Work-ops Chatbot FAQ Prompt Template

One-shot ChatGPT prompt (Markdown output) that produces deployable, policy-safe chatbot FAQ packs for work-ops support

What You'll Get

Single one-shot prompt template (Markdown-ready) with explicit input slots: product_area, tone, policies_list, escalation_rules, kb_base_url, product_facts, stakeholders, constraints.

Up to 12 prioritized FAQ intent blocks. Each block contains: Intent name; 3 example user utterances; 2-4 diagnosis questions (Ask:); concise canonical response (Answer:) tuned to tone; Empathetic and Firm variants; 1-2 short follow-up prompts; one-line microcopy (<=8 words); a KB link using kb_base_url; 2-4 measurable handoff triggers; one-line private escalation message with required ticket fields; and a one-sentence refusal tied to a policy key.

Global sections: CSV-compatible index table (Intent | KB path | Handoff? | Escalation owner); Facts vs Assumptions (with verification questions); 'When to hand off' prioritized rules; synthesized refusal rules; and a Confidence & fallback policy (hand off if <70% confidence or repeated question).

Strict formatting & ingestion rules: metadata comment lines () above each block, customer-facing lines â‰¤280 chars, no absolutes, and redaction prompts for PII.

Safety guardrails: explicit refusal templates referencing provided policy keys, instructions not to perform refunds/account changes unless an allowed-action policy is supplied, and guidance to update escalation owners/SLA targets before deployment.

How to Use

1Provide required inputs before running: product_area, tone, policies_list, escalation_rules, kb_base_url, product_facts. If any input is missing the prompt will request clarification.

2Run the prompt in ChatGPT and copy the resulting Markdown output into a staging KB or bot builder (Intercom/Zendesk/ChatGPT/Gemini).

3Verify the 'Facts vs Assumptions' section: replace each assumption with a confirmed fact or answer the suggested verification question, then rerun if needed.

4Map each FAQ's handoff triggers and escalation message to your internal owners and SLA targets; update the escalation_rules input accordingly.

5Sanity-check customer-facing answers for compliance and tone; perform a small pilot (10-50 tickets) and adjust language or triggers before full rollout.

6Import canonical answers into your help widget and set up monitoring: escalation rates, handle time, and CSAT to validate effectiveness.

Provide conservative policies and an authoritative KB base URL

Always supply an explicit policies_list and kb_base_url before first run. Use conservative escalation_rules (owner, severity, SLA) and mark anything uncertain as an assumption so the template generates verification questions. This reduces risky hallucinations and speeds safe deployment.

Works Best With

Best used with structured inputs and defined policies

Provide explicit policy keys (e.g., NO_SENSITIVE_DATA_SHARING) so refusal lines are tied to exact rules

Supply a KB base URL to generate correct links and avoid invented hosts

Include escalation rules with owners and SLA windows to produce measurable handoff triggers

Give product facts (limits, trial lengths, refund windows) to populate 'Facts' and avoid assumptions

Prompt Template


# Work Ops (Meetings, Docs, Support) Prompt Blueprint

## Goal
- Produce a deployable Chatbot FAQ pack for customer-facing support in a specific product area: canonical answers that resolve intent, include 1-2 follow-up prompts, explicit handoff/escalation triggers, short microcopy, and KB links â€” ready for direct ingestion into bot platforms or support playbooks.

## Inputs
- Context: target product area, typical customer scenarios, known sensitive topics.
- Stakeholders: support leads, escalation owners, content reviewer.
- Constraints: policy rules, allowed actions, maximum reply length, KB base URL.
- Desired tone: primary voice for answers (e.g., friendly, professional, concise).
- Required format: for each FAQ produce Diagnosis questions â†’ Response draft â†’ Next steps.
- If any of the above are missing the prompt will ask clarifying questions before generating content.

## Prompt
Act as a support knowledge author and chatbot copywriter for {{product_area}}. Using the inputs below, produce a canonical Chatbot FAQ pack optimized for deployment in chatbots and help widgets (Intercom/Zendesk/Gemini/ChatGPT). Output MUST be structured so each FAQ entry contains: Intent, Example user utterances, Diagnosis questions, Canonical response draft (concise), Tone-variant (empathetic and firm) drafts, 1-2 follow-up prompts, One-line microcopy (button/tooltip), KB link, Clear handoff triggers (measurable), and a one-line escalation message for agents.

Inputs you must use:
- Primary tone: {{tone}}
- Policy and refusal rules: {{policies_list}}
- Escalation rules and severity mapping: {{escalation_rules}}
- Knowledge Base base URL (use to form KB links): {{kb_base_url}}
- Product facts and limits: {{product_facts}}

