Litigation Prompts for Source-Linked Matter Review

Set the record before using a prompt

Replace every bracketed variable. Name the source inventory and version, the exact question and the intended reader. Supply originals or inspectable extracts with stable references. State what is missing. Ask for “not established from supplied material” when the record does not answer the question.

Keep the tool and run record where available. An output with references still needs passage checks. NIST's Generative AI Profile describes risk management and evaluation considerations, including confabulation; this collection is a working aid, not a validated benchmark. 2

Choose the task

Use extraction for candidate rows, verification for a particular assertion, comparison for competing accounts and gap review for a missing fact. Use the new-document and downstream prompts together without treating a proposed change as a completed edit. For briefs and handovers, start with the reviewed record rather than an unexamined first summary.

Witness preparation is limited to organising documents for the lawyer. It must not generate the witness's account. Check the applicable court's current requirements, including any AI disclosure requirements; GPN-AI is one Australian Federal Court reference. 3

Reusable prompts

Purpose: Prepare candidate matrix rows from approved documents. Required inputs: [factual question], [source inventory and version]. Expected output: Issue ID, candidate proposition, attributed source, pinpoint, support limit, contrary material and open check. Review checks: Check quantities, event types, qualifiers and the independent search for adverse material.

Purpose: Test support rather than citation existence alone. Required inputs: [exact assertion], [cited source], [inventory]. Expected output: Existence check, passage meaning, support level, contrary source and corrected wording. Review checks: Open the original passage and surrounding context; inspect the adverse material independently.

Purpose: Locate the precise factual difference between accounts. Required inputs: [account A], [account B], [relevant source set]. Expected output: Shared ground, specific differences, source limits and next checks. Review checks: Confirm the accounts refer to the same event before calling them contradictory.

Purpose: Convert an unresolved proposition into inspectable next checks. Required inputs: [proposition], [source inventory], [known missing records]. Expected output: Known source, missing item, consequence of absence, request and proposed owner role. Review checks: Check collection limits and whether the proposed record is actually known to exist.

Purpose: Connect stated terms with reported performance. Required inputs: [milestone], [agreement version], [event records]. Expected output: Term pinpoint, candidate event, relevant quantity/date, qualification and question. Review checks: Check the full agreement and whether the units of count match.

Purpose: Record a proposed change while preserving the prior position. Required inputs: [new source], [earlier wording and version], [current inventory]. Expected output: Earlier wording, new account, proposed change, reason, limits and affected IDs. Review checks: Check the old version, actual receipt date, reviewer decision and completion separately.

Purpose: Find work products that need inspection after an accepted change. Required inputs: [accepted change and limits], [named work product versions]. Expected output: Affected location, specific check, proposed owner and completion evidence needed. Review checks: Open each output version and record real completion evidence.

Purpose: Organise reviewed facts for a named internal decision. Required inputs: [reader and decision], [reviewed record], [source version]. Expected output: Scope, factual position, competing accounts, gaps, key sources and questions for lawyer. Review checks: Verify every material assertion and ensure the brief answers the named decision.

Purpose: Give the next reviewer current state and owned work. Required inputs: [recipient], [current versions], [open actions]. Expected output: Current position, source access, people, recent changes, open checks and acknowledgement fields. Review checks: Confirm access, real owners, procedural dates and actual receipt of responsibility.

Purpose: Prepare document-based neutral questions for the lawyer. Required inputs: [witness role], [approved source set], [topic]. Expected output: Event, recorded connection, source, competing material and neutral question. Review checks: Lawyer checks duties, permitted AI use and whether each question assumes a fact.

Purpose: Separate transmission, receipt, attendance and later account. Required inputs: [process question], [approved employment packet]. Expected output: Event/document dates, attributed account, pinpoint, limitation and missing record. Review checks: Inspect metadata, addresses, attachments and actual attendance evidence.

Purpose: Define quality checks before running an evaluation. Required inputs: [one task], [approved source packet], [reviewed reference answer]. Expected output: Inputs, expected output, critical failure cases, review steps and measured-effort fields. Review checks: Agree acceptance criteria with the responsible reviewer before any run.

Inspectable editorial tests

We applied the proposition, summary-support, competing-account, evidence-gap, milestone and new-document instructions to the fictional equipment packet during preparation of this collection. The test record preserves the inputs, generated candidate answers, source checks and corrections. These are actual editorial assistant exercises in this preparation session, not Grella product runs, observed customer results or a model-performance benchmark. The equipment record is a connected running example across this library. Reusing it lets you follow the same sources through different review tasks. Apply the source and review checks to the records in your own matter.

One useful check is the flawed assertion that all 12 pilot units arrived by 30 April. DOC-005 exists, but it distinguishes 10 pilot units from two borrowed units and is dated 3 May. A correct support check must reject the broader assertion. Another check is the depot scan: DOC-007 records a delivery status at the depot gate with no identified recipient, so the warehouse-receipt question stays unresolved.

The worked answers in the TXT test record can be checked against the enclosed originals. They are examples, not an acceptance result for your tool or matter. Run your chosen prompt against a suitable approved packet, retain the actual output and use the review checks before adopting any row or paragraph.

The working kit

The complete collection includes all 12 prompts, the editorial test record and the synthetic source packet. Each prompt includes purpose, inputs, expected output and review checks. The complete ZIP contains all files.

Use the evidence matrix, verification guide, new-document register and one-matter pilot for the surrounding review workflow. The existing chronology guide retains its own chronology prompt pack. Contact Grella to discuss your matter workflow.