Your AI Action Plan

Create a research brief on [TOPIC] for [AUDIENCE]. Include: current landscape, 5 key facts, 3 risks, 3 open questions, and source links for every claim that affects a decision.

Review this research draft. Split it into: source-backed facts, reasonable inferences, unsupported claims, and questions to verify. Then rewrite the summary so unsupported claims are removed or clearly caveated.
Draft: [PASTE DRAFT]

Research [OPTIONS / VENDORS / TOPIC], compare them against [CRITERIA], and produce a recommendation memo. Include a table, tradeoffs, risks, source links, and the decision I should make if the priority is [COST / SPEED / QUALITY / RISK].

Create a research intake template for [ROLE / TEAM]. It should capture: decision to support, scope, time period, must-use sources, sources to avoid, comparison criteria, output format, approval owner, and caveats required before sharing.

Clean this data before I enter it into [SYSTEM]. Return a table with columns: [COLUMNS]. Standardize dates, names, categories, and phone/email formatting. Add a final column called Review needed for anything uncertain. Do not invent missing values.

Raw data:
[PASTE]

Map this incoming form to my database fields. Required destination fields: [FIELDS]. Validation rules: [RULES]. Return a field mapping, transformations needed, and exception cases that should stop for human review.

Extract records from this input and prepare them for [SYSTEM]. Return: 1. clean records table, 2. duplicate warnings, 3. missing required fields, 4. values that need human review, 5. a short change log.

Input:
[PASTE]

Create a data-entry QA playbook for [PROCESS]. Include: required fields, allowed formats, duplicate checks, sensitive-data warnings, examples of good records, examples of records to reject, and a final human review checklist.

Summarize this meeting.
Notes or transcript: [PASTE]
Return: 1. short summary, 2. decisions made, 3. action items with owner and due date, 4. risks, 5. follow-up message draft.

Review this meeting transcript and create a follow-up note. Separate exact decisions from discussion points. Flag anything unclear. Transcript: [PASTE TRANSCRIPT].

Convert this meeting transcript into project tasks. For each task include owner, due date, dependency, priority, and a follow-up email paragraph. Transcript: [PASTE].

Create a meeting memory template for [TEAM / CLIENT]. For each meeting, capture decisions, action items, repeated themes, unresolved questions, stakeholder commitments, and items to revisit next time.

Write a [TYPE - cold outreach / follow-up / proposal / status update] email.
From: [YOUR ROLE] at [COMPANY]
To: [RECIPIENT ROLE] at [THEIR COMPANY]
Context: [1-2 sentences of background]
Goal: [what you want them to do]
Tone: [professional / friendly / direct]
Length: [short = 3 sentences / medium = 1 short paragraph / full = structured email]

Write a weekly marketing performance report.
Period: [DATE RANGE]
Metrics to include: [list your KPIs]
Highlights: [what went well]
Issues: [what underperformed and brief reason]
Next week priorities: [3 bullet points]
Audience: [manager / team / client]
Tone: factual, no fluff. Use bullet points for metrics, short paragraphs for narrative.

Turn these rough notes into a clear [EMAIL / STATUS UPDATE / REPORT].
Audience: [WHO WILL READ IT]
Purpose: [DECISION, UPDATE, REQUEST, OR ESCALATION]
Raw notes:
[PASTE NOTES]

Return:
1. Suggested subject line
2. Short summary
3. Main message in my tone: [DIRECT / WARM / EXECUTIVE]
4. Action items with owners and dates
5. Risks or open questions

I'll share 3 examples of emails I've written. After reading them, identify:
1. My typical sentence length and structure
2. Words or phrases I use often
3. My tone (formal / casual / direct / warm)
4. Things I never say

[PASTE EMAIL 1]
[PASTE EMAIL 2]
[PASTE EMAIL 3]

Now write a [TYPE] email using my style. Here's the context: [CONTEXT]

Summarize this compliance-sensitive document for human review. Return: key facts, deadlines, obligations, missing information, ambiguous language, and questions for the responsible reviewer. Do not give legal, tax, or employment advice.

Document:
[PASTE]

Turn this policy or form instruction into a checklist. Include required fields, approvals, deadlines, evidence needed, common errors, and escalation triggers. Mark anything that needs a qualified human decision.

Text:
[PASTE]

Prepare a reviewer packet for this document. Return: 1. plain-English summary, 2. evidence table with source excerpts, 3. missing information, 4. risk questions, 5. reviewer decision log template. Keep all final decisions blank.

Document:
[PASTE]

Create a reusable compliance-review prompt for [ROLE]. It must require AI to: summarize only, cite source text, flag uncertainty, list missing information, avoid final legal/tax/employment decisions, and produce questions for the qualified reviewer.

