Your Clerk Used AI. Now What? The Cite Check Every Small Firm Needs

On Oct. 22, 2026, a federal appeals court set aside a judge's order and sent the case to a different judge, saying the orders contained invented facts and a case that doesn't exist; the judge had blamed a law clerk's use of AI. The fix for a small firm isn't banning AI. It's a cite list on every filing, where the lawyer who signs opens every link, confirms each case exists and says what the filing claims, checks the facts against the record and initials the list. Our podcast's Episode 3 also covers why clients shouldn't discuss their case with AI chatbots, what one billing company's data says about flat fees, and who should write your intake script.

Where does this come from?#

This guide is the written companion to Episode 3 of our podcast, "Your Clerk Used AI. Now What?". The episode covers legal AI news from Oct. 19 to 22, 2026. Every fact here comes from the sources listed in the episode's show notes, and we link the primary documents where we can, such as the court opinion and the agency decision. The advice sections are practical suggestions for solo and small firms, not rules. This isn't legal advice, and your state's rules control.

If you'd rather listen, the episode page has the audio and the full transcript.

What happened with the judge and the law clerk?#

On Oct. 22, 2026, the U.S. Court of Appeals for the Fifth Circuit, one of the federal appeals courts, vacated an injunction a federal judge had issued. Vacated means set aside by a higher court. An injunction is a court order requiring someone to do, or stop doing, something.

The appeals court then ordered the case reassigned to a different judge. It said the judge's orders contained invented facts and a nonexistent case. The judge had blamed those errors on a law clerk's use of AI. A law clerk, often a recent law graduate, works for a judge, helping with research and drafting.

You can read the Fifth Circuit's opinion, No. 25-60496 (PDF) yourself, or our short news item on the decision.

Blaming the clerk didn't save the order, and it didn't keep the case with that judge.

Is this a one-off?#

No. A day earlier, on Oct. 21, the Federal Labor Relations Authority, a federal labor board, said 35 of the 42 legal or record citations in a Health and Human Services brief were fictitious, erroneous or irrelevant. It said the agency may have used a generative AI tool. Generative AI is software that writes new text, such as a draft brief, in response to a prompt.

The FLRA decision is public (PDF), and we covered it in our news item on the labor board's finding.

Neither case involved a small firm. Both involved a document that went out with errors in what it cited.

What does this mean for a small firm?#

Picture a five-lawyer firm. There's no law clerk. There's a paralegal or a junior associate drafting, maybe with AI. And the signature on the brief belongs to the lawyer who files it.

The episode's take: the lesson from both cases isn't that the software failed. It's that nobody checked. That's a process problem, and a process is something a small firm can fix.

What this means for you: If anyone at your firm drafts with AI, the person who signs needs a fixed, written check on every citation before anything gets filed.

Why not just ban AI drafting?#

It's a fair instinct, and the episode takes it seriously before rejecting it. A ban feels clear. But a ban doesn't add a check. People keep using the tools quietly, and now you don't even know AI was involved in the draft on your desk.

A check works whether or not AI touched the document. As the show put it: a ban says "don't." A checkbox says "who." In other words, a checkbox names the person who checked.

How does a cite list work?#

A cite list is a simple document that travels with every filing. Here's the routine from the episode:

1. Whoever drafts builds the list. Every case cited in the filing goes on it, each with a working link to the opinion.
2. The signing lawyer opens every link. Not a sample. Every one.
3. The lawyer asks two questions per case. Does the case exist? Does it say what the filing says it says?
4. The lawyer checks the facts against the record. The record is the official file of evidence and filings in the case. The judge's orders in the Fifth Circuit case had invented facts, not only a fake case, so facts need checking too.
5. The lawyer initials the list. That makes it clear who verified the filing.

That catches a case that doesn't exist, and the quieter error of a real case cited for something it doesn't say.

Does the firm need a written AI policy?#

The episode says yes, and it can be one page. It suggests the policy answer which tools are allowed, what gets verified and who signs off.

Then build the check into the system people already use. Add a "cite list verified" step to the filing step in your practice management software, the system your firm uses to run matters, calendars and billing. If the box isn't checked, the filing doesn't go out.

What about the ethics rules?#

Check your state's ethics rules and any bar guidance on AI before you roll this out. (Not legal advice.) In practice the takeaway is simple: check the output, and supervise whoever drafts.

Before you file, also check whether your court has a standing order on AI use. A standing order is a court's or judge's standing rules for the cases in front of it.

Can the other side ask for your client's chatbot history?#

Someone is arguing that it can. In a federal defamation case (a lawsuit claiming someone made false statements that damaged a reputation), the plaintiff argues it may subpoena AI providers for the defendant's prompts and chats. A subpoena is a legal demand for documents or testimony; here it would be aimed at the AI companies, which aren't parties to the case.

The plaintiff's theory is that the chats show intent to harm, and that in that case this kind of evidence can lift the cap on punitive damages, damages meant to punish, not just to compensate.

As of Oct. 19, 2026, no court had ruled on it. It's an argument, not a ruling, and we're treating it that way. The reason it matters anyway is simple: a chatbot is not a diary. If your client typed their side of the story into one, someone may come asking for it.

What should you tell clients about AI chatbots?#

Start with one plain-English line in your new-client welcome packet. The episode suggests this wording:

"Don't discuss your case with AI chatbots. What you type may not be private, and it could be sought in litigation."

Then say it out loud at the first meeting, because a line in a packet is easy to skim past.

