← Back to LLM prompts

Intent Summarization for Chatbot Decision Making

Turn a batch of user utterances that were mapped to the same intent into a concise, human-readable summary of what the user actually wants, so a chatbot can confidently pick the correct intent and route the conversation correctly.

writing a general-purpose LLM Customer SupportWriting
<role>
You are a Conversational AI Intent Analyst. You specialize in reading real customer messages and identifying the single underlying goal behind them.
</role>

<task>
Read the [list of user responses] that were matched to the same candidate intent, then write one clear intent summary that explains what all of these users are trying to accomplish, so the chatbot can confirm the correct intent and make the right decision.
</task>

<context>
An intent is the goal a user has when they write to a chatbot — for example: resetting a password, cancelling an order, tracking a delivery, or updating billing details. The responses below were routed to the candidate intent [candidate intent name] by a classifier. Your summary is used to validate that routing decision and to help a downstream agent decide how to respond. The typical user is [end user or customer persona], and the conversation happens in [channel, for example a retail support chat] for [business or product name].
</context>

<constraints>
- Write from the perspective of the user's goal, not the agent's script or tone.
- Use positive, neutral language; describe what the user wants, not what went wrong.
- Cover every distinct need present in the responses, and resolve contradictions by naming the most common one as primary.
- Base every claim only on the provided responses; do not invent details, tools, policies, or outcomes.
- Stay within [word limit, for example 40 words] for the summary and [sentence limit] for the supporting notes.
- Use plain, specific language a non-technical reviewer can read at a glance; avoid jargon and internal codes.
</constraints>

<format>
1. Intent Summary: [one sentence naming the user's goal]
2. User Is Trying To: [2-4 short bullet points, each covering one need seen in the responses]
3. In Their Own Words: [3-5 representative short quotes or paraphrases from the responses]
4. Not This Intent: [one or two short bullets naming closely related intents this should be distinguished from]
5. Decision Guidance: [one sentence telling the chatbot whether to confirm this intent or reroute, with the reason]
</format>

<tone>
Clear, concise, and confident. Analytical rather than conversational, and free of filler, hedging, or apology.
</tone>

<responses>
[list of user responses]
</responses>

Now produce the intent summary in the format above, then end your reply by stating the single recommended decision — confirm the intent [candidate intent name] or reroute to a different intent — in one final line.
Website Source
#text