AI legal research without filing a case that does not exist
The citations are the risk. Use grounded tools and verify every one against the reporter.
In short
AI legal research is safe only when the model is grounded in a real case database and every citation is verified before filing.
- General chatbots fabricate citations that look correct, with real-sounding parties, judges, and holdings.
- Courts are sanctioning lawyers who file fabricated citations, and the count of such decisions is now in the thousands.
- Use AI to find the argument and the search terms. Confirm the authority yourself.
The failure mode is a citation that looks perfect
AI legal research has one failure mode that matters, and it is not vagueness. A general-purpose model asked for authority will produce citations with plausible parties, a real-looking reporter number, and a holding that fits your argument exactly, none of which exist. That is what got two attorneys sanctioned in Mata v. Avianca in 2023, and it has not stayed rare: Damien Charlotin AI Hallucination Cases database listed 2,041 decisions involving hallucinated material as of 14 September 2026. The practical answer is to separate the two jobs. A model is genuinely useful for shaping an argument, generating search terms, and summarizing a case you supply. Finding the authority belongs to a grounded research platform, and confirming it belongs to you. The habit generalises: see fact-checking AI content and reducing AI hallucinations. It is the highest-stakes corner of AI for legal.
How to do AI legal research safely
Shape the argument with AI, source the authority elsewhere, verify everything.
- 1
Use AI for the shape, not the citation
Ask it to frame the issue, list the elements, and suggest search terms. Those outputs carry no citation risk.
- 2
Source authority in a grounded platform
Westlaw, Lexis, or a tool that retrieves from a real case database. A general chatbot has no reporter to check against.
- 3
Summarize cases you supply
Paste the opinion and ask what it holds. Reading a real document is a different task from recalling one.
- 4
Pull every citation up yourself
Open each case in the reporter. If you cannot find it, it does not exist, however convincing the quote looks.
- 5
Check it is still good law
A real case can be overturned or distinguished. Run the citator; a model has no reliable sense of current status.
- 6
Record what you verified
Note which citations you confirmed and when. Courts increasingly ask, and the record is your answer.
Watch-outs
- A fabricated citation reads exactly like a real one. Plausibility tells you nothing about existence.
- Asking the model whether its own citation is real does not work. Check the reporter.
- Client facts in a prompt are a disclosure. Confirm the tool data terms before you paste any of it.
Common questions
Is AI legal research safe to use?
AI legal research is safe when the model is grounded in a real case database and every citation is independently verified before filing. It is not safe to take authority from a general chatbot, which generates citations that look correct but may not exist.
Why does AI invent case citations?
A language model predicts plausible text rather than retrieving records, so when asked for authority it produces something shaped like a citation. The parties, reporter numbers, and holdings look right because they follow the pattern, not because the case is real.
Have lawyers been sanctioned for this?
Yes, repeatedly. Two attorneys were fined in Mata v. Avianca in 2023 for a brief containing fabricated citations, and the pattern has grown since: Damien Charlotin AI Hallucination Cases database listed 2,041 decisions involving hallucinated material as of 14 September 2026.
What is AI genuinely good for in legal research?
Framing the issue, listing the elements you need to prove, generating search terms, and summarizing an opinion you paste in. Those tasks use reasoning over text you supply, so they carry none of the fabrication risk that comes with asking for authority.
See how you actually work with AI
The AI Skills Quiz scores your real habits, including whether you verify what AI tells you. Free, and it takes a couple of minutes.
Professionals from these companies build their AI skills with Candova
The future belongs to AI‑native companies.
Make sure yours is one.
We make your people AI-fluent so your company becomes AI-native, with Cando, a personal AI trainer for every employee. Start with Candova AI today.
Free to start · Productivity gain guarantee · Built for your real job