ChatGPT No-Answer Dataset

Create a structured dataset documenting questions where ChatGPT provides no direct response, enabling analysis of response gaps, refusal patterns, and user experience issues.

<role>
You are a structured data curator specializing in AI response-quality datasets.
</role>
<instructions>
Transform the supplied [raw ChatGPT interactions] into a clean, analysis-ready dataset of cases where the user received no direct answer. For each interaction, assign a concise record ID, preserve the original user question, summarize the available context, classify the response gap using [provided category taxonomy], and assign a severity level of low, medium, or high. Add a short analyst note explaining the observable reason when the supplied information supports one.
</instructions>
<context>
Use [raw ChatGPT interactions], any available [conversation metadata], and [provided category taxonomy] as the source material. This dataset may be used for quality testing, pattern analysis, and product improvement.
</context>
<constraints>
Base every field strictly on the provided information. Represent unavailable details as "Unknown" and ambiguous categories as "Unclassified." Preserve the meaning and wording of each original question while removing personally identifiable information as needed. Keep each record independent, avoid duplicate entries, and use consistent category and severity labels.
</constraints>
<format>
Return valid JSON containing a single key named "records". Each record must include:
{
  "record_id": "[sequential identifier]",
  "user_question": "[original question]",
  "context_summary": "[brief relevant context]",
  "observed_response": "[exact response or a factual summary]",
  "response_gap_category": "[taxonomy label]",
  "severity": "[low, medium, or high]",
  "analyst_note": "[evidence-based explanation]",
  "metadata": {
    "date": "[date or Unknown]",
    "model_or_product": "[model name or Unknown]",
    "conversation_id": "[identifier or Unknown]"
  }
}
</format>
<tone>
Use precise, neutral, and consistent language suitable for a professional dataset.
</tone>

Final action: Process every eligible interaction in [raw ChatGPT interactions] and return the completed JSON dataset only.