---
title: "Before you file: a citation verification checklist for AI-assisted work (with the sanctions data)"
url: https://thematteros.com/blog/ai-citation-verification-checklist-for-lawyers
updated: 2026-10-27
author: Raghav Handa
---

# Before you file: a citation verification checklist for AI-assisted work (with the sanctions data)

To verify AI-assisted legal work, confirm that every cited case, statute and quotation exists, says what you claim, and is still good law, by reading it in a trusted source, not in the AI's output. Do this for every authority before filing, record who checked it, and check your court's AI rules. The 12-step checklist below does exactly that.

## Key takeaways

- Treat AI output as an unverified draft: every authority must be read in a primary or trusted source.
- Check three things for each citation: it exists, it says what you claim, and it is still good law.
- Record the checker and the date on the matter before you file.
- Counts of AI-citation cases differ by snapshot and scope, so always cite the date and source.

## How do you verify AI-generated legal citations?

Open every authority in a source you trust, such as an official reporter, your legal research database or the court's own website, and read the relevant passage. Do not use the AI to confirm itself. Then confirm the authority supports the proposition, that quotations are word-for-word, and that the authority has not been overturned or superseded.

The risk is real and public. Courts in several jurisdictions have dealt with filings containing authorities that do not exist, and reports track the growth, as the next section shows. This page is organisational guidance for verification workflow, not legal or ethics advice; check your court's rules and your regulator's guidance. Last reviewed: 30 October 2026.


## The 12-step pre-filing checklist

1. 1. List every authority: Extract every case, statute, regulation, rule and secondary source cited in the draft into a table, one row each.
2. 2. Confirm existence: Find each in an official or trusted database by name and citation. If you cannot find it, treat it as nonexistent until proven otherwise.
3. 3. Check the citation details: Court, year, volume, page and parties must match the real record, not just look plausible.
4. 4. Read the passage: Open the decision or text and read the part relied on, not a summary.
5. 5. Check the proposition: Does the authority actually say what the draft says it says? Note the page or paragraph.
6. 6. Verify quotations: Compare each quotation word for word with the source.
7. 7. Check the status: Use a citator or equivalent to confirm the authority has not been reversed, overruled or superseded.
8. 8. Check jurisdiction and weight: Is it binding or persuasive in the court where you are filing?
9. 9. Check statutes and rules are current: Confirm the version in force at the relevant date.
10. 10. Check the court's AI rules: Look for standing orders, local rules or certification requirements about AI use in your court, and follow them.
11. 11. Second-person review: Someone other than the drafter spot-checks or fully checks the table.
12. 12. Sign off and record: Record the checker, the date, the sources used and any authorities removed, on the matter before filing.


## How many AI-citation cases are there? The data, with dates

Several trackers and articles count cases in which AI-generated false content appeared in court filings. A widely cited tracker is the database maintained by Damien Charlotin. The figures below are as reported in the articles linked in the sources; we have not independently counted them, and the database changes continually, so check the current count at its source before quoting anything.

| Figure | As reported by | Date or period | Scope note |
| --- | --- | --- | --- |
| 37 cases | MSBA article citing the Charlotin database | 2024 | Calendar-year tally |
| 524 cases | MSBA article citing the Charlotin database | 2025 | Calendar-year tally |
| 668 cases | MSBA article citing the Charlotin database | 2026 to 27 July | Year-to-date at that snapshot |
| 1,397 US cases | ABA GPSolo, May/June 2026 | As of 6 May 2026 | US only, cumulative, different snapshot |

Why do the numbers differ? Each one counts a different slice (a calendar year, a year to date, or a cumulative US total) at a different snapshot date. The database is crowd-sourced, so completeness varies. Do not add the rows together or compare across rows as if they were the same measure. The consistent message is the direction: the count has risen year over year.


## What kinds of errors should you look for?

| Error type | What it looks like | How to catch it |
| --- | --- | --- |
| Nonexistent case | A plausible name and citation that no database returns | Search by citation and by party names in a trusted database |
| Real case, wrong details | Right name, wrong court, year or reporter | Compare against the official record |
| Real case, wrong proposition | The case exists but does not say that | Read the passage |
| Invented quotation | Quote marks around words not in the opinion | Word-for-word comparison |
| Outdated authority | Case later overruled; statute amended | Citator and current-version check |
| Wrong jurisdiction | Persuasive authority presented as binding | Check the court hierarchy |

Notice that only the first error is caught by 'does it exist?'. The others are why steps 4 to 7 matter.


