Bringing Generative AI to the Judiciary with Judge Allison Goddard

As generative AI reshapes industries across the globe, the legal profession is facing a pivotal moment, where attorneys and judges alike must partner technical awareness with thoughtful, responsible integration.

Few figures embody this shift more than Judge Allison Goddard, a federal magistrate judge who has emerged as a leading voice on the use of AI in the judiciary. With a career rooted in both technology and law, she brings a rare blend of hands-on experimentation and judicial insight to the conversation—pushing forward the idea that technological fluency is now essential for lawyers and judges alike.

Judge Goddard’s work is helping to normalize the use of generative AI in courtrooms and chambers, offering a pragmatic roadmap for how these tools can enhance—not replace—human judgment. Whether using AI to accelerate legal research, support pro se litigants, or manage large caseloads with limited resources, she sees immense potential for these technologies to improve access to justice and reduce cognitive load for overburdened courts. Judge Goddard spoke with Everlaw about generative AI, privilege logs, and more.

During your career on the bench, you’ve become a recognized voice at the intersection of law and technology. Can you share a bit about your path to the judiciary and what drew you to focus on areas like legal technology?

I would say there are two different things that drew me in.

One is that I moved to San Diego first thing after college. I worked for a startup company that was doing telecom, so it was kind of a tech company. I’d always been interested in tech. One of my roles there was as a kind of contract administrator. I’d ask questions of the lawyers about why we were doing this or maybe we should try and negotiate that, and I felt like they were really paternalistic with me and I didn't like it. I felt like I could do their job, too. And if I had a law degree, then I would be able to do it with the respect that they got for being lawyers. So that helped me realize I needed to go into the law. Otherwise, I was going to be stuck without a place to move up in the corporate world, doing what I wanted to do. I didn't want to do marketing. I didn't want to do business development. I was really attracted to the legal side, so I figured I had to get a law degree.

And then after law school I worked at Cooley, which had a lot of biotech and high-tech clients, which kept my interest going. And then when I went on and had my own firm, where I was basically the tech person.

"Having technological competence gives the courts confidence that you are a competent lawyer."

• Share this quote:

What got me really interested in AI is that I teach a high school class, and one of my students turned in an essay that was a little surprising because it was more coherent than I would have expected. My daughter was also in the class, and, without using real names, I said to her, "Tom turned in a really good essay. That was interesting." And my daughter said, "Oh, mom, that's ChatGPT."

I immediately said, "I need to figure out what this is." And my husband, who’s also a high school teacher, said "How have you not heard of this?" And I told him, "Some people are using WordPerfect in the courthouse. This is not something that we learn at work.”

So my husband showed me how ChatGPT worked, and I could see that this was a total game-changer. That’s what really sparked my interest in AI.

I know a lot of people are in despair about this, thinking people aren’t going to know how to write anymore, but I think it’s just going to change the way people do it. We're going to become more second drafters than first drafters, which is not necessarily a bad thing as long as we keep humans in the loop.

During our Everlaw Summit panel, you spoke about the idea that being comfortable with technology is a differentiator and how those who can use it effectively can really separate themselves in their practice. Can you speak about that a little more?

I don’t think anyone expects people to walk in and be able to start coding in front of the court, but if you show that you're technologically competent, it gives judges confidence that you’re up to date on how law is practiced. It provides confidence that you're doing a good job representing your client.

For example, with the transition to Zoom during COVID, I was really impressed with a lot of lawyers, even ones who have been practicing for decades, who quickly adapted. And with the ones who didn't adapt, it makes you wonder that if your secretary is still logging you on to Zoom, are you really able to best represent your clients in today's age? Are you able to understand that if you have an individual employment case, you need to tell your client to preserve data at the beginning of the litigation, including their phone? And if you don't even know how your phone works, that's really a concern.

I worry about some lawyers not being competent. They're not going to be able to really represent their clients. Having technological competence gives the courts confidence that you are a competent lawyer. If you want someone to represent transactions, for example, and be the class counsel, showing that you're technologically competent is really important because that's how we know as a court that you can manage litigation in an efficient way.

You’ve done a pretty extensive amount of experimentation with generative AI tools, especially for a judge. Can you speak about those a little more, and what potential you see for generative AI from a judges perspective?