N/A — document management / template-driven.

You are drafting a first pass of a legal document for attorney review. Do NOT provide legal advice. Every citation must be labelled 'VERIFY — attorney must confirm against primary source' — never fabricate or assume case law exists.

Inputs:
- Document type: [motion / brief / letter / affidavit]
- Firm template skeleton: [PASTE OR LINK]
- Jurisdiction and court: [DETAILS]
- Case facts: [PASTE]
- Key arguments from the supervising attorney: [LIST]
- Firm voice / tone notes: [PASTE]

Return:
1) A draft in the firm's voice matching the template skeleton.
2) A list of every citation used, each flagged 'VERIFY'.
3) Assumptions you made that the attorney must confirm.
4) Open research questions.

Polish the attached legal draft in the firm's voice without changing any citation, holding, or argument of record. Return marked-up output: original text, revised text, and rationale per edit.

Inputs:
- Lexis+ AI (or Westlaw Precision AI) generated draft: [PASTE]
- Firm voice guide: [PASTE OR LINK]
- Page limit or word budget: [NUMBER]
- Attorney preferences: [LIST]

Return:
1) Edited draft inline with change markers.
2) Reasons for edits (tone, concision, active voice, etc.).
3) Anything that looks citation-shaped but needs attorney verification (pinpointing, parenthetical accuracy).
4) Suggested structural moves if still over the page/word limit.

You are a citation checker for one legal document. Return JSON only. Do NOT assert the document is citation-accurate — only flag extraction and consistency issues.

Inputs:
- document_text
- jurisdiction
- allowed_citation_sources: [Westlaw, Lexis, state reporters, etc.]

Return:
{
  "extracted_citations": [{"cite": "...", "context": "...", "type": "case|statute|regulation|secondary"}],
  "parenthetical_consistency": [{"cite": "...", "issue": "..."}],
  "likely_hallucinations": [{"cite": "...", "reason": "..."}],
  "missing_pinpoint": ["..."],
  "reviewer_questions": ["..."]
}

Do not confirm a citation is real — attorneys verify in Westlaw or Lexis. Only flag structural problems.

N/A — manual chronology maintenance.

You are building a case chronology for attorney review. Keep entries strictly grounded in the provided exhibits. If a date or actor is ambiguous, mark 'UNCLEAR — verify'.

Inputs:
- Exhibit notes (bullet form): [PASTE]
- Parties of interest: [LIST]
- Matter type: [litigation / transactional / regulatory]
- Known date range: [RANGE]

Return a chronology table:
| Date | Event | Parties | Document / Bates ref | Source confidence |

Below the table include:
- Gaps or conflicts to resolve.
- Questions for the attorney.
- Exhibits referenced that appear incomplete.

Draft a fact memo based strictly on the attached chronology. Do NOT add facts not in the chronology. Return narrative plus open questions.

Inputs:
- Chronology exported from [Relativity aiR / Everlaw]: [PASTE]
- Matter theory of the case: [PASTE]
- Opposing party's likely argument: [PASTE]
- Desired memo length: [WORDS]

Return:
1) Narrative fact memo grounded in chronology entries only.
2) Events supporting our theory.
3) Events supporting the opposing argument.
4) Events unexplained in the current record.
5) Questions for attorney follow-up.

You are extracting chronology events from one document. Return JSON only. Do not extrapolate beyond the text.

Inputs:
- document_text
- bates_reference
- matter_parties
- matter_theory (for tagging relevance only)

Return:
{
  "events": [
    {
      "date": "YYYY-MM-DD or 'unclear'",
      "parties": ["..."],
      "action": "...",
      "bates": "...",
      "relevance_tags": ["..."],
      "confidence": "low|medium|high"
    }
  ],
  "unresolved_references": ["..."],
  "verify_before_use": true
}

Any event with confidence below 'high' is flagged 'verify_before_use: true'.

N/A — template-based.

You are drafting a discovery request set for attorney review. Do NOT include FRCP or state-rule citations unless you can clearly name and quote the rule — otherwise mark 'attorney to add local citation'.

Inputs:
- Case facts: [PASTE]
- Legal issues / causes of action: [LIST]
- Jurisdiction and rule set (FRCP / state): [DETAILS]
- Privilege concerns already raised: [LIST]
- Desired scope (broad / narrow): [CHOICE]

Return:
1) Interrogatories mapped to each issue (numbered).
2) Requests for production mapped to each issue.
3) Requests for admission where useful.
4) Objection categories the responding party is likely to raise.
5) Notes on proportionality and local rule caveats that the attorney should confirm.

