Record Interrogation Session Setup
The standing-rules prompt: paste this once at the start of a chat session and every answer after it comes cited, sourced, and honest about gaps.
What is an AI record interrogation session setup prompt?
It is a CITE prompt that asks an AI tool to set standing rules for questioning case records, deposition transcripts, or other records. The standing-rules prompt: paste this once at the start of a chat session and every answer after it comes cited, sourced, and honest about gaps. You fill in three fields; the Evidence block requires a page:line, Bates, page, or timestamp cite for every finding so a person can check each one against the record.
Sample prompt
Synthetic data
Synthetic example: The names, dates, case details, and issues below are fictional CasePrompts sample values, not client data or evidence from a real matter. They demonstrate the prompt format, not the accuracy of any AI model's output.
Preview the complete sample prompt
CONTEXT: You are a litigation support analyst assisting counsel's paralegal in a construction defect matter. The attached materials are the case record. The issues that matter are: water intrusion origin; notice to the GC; repair scope disputes. INSTRUCTIONS: I will ask you questions about these materials over an ongoing session. Standing rules for every answer: 1. Answer only from the attached record. Your general knowledge may inform how you search, never what you assert. 2. Distinguish what the record STATES from what it merely SUGGESTS, and label inference as inference. 3. When a question has no answer in the record, the entire answer is "Not found in the record," optionally followed by the closest related material, clearly labeled as such. 4. When sources conflict, present both sides with cites rather than resolving the conflict yourself. 5. End substantive answers with up to three follow-up questions I could ask next, drawn from what you encountered while answering. Acknowledge these rules in one sentence and wait for my first question. TEMPLATE — format your output exactly as follows: Format every answer as: **Answer:** [direct response] **Stated in the record:** [verbatim support with cites] **Inference (if any):** [labeled, with the cited basis for the inference] **Conflicts (if any):** [both sides, cited] **Worth asking next:** [up to three follow-ups] EVIDENCE REQUIREMENTS: For every finding, identify the source document by name and cite page:line for transcripts, timestamps for audio/video, and Bates/page for documents. These rules persist for the entire session — apply them to every answer without being reminded. If a later question tempts you to speculate beyond the record, decline and say what the record would need to contain to answer it. Unless a section above specifies its own marker, write "not found in the record" wherever the provided record does not support a requested finding. Do not infer or fill gaps from outside knowledge. The scope is limited to the requested task, stated issues, and provided record. Do not introduce outside subjects or sources. A human reviewer must verify every material statement and citation against the source record before relying on or sharing the output.
Through the Civil / insurance defense lens
In every answer, distinguish what the record proves from what the plaintiff merely alleges, and say which it is.
Through the Commercial litigation lens
In every answer, cite the governing document first and the testimony second, and flag wherever the two diverge.
Through the Criminal defense lens
In every answer, note whether the supporting material came from the State's production and flag anything referenced in the record that does not appear to have been produced.
Through the Employment litigation lens
In every answer, note when each document was created relative to the challenged decision, and attribute knowledge to specific people with cites.
Through the Family law lens
Attribute every answer to its specific source and author; in this practice, who said it and when is usually the point.
Through the Medical malpractice — defense lens
In every answer, state whether the support comes from the contemporaneous chart or later accounts, and flag wherever the plaintiff's claimed history conflicts with the record.
Through the Medical malpractice — plaintiff lens
In every answer, distinguish what the chart documents from what witnesses later recalled, and flag chart entries that appear altered, late, or out of sequence.
Through the Personal injury — plaintiff lens
Whenever an answer touches damages, state whether the record quantifies them and where; unquantified damages are follow-up work, so flag them.
Through the Probate & estate litigation lens
Attribute every answer to its source and date; in estate disputes, who was in the room and who controlled the money is usually the whole case.
Through the Prosecution lens
In every answer, note material that bears on disclosure obligations, and keep proof and inference visibly separate.
Through the Workers' compensation — claimant lens
Whenever an answer touches benefits, state whether the record documents the supporting wage, status, or authorization evidence and where; missing documentation is follow-up work, so flag it.
Through the Workers' compensation — employer/carrier lens
In every answer, distinguish what the record proves from what the claimant alleges, and note compliance and work-status evidence wherever it appears.
AI output is a starting point, not work product. Verify every citation against the record before you rely on it, file it, or send it.
What you'll fill in
- 01Your role
- 02Case typee.g., "trucking liability," "first-party property," "medical malpractice"
- 03Key issue(s) — 1 to 3e.g., "vehicle speed at impact; brake maintenance; visibility"
Verification checklist
- Mixed / full case record: open each cited page:line, Bates, page, or timestamp location and confirm it supports the finding. The prompt asks: “For every finding, identify the source document by name and cite page:line for transcripts, timestamps for audio/video, and Bates/page for documents.”
- Deposition transcript: open each cited page:line location and confirm it supports the finding. The prompt asks: “For every finding, cite page and line in the form page:line (e.g., 47:12–48:3) and quote the operative testimony verbatim.”
- Medical records: open each cited Bates or page location and confirm it supports the finding. The prompt asks: “For every entry, cite provider, date of service, and the page or Bates number of the source record.”
- Discovery responses: open each cited Bates or page location and confirm it supports the finding. The prompt asks: “For every finding, cite the request/interrogatory number and the page or Bates number of the response.”
- Re-check every item labelled “not found in the record” against the full source; these labels flag gaps or judgment calls for a person to resolve, not proof of absence.
- A human reviewer must verify every material statement and citation against the source record before relying on or sharing the output.
Pro tip This is the one to put in Claude project instructions or a saved Rev prompt — set it once, and the whole matter chat inherits the discipline.
Related verification workflow
Related templates
Admissions & Key Testimony Extractor
Pull every admission, damaging concession, and 'I don't know' on your key issues out of a transcript — with verbatim Q&A and page:line cites.
Demand Letter Facts Section
Draft the facts and damages narrative of a demand from the record itself — every sentence cite-backed, gaps flagged instead of papered over.
AI Deposition Outline Prompt from the Record
Turn the verified case record into a topic-based deposition outline with fact-locking goals, record-cited exhibits, conflict probes, and open-to-closed question sequences.