I can remember about 18 months ago, once some of these generative AI tools really started picking steam, people were like, "Does this mean that the pro se litigants are going to flood the courts?” And we haven't really seen that at all. I haven't seen pro se adoption to the level I thought I would. Now, that also might be because a lot of our pro se litigation comes from prisoners, and they don't have access to anything.

One of the first things we did as an experiment with generative AI was upload an excessive force 1983 complaint, and also, the motion to dismiss that the government agency had filed.

We ran it through a few different tools to draft our opposition, and it was actually really good. It contained about 25% hallucinated citations, but I asked our pro se lawyer clerk about that, and she said, "That's something we would be able to check and catch quickly." But if we could get the litigants to have such coherent arguments tied right back to the allegations in the complaint, that would actually be much better for the court and for the litigants.

That was when I realized we shouldn't be so afraid that pro se litigants are going to flood us with lawsuits. We ought to look at it and say, "Wow, this could be a really amazing way to increase access to justice." I still think we're a ways away from that.

I've given this example before, but I've created something like my own little LLM of Social Security orders (because I do Social Security appeals) through tools like NotebookLM and CoCounsel. And while these orders are very fact-specific, there's also a lot of the fundamental law that’s the same. It's also different depending on what the facts are of the particular appeal.

For example, if it's being challenged because of the plaintiff's testimony, or if it's being challenged because of the evaluation of the medical evidence, you need to tweak exactly what your order says based on that.

We don't have a document management system, so all I have is a collection of Word documents in a folder. I think Windows is just terrible for indexing and searching and File Explorer.

I was able to take those orders, which are all publicly available, and put them in both NotebookLM and CoCounsel just to try out different programs. Then I can prompt against my own documents and say, "Draft the legal standard for this type of issue," or "Pull all orders where I address whether or not the person claimant who has schizophrenia was disabled." And then I can pull up all those really specific cases and look for the most factually similar cases to support my decision, which has been a game-changer because it's much faster.

Then, depending on the tool I'm using, I can actually share the database with my clerks and they can update it for me and they can use it to do the first draft of orders. When I say do the first draft of orders, there is no tool where you can put the party's moving papers in and then hit the easy button and then it turns out a first draft. I usually use it to do parts and pieces of the order, which I think is a much faster way to go about it. It's a much more efficient use of my time.

Have you seen pro se litigants using generative AI in the courtroom yet?

I've only seen it a few times that I've recognized.

It's come with things like, when requesting a status conference with the court or something small, they'll cite to California District rule this or that. Well, there is no California District. There's a Southern District of California. But I can look at it and say, "Okay, I think what they're asking for is that they're having some problems in discovery and they need the court to intervene." For me, it’s important to see that at least they're trying.

"I would love to see courts, at least, start to experiment with literally just having a generative AI tool do the first iteration of the order."

• Share this quote:

In fact, unless I feel like a pro se litigant is trying to purposefully mislead the court, I probably wouldn't even make a big fuss about it. I might drop a footnote that says something like, "The court just wants to caution the litigant that they need to double-check all the citations in their briefs because it appears these are not valid cases." But I don't know that I would come down really hard on them. It'd be different for a lawyer. Lawyers don't have the same excuse.

It's going to be a much easier brief for me to read and understand if the average pro se litigant gets help from an AI tool in drafting it. I can detect the fake citations. I can work around that. And I would give them the benefit of the doubt that they were trying their best.

You stated in a previous interview that “the important thing is that the playing field is level,” when two parties are in the courtroom. While this isn’t always necessarily the case in each matter, do you fear technologies like generative AI could make the playing field even more uneven?

Even with the most easily accessible tools, if you know how to use them, that itself could help level the playing field.

With discovery, it's a little bit of a different issue because those discovery-specific generative AI tools are still coming out.

Let's say that you are a pro se litigant, for example, you could upload a bunch of documents to a tool that's relatively inexpensive and start prompting against it and asking it to review everything. However, it's probably going to be a little clunkier than one of the more polished programs. And I think if you have the resources, you’d probably get the polished program. But I do think the discovery-specific generative AI tools can actually help. I think they're more likely to level the playing field than increase the disparities.

