The End of the Highlighter? What AI-Powered Transcript Review Looks Like in 2025
The Death of the Yellow Highlighter
For decades, the ritual was the same: print the deposition transcript, grab a stack of yellow highlighters, and start combing line-by-line for the few exchanges that actually mattered for your case. Reviewing depositions routinely consumed days, sometimes entire weekends. In a big case, dozens of witnesses meant weeks of manual markup.
“Read the deps. I don’t care if you read them last year. Read them again.”
James Mason in The Verdict (1982)
That muscle-memory is disappearing in 2025. AI transcript tools can ingest hundreds of pages in seconds, flag every mention of a disputed fact, and serve up a clean summary before you finish your cup of coffee.
The shift is so dramatic that litigators around the country are asking a new question: If the machine can read for me, what should I be doing with my time?
The purpose of this post is not to dissect the neural networks underneath the hood. Your clients don’t care, and frankly neither do most lawyers. Rather, we’ll look at how these tools collapse time, sharpen strategy, and—importantly—how to use them without violating core ethical duties.
Deposition Review at Machine Speed
Early adopters report that a first-pass review that once ran four to six hours per witness can be front-loaded by AI in well under sixty minutes. Tools such as CaseChat from esumry and Transcript Genius from Steno read the text, answer questions, generate summaries, and include links to the page-line in the transcript. esumry’s FAQ boasts that a 400-page transcript processes “in several minutes,” with smaller transcripts completing in under one minute. Steno markets the same transformation as “hours, not days.”
What changes for the lawyer? The first pass is no longer a slog through thousands of lines of verbatim Q&A but a curated highlight reel, ready for analysis. Litigators move from doing the time-consuming work of reading and annotating to strategizing with the results much faster.
The tedium is gone. The analysis and judgment about the evidence remains all yours.
Spend More Time on Strategy, Not Summaries
Once the transcript is processed, modern systems don’t stop. They draft. In esumry, CaseChat identifies admissions and contradictions, extracts names, places and dates, and functions as an assistant to perform tasks that you assign. Plus, it delivers a narrative summary suitable for client reporting or quickly getting a newly assigned team member up to speed.
Lexitas’ new Deposition Insights+ promises customized summaries that extract “key admissions, contradictions, and exhibit references.” While not strictly a litigation software product, Thomson Reuters’ CoCounsel Core offers comparable “skills” that deliver initial outlines lawyers can refine.
Across platforms, practitioners report receiving a clean, structured summary in a few minutes—about a 70 percent head start on the dreaded “depo memo.” Think of it as work from a junior associate or paralegal who works at hyper speed but still needs your review and edits.
That head start is more than convenience: it frees up precious mental bandwidth to perform high level work such as developing dispositive motion strategy, refining settlement position statements, and planning witness sequencing if the case goes to trial.
Here’s a comparison of key features of several leading platforms you can use for transcript analysis:
Source: Vendor websites (surveyed 7/17/2025)
Finding Needles: Inconsistency Detection Across Depositions
Cross-exam prep once required a spreadsheet and a lot of caffeine. Now, AI can compare every answer a witness has ever given—across cases, across firms—and flag contradictions instantly.
One of AI’s superpowers is pattern recognition. Feed it a bundle of transcripts and it will happily cross-reference every answer, looking for conflicts. Steno’s Transcript Genius flags when a witness says “I never reviewed the invoice” but also said “I read it last year” in the same deposition. esumry and CoCounsel offer similar cross-deposition comparison, while Lexitas puts the feature front-and-center in Deposition Insights+.
Trial teams can even receive real-time “contradictions” reports—while the witness is still testifying—using Deposition CoPilot from Filevine. That just-in-time awareness can change the tone of follow-up questions and, potentially, the value of the case. We believe future AI holds the promise of being able to provide that information in real time when multiple depositions are being taken simultaneously.
Impeachment Preparation and Finding Testimony on the Fly
Some tools like CaseChat from esumry and Steno’s Transcript Genius can generate an “Impeachment Outline” of inconsistent statements and even suggest the order you might consider using them on cross. Added bonus: you can do the same thing with your own witnesses to identify areas to review when preparing them to testify at trial.
How many times have you flipped through binders of deposition testimony at trial or in a hearing to find the critical line of testimony, silently cursing yourself for not tabbing it during prep? In just seconds, these modern platforms can retrieve it for you (e.g., “Find the testimony where this witness said Topic X?”), along with page-line citations ready to be read into the record. For the seasoned trial lawyer this means less time paging through printouts under pressure—and more time listening to the witness, adjusting tactics on the fly.