## What do courts and bars say about AI use?

ABA Formal Opinion 512 (July 2024) addresses lawyers' use of generative AI and describes existing duties applying to it, including competence, confidentiality, communication and supervision. Some state bars and individual courts have issued their own guidance or orders, and some courts require certification or disclosure. The picture changes quickly and varies by court, so we do not list rules here. Read the primary text of the opinion and your own court's current orders and local rules.

- Legal research with AI: a verification-first guide: https://thematteros.com/resources/legal-research-ai-guide
- Legal AI ethics: a practical guide: https://thematteros.com/resources/legal-ai-ethics-guide
- How to use AI on a matter without losing control: https://thematteros.com/blog/use-ai-on-a-legal-matter


## How do you record verification on the matter?

Keep the checklist table as a record in the matter file. A minimal row set is: authority, source opened, proposition checked, status checked, checker, date. Add a footer with a sign-off line: AI tool used (if any), prompt purpose, verified by, date, court AI rule checked (yes or no).

That one record answers the questions a partner, insurer, court or regulator may ask: what was used, who checked it and how. It also makes the next filing quicker, because authorities you have already verified can be reused with their status re-checked.


## Where MatterOS fits

We make MatterOS, so weigh this accordingly. The Research library holds notes linked to matters, and the work review and sign-off feature records that a second person has reviewed work. You can paste the checklist table into a matter page. MatterOS does not verify citations, does not connect to legal databases or citators, and cannot detect AI-generated errors; the checking is a human task.

- MatterOS: the Notion operating system for lawyers: https://thematteros.com/product
- The Matter Method: https://thematteros.com/method
- Pricing: one payment, no subscription: https://thematteros.com/pricing
- Work review and sign-off: https://thematteros.com/features/work_review_signoff
- Legal research notes in Notion: https://thematteros.com/features/research


## FAQ

### How do I check if a case cited by AI is real?

Search for it by name and citation in an official reporter, your legal research database or the court's website. If you cannot find it, treat it as nonexistent. Then read the passage to confirm it supports the proposition, and run a citator check on its status.

### Can I use ChatGPT or another AI for legal research?

Lawyers do use AI for first drafts and issue-spotting, but every authority must be verified in a trusted source before reliance or filing. Professional-conduct duties of competence, confidentiality and supervision still apply. Check your regulator's guidance and your court's rules on AI use.

### How many lawyers have been sanctioned for AI fake citations?

No single reliable figure exists. Trackers count cases, not lawyers, and numbers vary by date and scope; one article reported 668 cases for 2026 through 27 July, and an ABA article cited 1,397 US cases as of 6 May 2026. Check the current database and cite the snapshot date.

### Do courts require disclosure of AI use in filings?

Some courts and judges have issued orders or local rules requiring disclosure or certification of AI use, and many have not. Requirements change and vary by court. Check the standing orders and local rules of every court where you file.

### Who should sign off on AI-assisted work?

Someone other than the drafter should check the authority table, and the responsible lawyer remains accountable for the filing. Record the checker, date, sources used and any removed authorities on the matter before filing, so you can show your process if asked.

## Sources

- [MSBA: confronting the hard numbers behind lawyers, AI and phantom precedents](https://www.msba.org/site/content/News-and-Publications/News/General-News/Confronting_the_Hard%20Numbers_Behind_Lawyers_AI_and_Phantom_Precedents.aspx)
- [ABA GPSolo (May/June 2026): suspect hallucination](https://www.americanbar.org/groups/gpsolo/resources/magazine/2026-may-jun/suspect-hallucination-when-generative-ai-finds-perfect-case)
- [ABA Business Law Today: ABA ethics opinion on generative AI](https://www.americanbar.org/groups/business_law/resources/business-law-today/2024-october/aba-ethics-opinion-generative-ai-offers-useful-framework/)