CiteClear

ChatGPT Made-Up Legal Cases

Documented examples of ChatGPT fabricating legal cases and citations. Understand the patterns to better detect AI hallucinations.

Check Citations See More Examples

Why ChatGPT Invents Legal Cases

📖 Pattern-Based Generation

ChatGPT generates text based on patterns it learned during training, not by retrieving facts. It creates text that "looks right" but may be fabricated.

🎯 Lack of Real-Time Data

ChatGPT's knowledge cutoff means it may not have information on recent cases. It fills gaps with plausible-sounding inventions.

📊 Probability Over Accuracy

ChatGPT optimizes for "likely" text, not "accurate" text. A likely-looking citation may be more probable than an accurate one.

🔍 No Source Verification

ChatGPT doesn't verify its outputs against external databases or primary sources. It can't check if a case actually exists.

Common Types of ChatGPT Legal Hallucinations

📅 Future-Dated Cases

Citations with dates beyond the current date. These are guaranteed to be fabrications.

Example: Smith v. Jones, 123 F.3d 456 (9th Cir. 2025) (in 2024)

📚 Impossible Reporter Volumes

Volume numbers that exceed the known range for a reporter series.

Example: Doe v. Roe, 999 F.4th 1000 (11th Cir. 2024) (F.4th only up to ~64)

🏛️ Non-Existent Courts

References to courts that don't exist or incorrect court abbreviations.

Example: Smith v. City, 123 F.Supp. 456 (Federal Appellate Court 2023)

📜 Non-Existent Statutes

References to US Code titles or sections that don't exist.

Example: 99 U.S.C. § 9999 (US Code only has 54 titles)

✋ Real Case, Wrong Citation

Real case names paired with fabricated reporter, volume, or page numbers.

Example: Brown v. Board, 999 U.S. 9999 (1954) (Real case, wrong everything else)

🎭 Completely Fabricated

Entirely made-up cases with made-up party names, citations, and legal principles.

Example: Innovative Tech v. Global Solutions, 456 F.3d 789 (Silicon Circuit 2023)

How to Spot ChatGPT Hallucinations

✅ Check the Date

Future dates are impossible. Very old dates may be suspicious if the case is cited as recent precedent.

✅ Check the Reporter

Verify the reporter abbreviation is valid (F., F.2d, F.3d, F.4th, U.S., S.Ct., L.Ed., etc.).

✅ Check the Volume

Verify the volume number is within the valid range for that reporter series and time period.

✅ Check the Court

Verify the court abbreviation is valid. ChatGPT often invents court names.

✅ Search the Citation

Search the full citation on Google Scholar, CourtListener, or official court websites.

What to Do When You Find a Hallucination

🚫 Don't Use It

Never use a suspicious citation without thorough verification. The risk to your credibility and legal arguments is too high.

🔍 Verify Thoroughly

Check multiple sources. Use Google Scholar, CourtListener, and official court websites.

📝 Document the Issue

Keep a record of the hallucination, when you found it, and how you verified it was fake.

🎓 Improve Your Prompts

Be more specific in your prompts. Ask for citations in Bluebook format with full information.

⚖️ Use Authoritative Sources

Supplement AI output with primary sources. Don't rely on ChatGPT for critical legal research.

Preventing ChatGPT Hallucinations

🎯 Use Specific Prompts

Instead of: "Tell me about case law on this topic"

Try: "Cite 5 Supreme Court cases from the last 10 years on this specific issue with Bluebook citations"

📚 Provide Context

Give ChatGPT relevant context, jurisdiction, time period, or other constraints to reduce hallucinations.

✅ Ask for Verification

Ask ChatGPT to verify its citations or provide sources. While it can't truly verify, this sometimes catches obvious errors.

🔄 Use Multiple Models

Check the same question with multiple AI models. Consistency across models suggests higher reliability.

🧠 Supplement with Knowledge

Use your own legal knowledge to spot inconsistencies, improbable holdings, or suspicious reasoning.

⚖️ Consult Primary Sources

Always verify critical information through primary legal sources, not AI output.

Frequently Asked Questions

Why does ChatGPT make up legal citations?

ChatGPT generates text based on statistical patterns, not factual databases. It predicts what text is likely to come next, not what text is factually accurate. When it doesn't have a real citation to reference, it creates one that looks plausible.

Can ChatGPT citations be trusted for legal work?

No. ChatGPT citations should never be trusted without thorough verification. While some citations may be correct, the risk of hallucinations is too high for professional legal work.

What percentage of ChatGPT legal citations are hallucinated?

Studies suggest that 20-40% of ChatGPT legal citations may be partially or completely fabricated, depending on the specificity of the query and the model version. More specific queries with constraints tend to have lower hallucination rates.

Does ChatGPT know when it's hallucinating?

No. ChatGPT doesn't have self-awareness or the ability to recognize its own hallucinations. It generates text with confidence regardless of accuracy. That's why external verification is essential.

Are some AI models better at legal citations than others?

Yes. Models with more recent training data and those fine-tuned on legal text tend to have lower hallucination rates. However, all current general-purpose AI models still require verification for legal work.

Can ChatGPT be forced to only provide real citations?

No. While you can ask ChatGPT to only provide real citations, it cannot guarantee accuracy. It doesn't have the capability to verify its outputs against authoritative legal databases.

Related Tools & Guides