Human Oversight Still Matters—And So Do Ethics
Automation does not equal abdication. Even the best large language models (LLMs) can hallucinate, oversimplify nuance, or miss sarcasm. Under ABA Formal Opinion 512, lawyers “must fully consider their applicable ethical obligations” when employing generative AI, including the duty of competence, candor to the tribunal, and—crucially—oversight of the tool’s output.
ABA Formal Op. 512 requires a qualified lawyer to supervise AI output for accuracy and completeness before relying on it in any filing or argument. (July 29, 2024)
In practical terms, that means AI is your first drafter—not your final reviewer. Treat the AI draft like a new associate’s memo: trust but verify. You still check citations, review context, and confirm that machine-generated language hasn’t overstated a fact. Shortcuts end where professional responsibility begins.
Want a deeper dive about the ethical rules? Check out our previous post, which includes a link to our state-by-state analysis.
Confidentiality & the Key to Protecting It
Why privacy is different in litigation. Depositions teem with confidential and private information, whether it is personal health information, sensitive financial data, or trade secrets. Model Rule 1.6 bars disclosure without informed consent, and ABA Formal Op. 512 places the burden squarely on the lawyer to safeguard data. Unfortunately, consumer grade LLMs such as OpenAI’s ChatGPT and Anthropic’s Claude do not provide those safeguards.
ZDR is the new gold standard. Under zero-data-retention (ZDR) models, prompts and outputs vanish after processing so nothing lingers on the LLM servers to subpoena or hack or find its way into another user’s activity. For example, OpenAI and Anthropic offer a ZDR endpoint for commercial customers, now used by many legal tech vendors, including esumry. esumry routes every user prompt or request through that ZDR pipeline. Steno’s agreement with Anthropic provides “zero-day retention.” CoCounsel touts private, dedicated servers and a zero-retention API for client content.
What about new entrants? Lexitas’ brand-new Deposition Insights+ encrypts data and deletes customer content within ninety days of contract termination—fine for many matters but not technically ZDR. This may change, so be sure to check if you decide to try it out. Lawyers needing absolute privacy should negotiate custom deletion windows before using these services if it is not addressed in their Terms of Service.
Vendor due-diligence checklist before you hit “upload.”
Confirm ZDR or a contract-backed deletion window.
Verify data encryption in transit and at rest.
Ensure data is not used to retrain language models.
Request SOC 2 reports or similar audits.
If a platform cannot meet those standards, your ethical duty is to redact or abstain from using it. ABA Formal Op. 512 makes clear that it is the lawyer’s responsibility to undertake reasonable efforts to ensure the protection of client confidences when using these generative AI products. It is not the vendor’s.
Source: Vendor websites (surveyed 7/17/2025)
What It Means for Litigation Teams
For attorneys. Less time on mechanical tasks means more quality time for strategy: case themes, jury presentation, filling evidence gaps, admissibility issues, handling “bad facts”, settlement negotiations, and more.
For paralegals and litigation support. Roles will evolve from manual markup to AI wrangling: managing transcripts, refining prompts, verifying output, and integrating exhibits. Far from replacing support staff, the technology elevates them into analysts and quality controllers.
For firms. Faster deliverables translate to competitive fees and happier clients. Early adopters also report more accurate work product and reduced risk of missed issues. This may translate into an underwriting benefit for professional-liability carriers watching the AI space.
Clients. AI powered tools will help law firms deliver faster answers and clearer budgets. Speed rarely hurts leverage in settlement talks.
The New Standard of Care?
A decade ago, refusing to use email was malpractice waiting to happen. We may be reaching the same inflection point with AI transcript tools. The question is no longer “Will they replace me?” but “Will I fall behind if I don’t learn to use them?”
Highlighters won’t vanish entirely—sometimes a witness says something so outrageous it deserves bright yellow. But the days of spending your Saturdays coloring hundreds of lines of testimony are ending.
The winners will be teams that pair machine speed with human judgment, that demand iron-clad privacy from their vendors, and that keep one eye on evolving ethics guidance. The losers will be those who cling to the yellow marker because “that’s how we’ve always done it.”
Which side of that line will your practice land on?
Want More?
Download the companion checklist “AI Deposition Tools: 10 Questions to Ask Your Vendor” or reach out if you’d like a demo of how esumry handles ZDR in your matters.
Author’s note: This post is current as of July 17, 2025. New tools and ethics opinions are emerging rapidly. Always confirm the latest requirements in your jurisdiction before onboarding a platform.
With tools like esumry, deposition transcript review is fast and strategic. Tag testimony, assess credibility, and get ahead of how the other side will use the record—before they do.
