Page:Line or It Didn't Happen: Making AI Cite the Record
A dangerous legal-AI failure is confident, uncited fabrication. Here is how to require page:line support in a prompt and review the result.
One dangerous AI failure in legal work is a plausible statement the record never supports. Clean, confident prose can carry that error into a summary, draft, or brief without making the invention obvious.
The fix is not simply a better model. It is a requirement you can put in the prompt and then verify in the output: page:line or it didn’t happen.
The rule
The prompt should require every factual claim about the record to point to where it lives — 47:12, 112:3–113:8. It should instruct the model to flag unsupported claims and return “not in the record” when the source does not answer the question. Those instructions make the result easier to check; they do not guarantee the model followed them.
A polished summary is easier to demo than a disciplined one. Either way, every citation still has to be checked against the record.
How to require it in a prompt
You don’t need a special tool — you need the prompt to require it:
- Give it the record as the only source of truth. “Answer only from the transcript provided.”
- Demand citations inline. “After every factual statement, cite the page:line. No citation, no statement.”
- License the non-answer. “If the transcript does not address the question, respond exactly: not in the record.”
- Ask it to quote, then characterize. Pulling the verbatim line before summarizing makes unsupported characterization easier to catch.
This is the backbone of the CITE method, and it is the reason CasePrompts exists: copy-ready litigation prompts that state page:line requirements explicitly, so you are prompting like you are assigning a task to a junior associate — not Googling. Because the requirement lives in the prompt, it can be adapted across models, but the behavior and each resulting citation still have to be verified.
A citation requirement makes model output easier to audit; it does not make the model authoritative. The prompt is one control. Professional review remains the gate before consequential work is used.
For the full review workflow, open the guide to AI deposition summaries. The CasePrompts case study links to the transcript Q&A prompt.