Steve Townsend on AI, the court record, and access to justice

Chief Product Officer and Co-Founder, TheRecordXchange®

"The question is not whether courts will adopt AI. It is whether they will manage the adoption that is already underway." — Steve Townsend, Chief Product Officer, TheRecordXchange

A court record can sit quietly in the background of a case until the moment someone needs it.

Then everything about it matters.

The record may be what allows a case to move forward, what an appeal is built on, or what someone needs to understand a decision that shaped their life. But those purposes depend on something fundamental: people being able to access and use the record.

When obtaining the record is slow, expensive, or difficult, the burden falls hardest on people representing themselves, people relying on legal aid, and others with limited resources. For them, the record can become a barrier to justice rather than a pathway to it.

Courts have made significant progress in capturing proceedings digitally. But capturing what happened is only the beginning. The recording still has to become something people can use: searchable, accessible, and, when required, certified.

That is where artificial intelligence has the potential to change what is possible.

Steve Townsend has watched that evolution from more sides of it than most.

He has spent his career at the intersection of courtroom recording, transcription, and technology. He led For The Record (FTR) through the years when digital courtroom recording expanded across more than 40 countries. In 2014, he and longtime colleague Erik Lige co-founded TheRecordXchange (TRX) to address what came next: everything that has to happen to a recording before someone can actually use it.

He has also helped shape the profession itself. In 1993, he co-founded the American Association of Electronic Reporters and Transcribers, which established early certification standards for electronic reporters and transcribers and is known today as CAPTUR.

Today, as Chief Product Officer at TheRecordXchange, Steve focuses on the future of the court record and the responsible use of AI. Those questions are becoming increasingly important as courts consider not simply what AI can do, but how it can be used without compromising the integrity and accessibility of the record.

This October, Steve will bring that perspective to two conversations about the future of courts. At the 2026 Access to Justice Chairs Meeting in Indianapolis, he joins a session on how AI-enhanced transcription can improve access to justice. In Williamsburg, Virginia, he takes part in "AI & Courts: What Capabilities Do Our Courts Want and Need?", a Center for Legal & Court Technology conference at William & Mary Law School, sponsored in part by the National Center for State Courts.

In the conversation that follows, Steve discusses what decades in the field have taught him about technology, where AI can make a meaningful difference, what should remain firmly in human hands, and why making the court record more accessible is ultimately an access-to-justice issue.

AI is being called transformative for courts. After watching technology evolve in this field for decades, what do you think courts need to understand first about what AI can and cannot do with the court record?

What decades in this field teach you is that any confusion usually starts with the conversation, not the technology. A new thing arrives, everyone talks about it as one undifferentiated capability, and courts end up either expecting too much of it or rejecting all of it. The first thing to do is understand the different kinds of AI.

Concerns almost always come from what people are reading about large language models (LLMs). Generative AI offers courts the greatest opportunity for transformation I have seen in my career. Transformation not just with the official record but across a wide range of court operations. It also carries real risks that must be managed. Automated speech recognition (ASR), however, is a different and much more mature technology that continues to improve rapidly. Importantly, it does not carry the risks people worry about with generative AI.

From there it comes down to what each one is good at. ASR produces a very good first draft. LLMs are good at context, so they can improve that draft and help court reporters and transcribers turn it into a certified transcript. What AI should not do is decide anything. It does not certify the record, and it should not be the final authority on what was said. The original recording stays the source of truth, and a well-designed workflow makes getting back to the recording quick and easy. With the right guardrails, that combination can genuinely improve how courts produce and deliver the record.

So where does that actually show up? Where does AI save courts time day to day?

For as long as courts have relied on recorded proceedings, the biggest problem has been the time and cost between recording and access to a usable transcript. Appellate courts wait months. Many attorneys and self-represented litigants pay more than they can afford or simply go without. That is where AI earns its place.

Speech-to-text turns a recording into a working draft in a fraction of the time and at a fraction of the cost. It also produces a great deal more than text. Timing, confidence scores, speaker attribution, all captured alongside the words. That data is what makes everything downstream work. It supports editing and certifying the transcript, searching it, analyzing it, and jumping straight back to a specific moment in the recording. A judge or a clerk gets to the exchange they need instead of listening through an entire hearing.

There is also a floor that did not exist before. A recording paired with an ASR transcript is not certified, but it is searchable and readable, and it costs the court almost nothing to deliver. For a person who needs to know what was said and cannot afford a certified transcript, that is often the difference between having the record and not having it.

A record that used to take weeks or months can be produced in days or even minutes. That is how a court cuts its backlog.

Accuracy is everything in a court record. How do you keep AI from introducing errors?

Two things, plus a third that follows from them.

First, capture quality. AI can only be as good as the audio it works from. Poor microphone placement or bad room acoustics will degrade a transcript no matter how good the technology is.

Second, humans stay in the loop. AI produces a draft, and trained people review it. The AI does the heavy lifting, but a human remains accountable for what goes on the record.

Third, verification. Because the transcript links directly back to the audio, a reviewer can check any passage against the source in seconds. That is a real change. Accuracy stops being something you take on faith and becomes something anyone can confirm.

