Rewrite few-shot prompts to stop placeholder leakage

Fixes prompt instructions and reference examples so models generate fresh answers instead of repeating dummy values. Use it when prompts leak sample names or dates.

You getIdentified Leakage Triggers, revised Prompt Text and verification Check

Prompt

Prompt
Act as a prompt reliability engineer. Review my prompt draft to stop language models from repeating placeholder text, sample company names, or fixed numbers from few-shot examples.

- Replace fixed, repetitive placeholder values in examples with variable, realistic variations.
- Add explicit negative constraints that instruct the model to derive values only from the live user input.
- Flag specific terms in my current text that invite pattern-matching copy errors.

Output format:
- Identified Leakage Triggers: bullet list of phrases or values causing the copy behavior.
- Revised Prompt Text: the ready-to-use prompt with corrected instruction phrasing and diversified examples.
- Verification Check: one specific prompt to run to confirm the fix works.

Here is my current prompt:

Example output

Example for: a lead enrichment prompt echoing Acme Corp and John Doe

Identified Leakage Triggers
  • Repetition of "Acme Corp" and "John Doe" across both examples teaches the model these are default outputs.
  • Static date "2024-01-01" appears in both output templates, causing the model to hallucinate that exact date.
  • Lack of explicit boundary instructions telling the model to output "Not specified" when fields are absent.
Revised Prompt Text

Extract contact details from the provided email signature. Extract only details explicitly present in the input text. Do not invent names, titles, or dates. If a field is missing, write "Unknown".

Example 1:
Input: "Best, Sarah Lin, VP Marketing at Northwind Shipping"
Output: Name: Sarah Lin | Title: VP Marketing | Company: Northwind Shipping | Phone: Unknown

Example 2:
Input: "Marcus Vance | Call me at 555-0192"
Output: Name: Marcus Vance | Title: Unknown | Company: Unknown | Phone: 555-0192

Extract details for this text:

Verification Check

Test with a minimal input containing only an email address like "info@techflow.io" to verify it returns "Unknown" for the name rather than Sarah Lin or John Doe.

Written by AI from this prompt. Your results will vary.

Paste your prompt draft into Claude, ChatGPT, or Gemini. If your prompt includes JSON schemas, ask the model to ensure all field names remain unchanged during the rewrite.

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

Build a state sheet for roleplay characters

Create a persistent persona state sheet that prevents roleplay models from breaking character. Use it for training simulations, scenario games, or creative writing.