Compress bloated instructions while preserving accuracy

Trim wordy system prompts down to their essential constraints. Use this to reduce token consumption, lower API latency, and eliminate confusing prompt contradictions.

You getToken Analysis, optimized Instructions and preserved Critical Constraints

Prompt

Prompt
Act as a prompt optimization engineer. Compress the instructions pasted below to reduce token count while keeping 100% of the functional logic, edge cases, and constraints intact.

- Cut conversational filler, redundant guidelines, and repetitive phrasing.
- Replace verbose paragraphs with concise imperative commands and compact bullet points.
- Maintain all negative constraints, formatting rules, and classification boundaries.
- Show the exact reduction achieved in estimated word or token count.

Output format:
- Token Analysis (original length, compressed length, percentage reduction)
- Optimized Instructions (clean copy-ready block)
- Preserved Critical Constraints (bullet list confirming preserved edge cases)

My instructions are below:

Example output

Example for: a 240-word customer support categorization prompt

Token Analysis
  • Original length: 242 words (~310 tokens)
  • Compressed length: 82 words (~105 tokens)
  • Reduction: 66% decrease
Optimized Instructions

Role: Customer Support Ticket Triage Agent.
Task: Categorize incoming tickets and extract metadata.

Rules:

  • Output valid JSON only.
  • Assign exactly one category: Billing, Technical, Shipping, Account, or Other.
  • Set priority: P1 (service outage, critical data loss), P2 (major feature broken), P3 (general inquiry).
  • Extract order_id (digits only) or return null.
  • If sentiment is hostile or threatens legal action, set urgent: true.
  • Never provide customer responses or explanatory text outside the JSON object.
Preserved Critical Constraints
  • Fixed enum values for ticket categories.
  • Objective criteria for priority assignment.
  • Strict exclusion of conversational conversational filler.
  • Boolean flagging for legal/hostile sentiment escalation.
Written by AI from this prompt. Your results will vary.

Paste your long prompt or system instruction right after the final line in ChatGPT, Claude, or Gemini. Afterwards, ask the model to rewrite it in YAML format if you need to minimize tokens further.

Make it repeatable

Keep using this prompt without pasting it in every time.

Keep exploring

More in AI Tools & Prompt Engineering
AI Tools & Prompt Engineering

Restructure instructions to eliminate attention drift

Reorganize complex, bloated instructions into a primacy-recency structure so models never ignore mid-prompt rules. Use this when your prompts fail to respect buried negative constraints.