That last point also raises a question about access to recordings. If the audio is available, people have a way to verify the transcript against its source. Withholding the recording does not make the record more accurate. It can slow production, raise costs, and leave people relying on a transcript they have no way to independently check. That is why the restrictions many jurisdictions still place on releasing recordings deserve another look.

TheRecordXchange describes itself as partnering with courts, not just selling to them. What does that partnership actually look like, and why does it matter?

I am wary of the word partnership, because every vendor at every conference uses it. What I can tell you is how we are set up.

We sell a subscription, not a system. There is no point at which we have been paid and the relationship is over. That creates a very different incentive from installing something and moving on. If we stop being useful, the court leaves, and that shapes how we build and who we hire.

The structure is one thing. The other is time. I have worked with some of these courts for 25 years, across three companies, through staff turnover on both sides, and I still talk regularly with people I met in the 1990s. Those relationships are personal. Valuing relationships that way is something I have tried to instill in every company I have built.

You do not get that by closing deals. You get it by being there when something goes wrong.

The record itself is the part I take most seriously. It is not data to us. A court hands us something that belongs to the people whose cases it documents, and that carries an obligation that does not expire when a contract does.

You're speaking on access to justice this fall, and it echoes something in the TRX mission, that the record belongs to everyone it affects. What does that line mean to you?

To me the line is about obligation more than ownership. The court holds the record. But it holds it for the people whose lives it documents. The person trying to appeal, the parent in a custody case, anyone who needs to know exactly what was said and decided. A record those people cannot get is not serving its purpose, however carefully it is kept.

The access-to-justice community understands this already. The audience that still needs to hear it is courts and legislatures, because the cost and the delay are set by rules and budgets, not by technology. For a self-represented litigant, the transcript is often the largest single cost of pursuing an appeal. Waiver processes exist, but they take time and frequently do not come through. And transcript production is the most time-consuming step in the appellate process. In many jurisdictions the delays are chronic and severe.

That is how a record becomes a barrier rather than a right. When it is slow, or costs more than a person can pay, it quietly shapes outcomes that should turn on the merits.

The reason I am hopeful is that the tools to fix the cost and the delay now exist. The technology is here. What it requires is responsible adoption, and the decisions that unlock it sit with courts and the legislatures that fund and govern them.

In August, California's Supreme Court unanimously ruled that when no court reporter is available, trial courts must create an official verbatim record by electronic recording for civil litigants who can't afford one, a narrow exception to the state's longtime ban. How do you read a ruling like this, and what does it signal about where courts are heading?

I read it as a court naming an access-to-justice problem out loud. The ruling was not really about technology. It was about people being left with no official record at all, and without a record there is no meaningful way to appeal. California's own data shows that more than 3 million hearings in family law, probate, and unlimited civil proceedings took place without an official verbatim record since 2023. As the Chief Justice put it, a verbatim record created by electronic recording is better than no record at all. I think that is exactly right. For years, the people hit hardest by the gap have been those with the least ability to pay their way around it, and this ruling treats access to the record as something courts have an obligation to provide. The accuracy concerns are real and worth taking seriously, but the honest comparison is not recording against a perfect transcript. It is recording against silence.

What strikes me is that California has been dealing with this problem for years. The court reporter shortage exposed a gap between the right to a meaningful record and the state's ability to provide one. The Supreme Court's ruling addresses that gap by making clear that electronic recording can be used when an official reporter is unavailable.

But tying the use of recording to reporter availability is itself an obstacle to responsible adoption. It leaves courts unable to plan, budget, or build a long-term strategy around a capability they can use only by exception. The ruling also leaves an important question about what happens next: the recording itself can be one of the most direct ways to address cost and delay, but creating the record is only the beginning. Courts still have to decide how that record will be accessed, used, and turned into something people can rely on.

We will have to see how California's courts respond. My hope is that they read the ruling broadly and act in a way that benefits the litigants and attorneys who have been denied a record for years.

Looking ahead, what will it take for courts to adopt AI well, and not just quickly?

I would challenge the idea that courts get to decide when adoption starts. It already has. Our CEO, Eric Parsons, has written about shadow AI. This is the pattern in any organization: if leadership does not provide secure and appropriate tools, people use whatever they can get to on their own, outside policy and outside anyone's visibility.

That is happening in courts right now. We recently saw an appellate judge call out a court reporter for filing a transcript with obvious AI-generated errors left in it. Judges and attorneys are using public chatbots to work through their cases. The risk is not only bad work product. It is confidential case information leaving the building, and in some cases being retained by whoever operates the tool.

The question is not whether courts will adopt AI. It is whether they will manage the adoption that is already underway. That means providing tools people can actually use, working with vendors who are accountable for how data is handled, and writing policies that are both effective and enforceable. A policy nobody can follow produces more shadow AI, not less.

Beyond that, the fundamentals do not change. Start from the problem, not the technology. Demand accuracy you can verify and a clear record of what the AI did and what a person decided. Never let the tool blur the line between assisting and deciding.