Draft objections and responses to each discovery request below. Every objection must include the legal basis label (relevance, burden, privilege, proportionality, etc.) and a short rationale.

Inputs:
- Incoming discovery requests: [PASTE]
- Westlaw research notes on scope: [PASTE]
- Matter context: [PASTE]
- Known privileged documents: [LIST]

Return:
1) Request-by-request objection summary.
2) Proposed response language.
3) Documents that look responsive but likely privileged (for attorney review).
4) Items to confirm with the client before serving.

You are running a privilege flagger over one document. Return JSON only. Do not assert privilege — attorneys decide.

Inputs:
- document_text
- bates_reference
- client_counsel_domains (emails/names that suggest attorney-client communication)
- work_product_indicators (litigation labels, draft markers, attorney annotations)

Return:
{
  "privilege_signals": [
    {"type": "attorney_client|work_product|common_interest|joint_defense", "evidence": "...", "confidence": "low|medium|high"}
  ],
  "responsive_signals": ["..."],
  "reviewer_action": "withhold|review|produce_pending_review",
  "attorney_questions": ["..."]
}

Any 'withhold' suggestion requires attorney sign-off. Never recommend production.

N/A — manual docket tracking.

You are extracting court deadlines from one input. Do NOT compute final dates — paralegal calendars them after rule verification. Return a structured table plus a warning list.

Inputs:
- Source text (scheduling order / local rule / standing order): [PASTE]
- Court and jurisdiction: [DETAILS]
- Matter type: [DETAILS]

Return table:
| Event | Trigger | Days | Rule reference | Notes |

Below the table include:
- Ambiguous triggers that paralegal must clarify.
- Counting conventions the attorney must confirm (calendar vs. business days, weekends, holidays).
- Rules citation format attorneys should verify.

Draft a weekly docket reconciliation memo from the inputs. Keep the tone factual. Group by matter and flag anything materially different from last week.

Inputs:
- Current docket export from [Aderant CompuLaw / BEC Legal / CalendarRules]: [PASTE]
- Prior-week export for delta comparison: [PASTE]
- Firm risk rules (e.g. dispositive motions to senior attention): [PASTE]

Return:
1) This-week-due deadlines by matter.
2) Next-week deadlines for planning.
3) Changed deadlines (new vs. moved vs. cancelled).
4) Risk flags per firm rules.
5) Confirmations requested from each attorney.

You are extracting deadlines from one incoming court document. Return JSON only. Do not compute dates — return triggers and day counts.

Inputs:
- document_text
- case_caption
- jurisdiction_rules_ref
- existing_docket_summary

Return:
{
  "extracted_deadlines": [
    {"event": "...", "trigger_text": "...", "days": 0, "day_type": "calendar|business", "rule_cited": "...", "confidence": "low|medium|high"}
  ],
  "docket_conflicts": [{"event": "...", "existing_entry": "...", "issue": "..."}],
  "attorney_review_required": boolean,
  "reviewer_questions": ["..."]
}

Any deadline with confidence below 'high', or any dispositive deadline, sets attorney_review_required to true.

N/A — manual eDiscovery review.

You are clustering documents by fact pattern and relevance. Do NOT output privilege assessments — those stay with the attorney.

Inputs:
- Document excerpts (redacted of privileged content): [PASTE]
- Matter issues: [LIST]
- Case theory: [PASTE]
- Known key actors / terms: [LIST]

Return:
1) Suggested fact-pattern clusters with supporting excerpt references.
2) Documents that don't fit any cluster.
3) Items the attorney should review for privilege or relevance.
4) Missing documents the record implies (gaps).

Draft an exhibit-list memo and attorney briefing packet from the inputs. Focus on documents labelled key or hot by the review platform.

Inputs:
- Export of 'key' documents from [Everlaw AI / Relativity aiR]: [PASTE summary]
- Case theory: [PASTE]
- Upcoming hearing type: [DEPO / motion / trial]
- Attorney preferences: [LIST]

Return:
1) Proposed exhibit list (label, Bates, one-line summary, theory it supports).
2) Documents held back for completeness but not primary exhibits.
3) Open follow-ups (witnesses, missing corroboration).
4) Packet table of contents for the attorney.

You are tagging one document for issue, relevance, and actors. Return JSON only. Do NOT set privilege or production flags.

Inputs:
- document_text_redacted
- bates_reference
- matter_issues: [...]
- key_actors: [...]

Return:
{
  "issue_tags": [{"issue": "...", "evidence_snippet": "...", "confidence": "low|medium|high"}],
  "actor_mentions": ["..."],
  "document_type": "email|memo|contract|notes|other",
  "reviewer_follow_ups": ["..."]
}

Never output privilege or production recommendations. Attorney retains those calls.