thinkhard.SI

Prompt repository Customer support & success

Answer a customer question from your help docs and policy

You need an accurate, friendly reply that only uses what your documentation actually says.

Registry no.
SUP-010
Collection
Customer support & success
Laid out for
Any chat model
Score
10 out of 10
Blanks to fill
6
Added

Customer support & successAny chat

The prompt

Score 10 of 10, Excellent

Use it when You need an accurate, friendly reply that only uses what your documentation actually says.

  • {{company}}
  • {{help_docs}}
  • {{policy}}
  • {{customer_details}}
  • {{customer_message}}
  • {{agent_name}}
Full prompt and why it works
Prompt
Write a reply to the customer message below, using only our help documentation and policy. An agent at {{company}} will review and send it, so accuracy matters more than speed: a confident wrong answer creates a second ticket and an unhappy customer.

Help articles, with their links:
"""
{{help_docs}}
"""

Policy that applies:
"""
{{policy}}
"""

What we know about the customer (plan, device, account details): {{customer_details}}

Customer's message:
"""
{{customer_message}}
"""

Shape the reply like this:
- The direct answer in the first one or two sentences. For example, "Yes, you can do that from your account settings. Here's how:" is the kind of opening wanted.
- If they need to do something, numbered steps with one action each, using button and menu names exactly as the help articles write them.
- A link to the most relevant article, if one is given above.
- If they asked more than one question, an answer to each, in the order they asked.

Keep it under 150 words, in plain text, signed "{{agent_name}}". The tone should be friendly and direct, the way a helpful colleague explains something, without stock lines such as "Thank you for reaching out" or "I hope this helps".

Rules:
- Use only what the help articles and policy say. Do not add steps, settings, limits, prices or dates that are not there, and do not promise refunds, credits or exceptions.
- Take only what answers the question; do not paste whole articles.
- If the answer depends on something we do not know, such as which plan they are on, ask one short question rather than covering every case.

If the documentation does not answer the question, or two sources disagree, do not guess. Add a first line starting "FOR THE AGENT:" that says what is missing or which sources conflict, and write only the parts of the reply the documents support.

Variables

{{company}}
Your company or product name.
{{help_docs}}
The help articles that relate to the question, with their links.
{{policy}}
Any policy that applies, such as refunds or account changes. Write "none" if no policy applies.
{{customer_details}}
What you know about the customer that matters here: plan, device, app version, account status.
{{customer_message}}
The customer's message, exactly as received.
{{agent_name}}
The name the reply is signed with.

Why it works

  • What to avoid. The reply uses only what your documents say, so the model cannot fill a gap with a setting or a limit that sounds right but does not exist.
  • Format. The answer comes first, then steps with the exact on-screen labels, which is what a customer needs to fix the problem without writing back.
  • If unsure. Missing or conflicting documentation goes to the agent in a FOR THE AGENT line, instead of turning into a guess the customer acts on.

How to use it

  1. Press Fill in variables and type your details into each blank. What you type stays in this browser.
  2. Copy the filled prompt into the AI tool you already use, such as ChatGPT, Claude or Gemini, or open it in the Grader or the Prompt Tester here.
  3. Before you rely on it, try it on three or four real examples. The Prompt Tester runs them all at once and shows which ones go wrong.
Use it from code

Get this prompt as JSON from the free REST API, with your own text in the blanks, or let an AI assistant fetch it through the MCP server. No key.

curl -s https://thinkhard.si/api/v1/prompts/answer-from-help-docs

With the MCP server added to Claude, ChatGPT, Cursor or VS Code, call get_prompt with {"prompt": "answer-from-help-docs"}, or just ask for it by name.