In the Summit panel, you and Judge Pennypacker (Santa Clara County, Cal.) both discussed the idea of cognitive load, which centers around the idea that our brains can only comprehend and process a certain amount of information at a given time. Do you believe generative AI has the potential to help judges take on more cases, or at least move them through the system faster?

I can't emphasize this enough. The State Court is extremely overburdened in California right now. I believe during the panel Judge Pennypacker said that she's on a little bit of a different calendar system than we have in San Diego, but I know that the San Diego civil judges have something like 1,200 to 1,400 cases on their calendar, with only one research attorney. I don't know how they do that. I'm just amazed all the time. So I think these tools could really help them.

And it’s interesting because I think, gosh, we're busy enough in the Federal Court, and I have a much lighter docket, plus two full-time clerks, and I know how much these tools could really help us.

For example, there are a lot of things judges do that are more administrative in nature, like with scheduling orders, where generative AI would be great.

I would love to see courts, at least, start to experiment with literally just having a generative AI tool do the first iteration of the order, and then the judge reviews it, and you can post it as a tentative, and then let the parties argue it. So, everyone has due process there. It's not the machine making the decision, but it is helping speed up the process.

Motions to dismiss and pleading motions on pleadings really clog up the courts. Most of the time they aren’t dispositive. They narrow the case, which is important and good. But we could have multiple rounds of pleading, and it just takes a long time. It delays justice in the case of a resolution on the merits for a long time. And if you're looking at the low-hanging fruit for how we might start to integrate generative AI into our workflows in a non-administrative way, that would be it for me.

You recently co-authored a paper forthcoming in Volume 26 of The Sedona Conference Journal titled “Navigating AI in the Judiciary: New Guidelines for Judges and Their Chambers.” Why did you and your co-authors feel the need to create these guidelines, and what are you hoping they’ll achieve?

I've been really lucky between the work I do with the Federal Judicial Center and the Sedona Conference and some other groups to meet judges from all around the country.

There's a group of us that coalesce together, and we have an email chain about these types of issues like ediscovery and also technology and generative AI. Dr. Maura Grossman has been a big part of that, too. She's kind of the technical tether for us because she actually knows all the technology inside and out.

"The more you use one of these tools, the more you learn to interact with it."

• Share this quote:

In our group, I think Judge Schlegel was the first person to suggest the idea for the paper. He and Dr. Grossman were working on the ABA AI Task Force together, and thought they should also be thinking about the judges. The ABA had come out with guidelines for lawyers, and we needed the guidelines for judges.

At the same time, I was creating my own kind of internal guidelines for how to use AI in chambers. One of the most important things about that is seeing those guidelines has encouraged other judges to try it because it's not so unknown. And knowing that there's a responsible way to use it makes it a little less scary.

I think the group's thought was, if we have something like that, then other judges will see that there is a responsible way to use it, and maybe they'll start to think about using it themselves.

The paper also lists out what are deemed appropriate uses for AI in judicial settings. Can you tell us about those? How did the group create this list, and are there any uses not listed that could provide value for judges in the right context?

I don't believe there were any use cases that didn't make it.

I know that Judge Rodriguez did something pretty interesting by using a generative AI tool to experiment with his own work. After a bench trial that took place over something like eight months with a ton of exhibits, he did everything traditionally. He drafted his findings of fact, his conclusions of law. And then once he was done with that, he fed everything into an AI tool and had it do the drafting, and it did a pretty good job. There's going to be an article in Judicature coming out about that soon.

And I know Judge Schlegel and I both use AI almost on a daily basis. And so, a lot of those use cases came from the things we've already done, and just try to give people some ideas to get started.

What generative AI or legal technology innovations are you most excited about?

The Judicial Conference just formed an AI task force for the federal government, and several of them are people who are using generative AI both in the administrative side of the courts, and also as judges. I'm hoping that will help propel us forward in the federal courts.

I’m also hoping that this year we’re able to move beyond the trial period for a lot of the generative AI tools we currently use and retain permanent access to them.

But in the meantime, most of the tools I use are publicly-available, where I've been able to review the terms of service and confirm that my data, my inputs, and outputs will not be used to train the model. I have my own subscription to some tools, where I get kind of a different level of privacy, but those are the ones that I use all the time because they're so easy.

And I use them to do silly things too, like for a trivia contest at the court with our externs, I used it to draft my questions.