Instructions:
1. Identify up to 12 high-priority customer intents for {{product_area}} (prioritize likely frequency and risk). For each intent produce the following block exactly in this order:
   - Intent name
   - Example user utterances (3 short variants)
   - Diagnosis questions (2-4 short, explicit questions to ask the user; label as "Ask:")
   - Canonical response (max 3 sentences; label as "Answer:") tuned to {{tone}}
   - Tone variant: "Empathetic variant:" (1-2 sentences) and "Firm variant:" (1-2 sentences)
   - Follow-ups: list 1-2 suggested follow-up prompts the bot should present (each <12 words)
   - Microcopy: a one-line button/tooltip text (<=8 words)
   - KB link: full link using {{kb_base_url}} plus an intuitive path placeholder (e.g., {{kb_base_url}}/path-to-article)
   - Handoff triggers: 2-4 explicit, measurable conditions that require agent escalation (e.g., "payment failed after 2 attempts", "user requests chargeback")
   - Escalation message: one-line private note for agents including required ticket fields to provide (e.g., "Include order_id, customer_id, screenshots, last 4 of card")
   - Refusal/deflection line: one-sentence refusal if request violates any item in {{policies_list}} (cite the policy key)

2. After all intents, produce:
   - Global "When to hand off" rules derived from {{escalation_rules}} (bullet list, prioritized).
   - Global refusal rules synthesized from {{policies_list}} with exact wording the bot should use (3-5 short rules).
   - A short "Confidence & fallback" policy: if the bot is <70% confident or if user repeats the question twice, hand off with a template message.

3. Formatting & safety:
   - Mark facts vs assumptions at top of the output. For any assumption include a short rationale and suggested verification question.
   - Do not assert policy or legal claims without referencing the named item from {{policies_list}}.
   - Keep customer-facing replies â‰¤280 characters and free of absolutes like "always" or "never".
   - Redact or request redaction of any personal data beyond ticket fields specified in Escalation message.

4. Deliverable constraints:
   - Output in Markdown.
   - For each FAQ entry include machine-friendly keys in a metadata comment line above the block: e.g., <!-- intent:refund; severity:high -->
   - Include a short CSV-compatible index table at the top listing Intent, KB path, Handoff? (yes/no), Escalation owner.

If any input ({{product_facts}}, {{policies_list}}, {{kb_base_url}}, {{escalation_rules}}, or {{tone}}) is missing or ambiguous, ask up to three clarifying questions before generating the pack.

## Output Format
Support: Diagnosis questions â†’ Response draft â†’ Next steps

For each FAQ entry follow this exact order:
- Diagnosis questions (label "Ask:")
- Response draft (label "Answer:") with two tone variants (Empathetic / Firm)
- Next steps: Follow-ups, Microcopy, KB link
- Handoff: Handoff triggers and Escalation message
- Refusal: one-line refusal tied to a policy key

At the top include:
- CSV-compatible index table: Intent | KB path | Handoff? | Escalation owner
- Metadata comment lines for each intent block as specified above

## Guardrails
- Separate facts vs assumptions at the top of the output. Facts = items present in {{product_facts}} or explicit policy items from {{policies_list}}. Assumptions = anything inferred; flag them and propose verification questions.
- Never provide refunds, account changes, or legal commitments in chat without an explicit allowed-action policy key from {{policies_list}}.
- Do not invent KB URLsâ€”use {{kb_base_url}} as the host and append clear path placeholders.
- When handing off, include required ticket fields and a concise private note for the agent.
- If a request touches disallowed topics from {{policies_list}}, use the prescribed refusal line and immediately offer a permitted alternative.
- No policy claims without the named policy key from {{policies_list}}.

## Examples
- Mini decisions/actions excerpt
  - Decision: For "Refund request â€” paid within 14 days" the bot will ask for order_id and payment method, then escalate if payment system shows failed refund attempts.
  - Actions: Owner: Billing Team; Action: attempt refund within 3 business days; Notify customer with templated reply.

Variables

Usage Notes

Not for

â€¢ Generating binding legal, financial, medical, or regulatory advice (use counsel or authorized teams)

â€¢ Direct execution of transactions, refunds, or account changes unless an explicit allowed-action policy exists

â€¢ Replacing human judgment for complex escalations involving compliance or PII beyond provided ticket fields

Do not include

â€¢ Do not include customer PII beyond the ticket fields specified in escalation messages (order_id, customer_id, screenshots, last4 etc.)

â€¢ Do not invent KB URLs â€” use only the provided kb_base_url and append clear path placeholders

â€¢ Do not assert guaranteed outcomes, metrics, or absolute statements (avoid words like 'always' or 'never')

â€¢ Do not reference internal model names, system prompts, or confidential implementation details in customer-facing text

Work-ops Chatbot FAQ Prompt Template

A single, fillable ChatGPT prompt that generates up to 12 prioritized FAQ intents in Markdown â€” each with diagnosis questions, concise canonical answers (â‰¤280 chars), empathetic/firm variants, 1-2 follow-ups, microcopy, KB links (uses kb_base_url), measurable handoff triggers, and a private escalation note.

Turn this into a working asset in minutes.

Import into Workspace to auto-fill variables, adapt output style, and iterate with OctiAI's prompt editor.

Auto-fill & personalize from your inputs

AI-assisted edits: tone, structure, constraints

Attach context so outputs stay consistent

Iterate and save versions

What you'll get

Output format

Markdown

Best used when

When working on template or detailed.

Avoid when

Generating binding legal, financial, medical, or regulatory advice (use counsel or authorized teams)

Made for

What to fill in

Use the guidance below to complete each placeholder with the right level of detail.

Product Area