Should AI chats go on your litigation hold checklist?#

Yes. A litigation hold is the instruction to your own side, such as your client, to preserve evidence once a dispute is likely. A preservation letter is a separate thing: a written demand you send to the other side asking it to keep evidence.

Add AI chat histories to your client hold checklist and to the preservation letters you send the other side, right next to texts and email.

Should every litigator demand the other side's chatbot history?#

No. If intent is part of your case, it may be worth asking about in discovery, the pretrial process where each side can request evidence from the other. Whether to seek it is a strategy call for the lawyer on that matter, not a line to paste into every template.

Privilege and discovery rules for AI chats are unsettled. They vary by court and by state. Privilege here means the legal protection that keeps certain communications, like those between lawyer and client, out of the other side's hands.

What about what your own staff type into AI tools?#

This is the uncomfortable part, and the episode turns the question back on the firm. Before your staff paste client information into a tool, ask whether the tool trains on what you type, meaning the vendor uses your inputs to improve its AI. If it does, talk to the client first.

Know which tools your people use and what those tools keep. Our guide to AI tools for small law firms has a short list of data questions to ask vendors.

Can AI fix the end-of-week timesheet problem?#

Partly, if you treat its output as a draft. The episode's billing segment starts with the Friday afternoon problem: trying to remember what you did on Tuesday. For a small firm, the money leak usually isn't the rate. It's time that never gets entered.

If software drafts your time entries from what you actually worked on, that can help, with one big condition. A draft is not a bill. Here's the habit the episode recommends:

• Each evening, take a few minutes. Approve, edit or delete the day's drafted entries.
• At month end, bill only from approved entries. Your billing person works from those alone and spot-checks the descriptions, so no client sees something vague or embarrassing.

There's a privacy catch. A tool that watches everything you do is also seeing client information. Check how the vendor stores, uses and secures that data before you turn it on. If you're comparing options, our billing and timekeeping listings are a place to start.

Should small firms still bill by the hour?#

In one large vendor data set, hourly billing still wins easily. 8am, a legal billing software company, says 48% of the more than 60,000 firms using its products billed only by the hour in the first half of 2026. Ten percent used only flat fees. A flat fee is one set price for a defined piece of work, no matter how many hours it takes.

The overall number hides a split by practice area. The company says flat fees made up 88% of invoices at traffic firms and 64% at immigration firms.

This is the company's own data about its own customers, so read it as one vendor's view, not a national survey. The source is 8am's press release, and we summarized it in our news item on the billing data.

Why traffic and immigration? The guess on the show was that those matters look much the same every time, and that's a useful test for your own practice.

How do you try flat fees without getting burned?#

Pull last year's matters by type and pick the one or two most predictable. Then write a scope sheet for each: a short document that says what's included, what isn't and what the milestones are. Milestones are the points where part of the fee is earned, such as a set share at filing and the rest at the hearing.

Two warnings from the episode:

• Flat-fee rules vary by state. Check yours before you change how you bill.
• Reasonableness doesn't go away. If AI makes the work much faster, ask whether your fee still feels reasonable to the client.

So a scope sheet isn't something you write once and forget. When the work gets faster, look at the fee again.

Who should write your answering service's intake script?#

A lawyer. If an answering service picks up your phones, whether that's AI, live agents or both, the script decides what gets asked and what gets said on your behalf.

The episode's order of operations:

1. Names first. The caller, the other parties and any opposing counsel (the lawyer on the other side). Those names feed your conflict check, the screening of names against your current and former clients before you take on a matter, so you never take both sides.
2. The story comes after. Once the names clear, the details can follow.
3. Ask only what you need. The script should gather just enough to decide whether to take the matter, and tell callers up front not to share details yet.

Some callers won't like giving names first. Tell them why, and keep it quick.

Supervise the service the way you'd supervise your own staff. If AI answers your phones, check your state's rules on telling callers they're talking to AI. Our phone and answering services listings and conflicts checking listings cover tools for both jobs.

Frequently asked questions#

Did a court rule that AI chats can be subpoenaed?#

No. In the federal defamation case discussed on the show, a plaintiff argues it may subpoena AI providers for the defendant's prompts and chats. As of Oct. 19, 2026, it was an argument, not a ruling. Privilege and discovery rules for AI chats are unsettled and vary by court and by state.

Why did the appeals court take the case away from the judge?#

On Oct. 22, 2026, the Fifth Circuit vacated the judge's injunction and ordered the case reassigned to a different judge. It said the judge's orders contained invented facts and a nonexistent case, errors the judge had blamed on a law clerk's use of AI.

Should my firm ban AI drafting?#

The episode argues no. A ban doesn't add a check, and people may keep using the tools without telling you. A one-page AI policy plus a required cite check before filing addresses the actual failure: nobody checked.

Is this legal advice?#

No. This guide summarizes the sources behind Episode 3 of our podcast and offers general practice suggestions. Ethics, flat-fee, discovery and AI disclosure rules vary by state and by court, so check yours.

Next step#

Listen to Episode 3, "Your Clerk Used AI. Now What?" and read the full transcript. Then draft your firm's one-page AI policy and add the cite check to your filing step. For more episodes, see the podcast page, or get the written version in our newsletter.

This guide is general information, not legal advice. Facts come from the sources listed in the show notes for Episode 3 of our podcast; figures from 8am are the company's own data. Some listings on AI Tools For Law may be sponsored or include affiliate or referral links, and paid placements are always labeled "Sponsored"; see our disclosure. None of the links in this guide are paid placements, and nothing in the episode is sponsored.