But also, if I'm drafting a quick order, it will be much faster if I can upload the party's motion and the opposition and I can have it summarize our arguments, and then use that in my order. I use it all the time when I get a lengthy.

In our court, we have a magistrate judge and a district judge for every case. And the district judge handles all the dispositive matters like motions for summary judgment, motions for class verification.

Or when I get a lengthy motion order from the District Court in one of my cases, one of the things I need to do is stay up on the case and see if it's time for them to talk settlement again. So instead of reading a 65-page order from the district judge, I'll just throw it into a generative AI tool and have it summarize the order.

The more you use one of these tools, the more you learn to interact with it. I'll ask it, "Okay, what claims are remaining?" Then it'll mention them. Or I'll ask it to summarize the order. And I'll notice that it'll say, "The court indicated that it was going to sanction this." I'll say, "Can you tell me more about the sanctions issue?" or "Why didn't the judge dismiss the claim for intentional infliction of emotional distress?" It's really interactive reading.

All I'm doing is just reading it to assess whether or not it's time for us to talk settlement again. So, it doesn't have to be perfect. I'm not using it to draft an order. I'm just using it to keep me up to speed on things. And that's a great way to do it.

Switching gears a little bit, you’ve spoken about how privilege logs are often an afterthought in litigation preparation. Why do you think that is, and how can they take on more importance in the eyes of attorneys?

They're an afterthought because they're such a pain in the neck.

And that's where I get really excited about some of the generative AI tools for developing privilege logs because I think they could take some of the tediousness out of it. In my own experience as a young associate, doing privilege logs with paper documents was a really imperfect science.

For example, as you were going through the documents and pulling things, you were often overly careful and cautious because it's much worse to inadvertently disclose a privilege document than it is to find out you've over-withheld and you need to produce something.

A lot of times, I felt like it would have been better to review as a team, and generative AI really simulates that team review if you're using a tool that's helping you with the privilege and it's telling you why it thinks it could be or could not be privileged. It's helping the lawyer who's reviewing the outputs of that tool and giving them a second opinion in some ways.

"It's a no-brainer. It's just that people haven't learned to think about doing it that way yet."

• Share this quote:

If you could get to a better starting place with getting the log itself put together, I think people would be much less hesitant about doing it in the first place.

But I can tell you from personal experience, when you go and do the log, you usually find a number of documents you shouldn't have withheld. It's an important exercise.

Do you have any advice for attorneys or judges who are getting started with generative AI tools, or anything you think they should look out for as they're kind of getting started with technology generally?

I think the easiest way to get started is just to use one of these tools.

It would even be fun to just give someone three exercises to do that would demonstrate how good these tools can be. Because once you see how it can make certain things easier, I think that people are much more likely to use it.

One of the things I would suggest is to take a really long order, upload it to a generative AI tool, and have it summarize the text. Then ask three more questions about the order to the AI tool, and see the types of responses you can get.

For lawyers, I think the number one thing to do before your client gets deposed would be to upload documents to a generative AI tool, and have it come back with the questions it might ask your client from the defense and plaintiff sides. And then you can prep your witness based on those questions.

Your witness is going to be much better prepared for the deposition than if you had never used generative AI. I can viscerally remember three or four depositions that I wish I’d had generative AI for, because I feel like I would have done better preparing my witness.

And you don't have to worry about ceding any of your lawyer tasks to the technology. Giving your clients some concrete ways that the other side may come at them in a deposition with sample questions based on the case itself is just much better preparation.

Lastly, is there anything you think it would be important to include in this interview that you think it might be important for any attorneys or judges to know?

I’d just like to encourage them to try generative AI.

Someone said, the best way to get started is by having it create a packing list or write a poem for you or whatever.

Like I said, if you're a judge, and you want to see how powerful it can be and how it can help with your work, upload an order and ask it a lot of questions. Particularly, if it's an order you're familiar with or a Supreme Court opinion, you'll see how powerful it can be in summarizing information. Or have it create a timeline for your case based on the trial briefs or something like that.

For lawyers, I can't say enough about using it for a deposition prep, using it to prepare for an oral argument or any sort of hearing before the court. You can ask "What questions do you think a judge might ask me based on these briefs?" It's such an easy way to prepare. It's a no-brainer. It's just that people haven't learned to think about doing it that way yet.