FAI Legal Research Assistant

Can ChatGPT or Gemini Do Pakistani Legal Research?

General chatbots are useful for explaining a legal idea or structuring an argument. They are a risky way to find a Pakistani citation, because they answer from memory and Pakistani law is thinly represented in their training data. This page shows one real test, what went wrong in it, and how to check any tool.

Test run on 2 October 2026. The question was typed into the web version of Google Gemini (logged out, free mode) and into FAI Legal Research Assistant. The question, word for word: "Give me three Supreme Court of Pakistan judgments, with full citations (reporter, year, page), on bail in non-bailable offences under section 497 CrPC." Every citation was then looked up in FAI's archive of reported Pakistani judgments. This is one run of one question, not a survey, and chatbot answers change from run to run.

What the chatbot (Gemini) returned

Citation givenCase name givenWhat we found
PLD 1995 SC 34Tariq Bashir v. The StateReal. Supreme Court, Tariq Bashir and 5 others v. The State.
PLD 2017 SC 733Muhammad Tanveer v. The StateReal. Supreme Court, Muhammad Tanveer v. State.
PLD 2020 SC 456Imtiaz Ahmed v. The StateWrong case. That citation is Khawaja Salman Rafique v. National Accountability Bureau.

Two of the three citations were real. The third was a real citation attached to the wrong case name, which is the hardest kind of error to spot because the citation itself passes a quick look.

What FAI returned

Citation givenWhat we found
PLD 2021 SC 903Real. Supreme Court, Muhammad Imran v. State.
2024 SCMR 28Real. Supreme Court, Muhammad Usman v. State.
2024 SCMR 1210Real. Supreme Court, Attaullah v. State.

All three citations were real Supreme Court judgments. FAI marked each one "not checked against its headnote: confirm it is on point", and it did not print the case names in its bottom line. We checked that the judgments exist. We did not check that each one decides exactly the point FAI attached to it, so read each judgment before relying on it.

The point is not that one tool is perfect. It is that a citation should come with a way to open the judgment, and that a tool should say when it could not confirm something. A real-looking citation with the wrong name is exactly what checking is for.

  1. Ask for the judgment link or the case title as well as the citation, then open it.
  2. Check each citation against the court website, a law report or a database (see the step-by-step guide linked below).
  3. Compare the case name you were given with the case name at that citation.
  4. Never file a citation you have not opened.

The full method is in How to verify a Pakistani case citation.

For prompts that work, and how to connect ChatGPT or Claude to real Pakistani judgments, read How to use ChatGPT for legal research in Pakistan.

Can ChatGPT do Pakistani legal research?

It can help you understand and organise, but it can produce Pakistani citations that look right and are wrong, so every citation it gives must be checked against the judgment before use.

Is Gemini better or worse than ChatGPT for Pakistani law?

We tested Gemini once on one question and have no clean ChatGPT result for it, so we do not rank them. The test above shows why you check either one.

How is FAI different?

FAI answers from an index of Pakistani statutes and judgments and marks a citation as verified only when the judgment was retrieved from the court record. It still tells you to read the judgment, and it flags what it could not check.

Try the same question yourself

Ask FAI Legal Research Assistant a Pakistani law question. Ten questions a month, free, no card required.

Open the assistant