Texas Bar Journal • June 2025

Speak, Type, Win

The Texas lawyer’s guide to AI transcription.

Written by Chase Hardy

The practice of law runs on conversation—client intake calls, strategy sessions, deposition testimony—yet human recall is unreliable. When every factual nuance may matter, a lawyer should concentrate on what the client has to say, not on frantically scribbling notes.

Transcripts created by modern speech-to-text models solve the problem with remarkable accuracy at a moment when Texas courts face an acute shortage of human stenographers.¹ At the same time, cost barriers have collapsed. ElevenLabs—an emerging leader in generative audio and speech-to-text—launched its Scribe model this spring, delivering secure, multi-speaker text transcripts at roughly 40 cents per recorded hour² orders of magnitude less than the cost of a manually created transcript.

The net result of these advancements is a fast, accurate, and inexpensive record of all important conversations—not just those that take place in a courtroom or deposition.

MODERN SPEECH-TO-TEXT TECHNOLOGY APPROACHES HUMAN ACCURACY WHILE COSTS CONTINUE TO DECREASE
A recording stores sound; a transcript converts that sound to text. State-of-the-art speech recognition systems are able to convert audio into text by using advanced neural networks to recognize speech sounds from audio spectrograms.³ From this it can predict word sequences and differentiate between speakers resulting in highly accurate transcriptions. Today, ElevenLabs reports a 97% accuracy rate on multilingual speech-to-text benchmarks—serving as the current state-of-the-art across all models.⁴ Independent testing confirms 96.7% accuracy in English.⁵

For comparison, the National Court Reporters Association requires only 95% accuracy for the Registered Professional Reporter exam at 225 words per minute.⁶ The gap is narrowing, and AI’s accuracy continually improves with every update while costs trend downward.

THE PROFESSIONAL ETHICS COMMITTEE FOR THE STATE BAR OF TEXAS SUPPORTS RECORDING CLIENT CONVERSATIONS FOR LEGITIMATE PURPOSES
The Professional Ethics Committee for the State Bar of Texas issued Opinion 575 (November 2006), which recognizes legitimate purposes for the recording of client conversations—including maintaining accurate records, gathering witness information, and protecting against false accusations.⁷

The opinion explicitly permits attorneys to record client telephone conversations—even without disclosure—provided certain conditions are met. For AI transcription specifically, the opinion establishes a clear framework: recordings are ethically permissible when they (1) serve a legitimate purpose for lawyer or client, (2) protect confidential information under Texas Disciplinary Rules of Professional Conduct Rule 1.05, (3) comply with all applicable laws, and (4) don’t contradict any representations made by the lawyer.

While client consent is not required under the opinion, best practice certainly involves informing clients when conversations are being recorded or transcribed.

COURTS WILL NEED TO DISTINGUISH BETWEEN PROHIBITED RECORDINGS AND PERMISSIBLE AI TRANSCRIPTIONS
As speech-to-text technology advances, courts will increasingly need to distinguish between prohibited audio recording and permissible AI transcription. After all, if individuals can hire stenographers to create verbatim transcripts of open court proceedings, shouldn’t attorneys be permitted to use AI tools that perform the same function without preserving audio? Already, wearable devices such as “The Bee” claim to convert speech to text in real time while deliberately not storing any audio recordings.⁸,⁹ The technical distinction between recording and transcribing may soon form the basis for legal arguments allowing AI transcription tools in venues where recording devices are traditionally prohibited.

TEXAS HAS ALLOWED NON-STENOGRAPHIC DEPOSITIONS FOR OVER 25 YEARS
The Texas Rules of Civil Procedure (TRCP) embraced alternative recording methods long before AI made today’s advancements possible.¹⁰ TRCP Rule 199.1(c) permits any party to “cause a deposition upon oral examination to be recorded by other than stenographic means, including videotape recording.” The rule requires only reasonable notice (at least five days prior to deposition) specifying the non-stenographic method to be used. Other parties may designate additional recording methods at their own expense. Already commercial vendors like Skribe¹¹ and Prevail¹² offer transcripts wholly created by AI for depositions conducted over Zoom.

WILL AI REPLACE COURT REPORTERS?
Given the widening gap between supply and demand—and the rising cost of human labor—the profession may have little choice. AI transcription systems are already flirting with parity of their human counterparts while operating without fatigue or scheduling constraints. When mistakes occur, lawyers can invoke the same safeguard that exists for human error: TRCP Rule 203.1’s “read-and-sign” process.¹³ Parties will continue to review the rough draft, note corrections, and lock in the transcript knowing that any substantial errors can be resolved through an analysis of the video.

For decades, technology has threatened to replace court reporters, causing stenography school enrollment to plummet. AI transcription is close to matching human capabilities, so things could shift. Depositions will more than likely convert first, pushing remaining stenographers to courtrooms until those positions eventually become automated too. With AI delivering comparable results for pennies, the economics are undeniable. The profession will not end because technology is perfect—it will transition because AI has become good enough at a fraction of the cost.

HIGH-IMPACT USE-CASES FOR LOW-COST TRANSCRIPTION
Client-intake calls. Attach accurate transcripts to the matter file and search for key facts, dates, and information months later. These digital records transform intake calls into permanent, searchable assets—enabling attorneys to instantly retrieve critical case information and detail that would otherwise blur with time. The transcript(s) can also play a key role when securely analyzing case information within large language models, providing data that can be used to identify or articulate potential legal issues, and preserve institutional knowledge when cases transition between attorneys.

AI non-stenographic depositions. On routine discovery depositions, counsel can dispense with a live reporter and order certification later if necessary. Even better, attorneys can stipulate to verify and accept the AI transcript as the official record unless objections arise, dramatically reducing client costs. This approach follows an established structure under TRCP Rule 199.1(c) while using the latest technology to streamline proceedings.

Witness preparation. Feed prior statements into a large language model to assist in witness preparation and aid in refreshing recollection. This AI-powered workflow can quickly identify patterns, inconsistencies, and key testimony points across multiple sources, providing attorneys with strategic advantages when preparing witnesses for deposition or trial.

Client interviews for affidavits. Transcribe detailed client accounts that can later be refined into declarations or affidavits, ensuring no important details are missed and reducing the time needed to prepare these documents.

Mediation sessions. Create accurate records of settlement discussions while preserving confidentiality. These transcripts can help track evolving settlement positions, document conditional offers, and ensure all parties have a common understanding of proposed terms once a settlement is reached. The final transcript could then be used to draft a detailed mediated settlement agreement.  

Hearing summarization. As technology advances, lawyers could create unofficial transcriptions of court hearings
(with the court’s permission), quickly transforming hours of proceedings into concise summaries for clients and colleagues. This application would enable immediate case updates, preserve crucial details of court rulings, and enhance collaboration across the firm. 

Board meetings and corporate governance. For attorneys serving as corporate counsel, AI transcription can create accurate meeting minutes and document discussions about legal risks, compliance issues, and key corporate decisions, helping satisfy fiduciary duties and creating contemporaneous records for later use.

Legal education and training. Record and transcribe mock trials and training sessions, allowing law clerks and new associates to review their performance, receive more precise feedback, and learn from observing others.

CHASE HARDY  is a personal injury attorney based in San Antonio. He is a nationally recognized speaker on the impact of large language models on the legal profession and has presented on this topic at over 30 legal education events across the United States since 2021; see chasehardy.com.