Home › Prompts › Claude › High-volume ticket classifier with JSON output

High-volume ticket classifier with JSON output

Classify and extract fields from support tickets or feedback at scale with a fixed label set and strict JSON, tuned for Haiku 5.5's effort setting.

ClaudeData analysisRuleStack original

Written for Claude Haiku 5.5 (claude-haiku-5-5) · Updated 2026-10-18

Note: Review and test before use. Provided as is, without warranty. See Terms.

Classify the ticket below. Use only these labels.

Categories: [category list, e.g. billing, bug, feature_request, account_access, shipping, other]
Urgency: low | normal | high (high = [your definition, e.g. customer cannot use the product or mentions a deadline within 24 hours])

Extract:
- product_area: one of [areas] or null
- order_id: the order number if present, else null
- sentiment: negative | neutral | positive
- summary: one sentence, no customer names

Return only JSON matching this shape: {"category": "", "urgency": "", "product_area": null, "order_id": null, "sentiment": "", "summary": ""}

Ticket:
<ticket>
[ticket text]
</ticket>

Replace everything in [square brackets] with your own details.

Settings & parameters

• Model claude-haiku-5-5 with output_config.effort low for bulk runs; compare against medium on a labeled sample
• Use structured outputs (a JSON schema in output_config.format) so every reply parses
• Batch large backlogs through the Message Batches API for lower cost

Variations

• Multi-label: let category be an array of up to two labels.
• Add "needs_human": true|false with your routing rule.

Tips

• Keep a 100-ticket labeled set and measure accuracy per label before switching models or effort levels.
• If the classifier must call a lookup tool (for example, to fetch the order), keep adaptive thinking on; Anthropic notes Haiku 5.5 can skip needed tool calls when thinking is off and a JSON format is required.