Module 1: When a Litigant Brings an AI Draft to the Counter

The VoxJurisPath Team

Module overview

This module teaches front-line staff how to respond when a self-represented person brings a document drafted with an AI tool. It takes about 25 minutes and needs no technical background.

It is written for court self-help center staff, clerks, law librarians, legal aid intake workers, paralegals, and volunteers. Attorneys at legal aid programs can use it to train the non-attorney staff they supervise.

After this module, you will be able to:

Explain why more litigants are arriving with AI-drafted papers.
Tell the difference between legal information you can give and legal advice you cannot.
Hold a short, consistent conversation about an AI draft without endorsing or criticizing the tool.
Point the person to what they must check before filing.

This module is training for staff. It is not legal advice, and it does not replace your organization's policies or your court's rules. Where they differ from anything here, follow them.

Why this is happening now

Free chatbots have made it fast to produce papers that look like a lawyer wrote them. Researchers studying federal courts have found more self-represented filings since these tools became widely available.

A working paper from researchers at MIT and USC, covering more than 4.5 million federal civil cases, found a sharp rise in self-represented filings after AI tools spread. Docket activity by self-represented litigants in their first 180 days rose 158% over pre-AI levels (Federal Bar Association).

Looking better is not the same as doing better. A separate study of about 2.8 million federal filings found no improvement in win rates for complaints that showed signs of AI drafting. Those complaints were more likely to be dismissed, and dismissed earlier (arXiv).

Courts are responding by training the people at the counter. The Alaska Court System first tried to teach litigants how to prompt AI tools, then moved to guidance for clerks, librarians, and self-help workers instead (Thomson Reuters Institute). At a 2026 judicial summit, a judge on the panel cautioned against harsh sanctions for good-faith AI mistakes and favored education (Thomson Reuters).

The person in front of you has probably already used a chatbot. Your job is to help them see what still needs checking.

The line between legal information and legal advice

You can explain how the process works. You cannot tell someone what to do in their case. An AI draft does not move that line.

Legal information is general and would be the same for anyone in the same situation. Legal advice applies the law to one person's facts and recommends a choice. The table shows the difference using questions staff hear about AI drafts.

Question from the litigant	Legal information (you can)	Legal advice (you cannot)
"Is this motion good?"	Explain what the court's rules require a motion to contain.	Say whether their motion will work or should be filed.
"The chatbot cited a case. Is it right?"	Show them where to look up a case and confirm it exists.	Tell them whether the case supports their argument.
"Which form do I use?"	Point to the forms list and the instructions for each.	Pick the form for them based on their facts.
"When is this due?"	Show where deadlines appear in the rules or the notice they received.	Calculate their deadline and tell them to rely on it.
"Should I mention that I used AI?"	Show them the court's AI disclosure order or local rule, if there is one.	Tell them whether disclosure helps or hurts their case.
"What should I say at the hearing?"	Describe how a hearing runs and who speaks when.	Script their testimony or argument.

Who may give advice depends on your role. Attorneys at a legal aid program can advise their own clients. Non-attorney staff, and attorneys staffing a public help desk without a client relationship, generally cannot. Your court or program may have written guidelines on this line; those control.

The counter conversation

Use the same four steps every time. Consistency protects the litigant, and it protects you from drifting into advice.

Acknowledge the draft without judging it.
- "Thanks for bringing this in. A lot of people are using these tools now."
- Avoid praising or criticizing the tool. Either one can sound like an endorsement or a legal opinion.
Ask what the tool was used for.
- "Did the program write the whole thing, or did you use it for parts?"
- "Where did the facts in here come from?"
- The answers tell you which checks matter most. You are not judging the content.
Name what the person must check before filing.
- Every case and statute cited exists and says what the draft claims.
- Every fact in the draft is true and is theirs.
- No blanks, brackets, or placeholder text remain, such as "[insert date]".
- Names, case number, and court match their actual case.
- The court's AI disclosure rule, if one applies.
Point to resources and stop.
- Self-help forms and instructions, the law library, legal aid intake, lawyer referral, or limited-scope help.
- "I can't tell you whether this will work. I can show you where to check it and who can give advice."

Disclosure rules vary by court, and they change. A federal judge in the Western District of Texas, Del Rio Division, requires a "Certificate of Generative Artificial Intelligence Usage" from every party who uses AI to draft a filing, including self-represented parties (Standing Order, June 25, 2026). Florida went the other way: since June 15, 2026, Florida state courts may not impose local AI disclosure rules, and every signer instead represents that the cited authorities exist and are accurately cited (AOSC26-12). Keep your own court's current order or rule bookmarked, and check it before you point anyone to it. Our Court AI Rules directory summarizes current orders by court, with links to each official source.

