Skip to content
DEMOinvented matters · nothing you type is stored or sent
LawyerSwipe Firms

The review record for AI findings

The AI found it. A licensed lawyer has to stand behind it.

For a partner at a Colorado family-law firm whose associates use AI. If the AI gets a figure or a quote wrong, it is your license on the line.

Check each quote against the exhibits. Sign, modify or decline, and keep a hash.

A real matter? Check it on your device.

Your own Colorado case? Count your dates.

Today and next

Today: invented matters to try, and your own matter checked on your device. Nothing is stored. Next: one Colorado family-law firm runs one real matter through it and tells us what breaks.

This site counts three steps, started, finished and carried on, from its own server log. What you type or decide is never in the count.

The workbench

The AI's quotes, checked against the exhibits. Your decision, hashed.

FINDING
Invented people and documents
The matter

The exhibits

Deciding words in red

What the AI's finding rests on

Quotes checked

The finding, as it would leave the office

    A finding, not a filing: it goes nowhere until a lawyer at the firm signs it. The checks read the quotes. The judgment is the lawyer's.

    Try to sign it. Then try Modify.

    Your own matter

    Paste a real finding. Get back a hashed record.

    The same checks, run on your exhibits. Nothing leaves this device: no upload, nothing saved. Close the tab and it is gone.

    What the AI says, then |, then the exact words from the exhibits it rests on.

    Paste, then press Check.

    To change a statement, edit it and press Check again. A signed record then says it was changed. Only a licensed lawyer at the firm should sign.

    Where it sits

    Not the tool that finds things. The record of who stood behind what it found.

    Your discovery tool produces the finding

    The tool your firm already uses does the reading and the drafting. LawyerSwipe does not draft, and your files never leave your device.

    A lawyer at the firm decides

    Sign, modify or decline, with the exhibit's words beside each statement. Declining is a complete answer.

    The record is what the other side can check

    A hash of the exhibits and a hash of the decision: which lawyer, which documents, which finding, when. Not a confidence score.

    Why the record has to exist

    Courts are counting AI-invented content. Family law is on the list.

    Every figure below links to its source.

    2,095

    court decisions found with hallucinated content from generative AI, updated Sept 28, 2026.

    Charlotin database
    1,208

    of those were filed by people without a lawyer. 829 were filed by lawyers. 137 were family law.

    Same database, by party and field
    17–33%

    of answers from leading legal-research AI tools were wrong or misleading in a Stanford test.

    Stanford RegLab
    92%

    of the serious civil legal problems of low-income Americans got no or too little legal help.

    Legal Services Corporation, 2022

    “When people are in court without a lawyer, bad things happen.”

    Indiana Chief Justice Brent Dickson, on family and civil courts where 60 percent of family law litigants had no lawyer in 2013 (IAALS)

    In August 2026 a lecturer with no lawyer won a Fair Work Commission case against Macquarie University on his own written submissions (Gregory Baker v Macquarie University [2026] FWC 3054). The decision does not say who drafted them. A review record would.

    From July to August 2026, three other people without lawyers lost. One filed a list of 38 exhibits the court inferred were AI-made and did not exist; the case ended without reaching its merits (Ba v Sterling Parts Australia [2026] FedCFamC2G 1245). One kept filing AI submissions that argued the wrong date after the test was explained twice, and was ordered to pay A$1,230 (Australian dollars) of the employer's costs (Sadnan Khan v Aldi [2026] FWC 3144). One had an AI sheet headed “Safe Answers” at his feet in the witness box; his claim failed on the merits, and the tribunal gave little weight to statements it could not tie to their authors (Amponsem v Calvary Administration Ltd [2026] NSWCATAD 229). A check catches the first. A named lawyer reading the record catches the other two.

    Firms will keep using AI on discovery. What is missing is a record of which lawyer checked each finding, against which documents, before it was filed.

    How it decides

    No model in the checks. No made-up percentages.

    The machine checks

    A quote passes only if its words are in the exhibits, word for word, and its numbers, dates and weekdays match. A quote that is not there is flagged as invented. A term the finding leaves out is flagged too.

    The machine does not decide

    A statement that goes beyond the exhibits' words, such as whether a transfer was dissipation, is marked as the AI's reading. The lawyer judges it.

    A lawyer signs, modifies or declines, and the record is hashed

    The record carries a SHA-256 of the exhibits and of the decision, made in your browser. There is no confidence score: none has been measured on legal drafts.

    Sources

    1. AI Hallucination Cases Database, Damien Charlotin, updated Sept 28, 2026. damiencharlotin.com/hallucinations
    2. Magesh, Surani, Dahl, Suzgun, Manning, Ho. Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools. Journal of Empirical Legal Studies, 2025. Stanford RegLab
    3. Legal Services Corporation, The Justice Gap, 2022. justicegap.lsc.gov
    4. IAALS, High Numbers of Self-Represented Litigants Left Underserved by Family Courts, citing Indiana's State of the Judiciary (2013). iaals.du.edu
    5. Gregory Baker v Macquarie University [2026] FWC 3054 (Commissioner Crawford, 12 August 2026). austlii.edu.au
    6. Ba v Sterling Parts Australia Pty Ltd [2026] FedCFamC2G 1245 (Judge Symons, 1 July 2026). austlii.edu.au
    7. Sadnan Khan v Aldi Pty Ltd [2026] FWC 3144 (Deputy President Easton, 19 August 2026). fwc.gov.au
    8. Amponsem v Calvary Administration Ltd trading as Calvary Health Care & Anor [2026] NSWCATAD 229 (Senior Member Little and General Member Maher, 5 August 2026). caselaw.nsw.gov.au
    9. American Bar Association, Formal Opinion 512, Generative Artificial Intelligence Tools, 2024. americanbar.org