AI Assistant Prompts & Task Templates

Welcome — make AI work the way your team expects

This library gathers tested prompt templates and lightweight workflows for recurring collaboration tasks. Use these as starting points: fill the expected inputs, run the assistant, apply the quality checks, and keep the human-in-the-loop that prevents silent errors or bias. Each template shows the intent, required inputs, expected output format, verifications, and a short sample conversation you can paste into an assistant session.

How to use this library

1. Choose a template that matches your task (meeting summary, draft email, research brief, etc.).
2. Populate the required inputs and organizational variables (audience, tone, deadlines, links).
3. Select model and parameters (e.g., prefer higher temperature for creative drafts, lower for extraction).
4. Run the prompt and apply the listed quality checks; if checks fail, use the suggested follow-up prompt to correct or verify.
5. Record approved outputs and decisions—consider storing results with the platform's submission capability for traceability.

Common variables (examples)

• org_name: Your organization or team name
• audience: e.g., Executives, Engineering, Customers
• tone: e.g., concise, friendly, formal
• context_links: URLs, docs, or ticket IDs the assistant should consider
• date_range: meeting dates, research windows, deadlines

Template: Meeting Summarization (Action-focused)

Purpose: Produce a concise summary that highlights decisions, action items (owner + due date), and open questions.

Required inputs: meeting_transcript or meeting_notes, participants list, meeting_date, preferred_length (short/medium/long).

Prompt template (replace placeholders):

"Given the meeting transcript below, produce: 1) a 3-line executive summary, 2) a bulleted decision list (with rationale if given), 3) action items with owner and due date (format: Owner — Action — Due), and 4) unresolved questions. Use the participant list to disambiguate speakers. Keep output under [preferred_length]. If you cannot find an owner or date for an action, mark it as 'owner TBD' or 'due TBD'. Transcript: {{meeting_transcript}} Participants: {{participants}} Meeting date: {{meeting_date}}"

Expected output format: clearly labeled sections (Executive Summary, Decisions, Action Items, Open Questions) with short bullets and explicit owners/dates.

Quality checks:

• Spot-check 2–3 timestamps or quotes against the transcript for accuracy.
• Confirm each action has a named owner or is marked 'owner TBD'.
• Flag any paraphrases that might alter meaning and ask for human review when high-risk decisions are summarized.

Sample conversation (validation):

User: Summarize the meeting and extract action items from the transcript below. Assistant: [Delivers labeled sections as requested]

Template: Decision-Item Extraction

Purpose: Pull discrete decisions from documents or meeting notes and capture the rationale, impacted teams, and next steps.

Required inputs: source_text, decision_context (optional: why this matters), impact_scope (teams/functions).

Prompt template:

"From the source text, identify each explicit decision. For each decision, provide: Decision statement; Rationale (1–2 sentences); Impacted teams; Required next steps (if any). If the text implies a decision rather than stating it, label it 'implied' and include the supporting excerpt."

Quality checks: verify quoted excerpts supporting each decision, confirm scope with a subject-matter reviewer when impacts are broad.

Template: Draft Communications (Email / Announcement)

Purpose: Generate a first-draft email, announcement, or message tailored to audience and tone.

Required inputs: purpose, audience, key_points (3–5), sender_name, call_to_action.

Prompt template:

"Write a [tone] email from {{sender_name}} to {{audience}} whose purpose is {{purpose}}. Include these key points: {{key_points}}. End with the call to action: {{call_to_action}}. Keep the subject line under 80 characters and provide two subject line options."

Quality checks: check for factual accuracy, unintended policy or legal language, acronyms explained for the audience, and a clear call to action. Have a human reviewer confirm tone and compliance before distribution.

Template: Research Synthesis (Evidence-first brief)

Purpose: Summarize findings across sources, surface conflicting evidence, and produce short implications and recommended next steps.

Required inputs: list_of_sources (URLs or text), question_or_scope, audience, length_limit.

Prompt template:

"Synthesize the evidence in the following sources to answer: {{question_or_scope}}. For each claim, cite the source (use source IDs). Provide: 1) 3–5 key findings with source citations; 2) any conflicting evidence and why it conflicts; 3) practical implications; 4) recommended next steps or research gaps."

Quality checks: verify citations exist and match claims; surface methodological limits (sample size, date, bias). When stakes are high, route the brief to a domain expert for verification.

Template: Onboarding Pack (Role-specific)

Purpose: Create a tailored onboarding checklist and 30/60/90-day plan for a new hire or role.

Required inputs: role_title, team_goals, access_needed, key_contacts, org_policies (links).

Prompt template:

"Build a 30/60/90 day onboarding plan for a {{role_title}} on {{team_name}}. Include initial setup tasks (access, tools), early learning goals, first-project suggestions, and a simple success metric for 90 days. Provide a checklist and suggested calendar items."

Quality checks: ensure access items match current IT provisioning processes and that contact names/titles are up to date.

Guardrails, verification & handoff patterns

• Human-in-the-loop: Always assign a named reviewer for outputs that influence decisions, compliance, or public-facing communications.
• Two-step handoff: Assistant generates → Human verifies (fact-check + tone) → Assistant revises if requested.
• Bias & safety checks: Ask the assistant to list assumptions and potential bias sources; have a human reconcile or annotate them.
• Audit trail: Store final approved outputs and the prompt used (consider the platform's submission capability) for traceability and future improvement.

Prompt hygiene & tuning tips

• Be explicit about format and length.
• Include examples in the prompt when you need a precise style.
• Use model temperature and max tokens to control creativity vs. precision.
• Iteratively test using short validation conversations; keep prompts as small, focused, and deterministic as possible for extractive tasks.

Customization checklist

Before wider rollout, tailor these templates to your organization:

1. Add organization-specific variables (names, channels, policy constraints).
2. Define verification owners and SLAs (e.g., reviewer responds within 24 hours).
3. Choose model presets for each template (model, temperature, max tokens).
4. Decide where approved outputs are stored and who has access.

Next steps & capability recommendations

Consider converting high-value templates into interactive forms that collect the required inputs (organizing variables, transcript upload, key points) and store approved outputs. You can also package this collection as an ownable toolkit so teams can copy and adapt it for their site or business unit.

Image suggestion: "ai assistant prompts library"