What to avoid

Most slips happen when staff try to be extra helpful. Watch for these.

Editing the draft or suggesting new wording for an argument.
Saying a cited case is "good" or "bad" for them. You can confirm a case exists; you cannot judge whether it helps.
Recommending or warning against a specific AI product by name.
Promising the clerk will accept the filing, or that the judge will not notice AI use.
Telling someone their draft is "fine" because it looks professional. Polished language can hide invented citations.
Taking or keeping a copy of the draft unless your program's policy allows it.
Typing the litigant's personal or case details into an AI tool yourself, unless your organization has approved that tool for client data.

If you are unsure whether an answer crosses the line, say so and refer the person. "That's a question for a lawyer" is always an acceptable answer.

Worked scenario: an eviction answer drafted with a chatbot

The names and the cited case below are invented for training.

Dana, a tenant, comes to the self-help desk two days before her answer is due. She hands over a three-page answer to an eviction complaint. She says a free chatbot wrote it after she described her situation.

You notice three things while she talks. The caption reads "[Landlord's name] v. Dana Ruiz." The signature block has no date. The second page cites a case called Harmon v. Delgado Properties and says it "requires dismissal" when repairs were not made.

A weaker response

"This looks really professional. That case sounds like it helps you, so I'd file it today. Just fill in the landlord's name."

This response endorses the draft based on how it looks. It judges a cited case without checking it exists, which is advice. It also skips the disclosure question and the other blank.

A better response

"Thanks for bringing this in. Before anything gets filed, there are a few things you'll need to check yourself. I can't tell you whether the arguments will work, but I can show you what to look at."

"I see a bracket in the caption where a name should be, and the date by your signature is blank. Those need to match the complaint you were served."

"This page cites a case. Chatbots sometimes cite cases that don't exist. Here's how to search for it at the law library terminal. If you can't find it, take it out or ask a lawyer about it."

"Some courts require you to say if you used AI. Here is our court's rule on that. And because your answer is due soon, here is the legal aid intake line and today's walk-in hours."

The better response follows the four steps. It names specific checks, gives no opinion on the merits, and ends with a referral that fits her deadline.

Counter checklist

Print this or keep it at the desk. Each item is something you point out; the litigant does the checking.

☐ Acknowledged the draft without praising or criticizing the tool
☐ Asked what the tool wrote and where the facts came from
☐ Pointed out any blanks, brackets, or placeholder text
☐ Pointed out that names, case number, and court must match the case
☐ Explained that every cited case and statute must be looked up and confirmed
☐ Shown where to look up citations
☐ Shown the court's AI disclosure order or rule, if one applies
☐ Noted the filing deadline from the papers they were served, without calculating it
☐ Gave at least one referral for legal advice
☐ Made no prediction about the outcome

Knowledge check

Answer all five, then check the key below.

A litigant asks whether the case their chatbot cited will help them win. What can you do?
1. Read the case and tell them whether it helps
2. Show them how to look up the case and confirm it exists
3. Tell them chatbots are unreliable and to remove all citations
Which of these is legal information?
1. Telling a tenant to file a counterclaim
2. Explaining what the court's rules say an answer must include
3. Choosing which defense a tenant should raise
A draft is well written and has no obvious errors. What is the right next step?
1. Tell the person it looks ready to file
2. Still point out that every citation and fact must be checked
3. Suggest stronger wording for the main argument
A litigant asks which AI program is best for court papers. How should you respond?
1. Recommend the tool other litigants seem to like
2. Warn them away from free tools
3. Explain that you can't recommend a product, and point to court resources
Your court has an order requiring disclosure of AI use in filings. What should you do?
1. Show the person the order and let them read it
2. Tell them disclosure will not affect their case
3. Advise them to skip disclosure if they edited the draft themselves

Show answer key

B. Confirming a case exists is information. Judging whether it helps is advice.
B. Describing what the rules require is general and applies to anyone.
B. Polished drafts can still contain invented citations or wrong facts.
C. Naming or warning against a product can read as an endorsement or opinion.
A. Point to the rule itself. Predicting its effect, or advising how to comply in their case, is advice.

What comes next

This module covers the conversation. The next four cover the checking itself, with sample filings and practice exercises:

Checking citations and authorities in AI-drafted filings
Catching placeholders, blank fields, and exhibit mismatches before filing
Preparing a litigant to testify without scripting the substance
Finding and applying court AI disclosure orders and local rules

Modules 2 through 5 are included with VoxJurisPath organizational licenses. Each uses the same verification tools your staff and clients work with in VoxJurisPath, so training and practice happen in one place.

Sources

Talk to us about org licenses