explicit follow-ups |
| Inconsistency flagging | Yes — surfaces candidates | Every flag: read both passages in full context |
| Follow-up question drafts | Yes — useful prompts | Strategic judgment is yours |
| Prep outline | Yes — compresses raw-material review | Shape the outline before the deposition |
| Final reliance on any output | No | Everything touches the record |
The ethics line
ABA Formal Opinion 512 requires competence in the AI tool you use, which means understanding what it can and cannot do. Deposition summaries fall into a high-stakes category: anything that influences your examination of a witness, your motion practice, or your advice to a client about what the record shows carries the duty of competence under Model Rule 1.1. An AI flag does not discharge that duty — it creates the obligation to verify.
Model Rules 5.1 and 5.3 apply to how you supervise staff or other attorneys using AI for deposition work. The work product that relies on AI-generated summaries is yours; the verification standard is yours to set and enforce.
Billing: under Rule 1.5 and ABA Opinion 512, AI-generated time compression is not billable as if the attorney did the work manually. If AI produced a first-pass summary in five minutes that would have taken an hour of reading, you do not bill the hour. What you bill is the time spent reviewing, verifying, and acting on the analysis — and making the strategic judgments that required it.
What to watch for
Transcript formatting quirks. Court-reporter transcripts have consistent formatting that AI handles well. Rough notes, Zoom-recording auto-transcripts, or non-standard formatting introduce errors — the AI will attempt to parse them, but the underlying quality determines the output quality. Run a sanity check on rough transcripts before relying on any summary.
Long transcripts and context windows. A full-day deposition can exceed the context window of many AI tools. Chunk intelligently — by day, by topic, by witness — and track what was covered. Do not assume a summary of chunk one covers something that was testified to in chunk three.
Confident but wrong. AI tools produce output in confident prose. A plausible-sounding inconsistency flag that is in fact not an inconsistency — because the prior testimony was in a different context, or the AI misread a question — looks indistinguishable from a real one in the output. Read the underlying text.
The tools that apply
For general transcripts, a general AI assistant on a business tier (ChatGPT Team, Claude Pro, Gemini Workspace) handles the summarization and issue-extraction workflow described above. For transcripts that require cross-referencing with case law or other legal documents, a legal-specific tool like CoCounsel extends the analysis. The tools at each tier are mapped in the best AI tools for lawyers.
For the confidentiality framework that governs which tier is appropriate for your client material, see client confidentiality and AI tools.
Related reading: The best AI tools for lawyers | AI for legal research | Client confidentiality and AI tools