Text Deepfake Detection
Response schema and example outputs for AI-ForensiX text deepfake / AI-generated content detection model.
Text Deepfake Detection
AI-ForensiX Text Deepfake Detection model analyzes linguistic patterns, perplexity metrics, and burstiness indicators to determine whether a text document is human-written or AI-generated.
It is compatible with both plain text and extracted text from PDF documents.
Response Schema
TextDeepfakeDetectionResult
| Field | Type | Description |
|---|---|---|
label | string ("human" | "ai generated") | Indicates whether the text is human-written or generated by AI. |
perplexity | number - optional | Measures text unpredictability. Lower values often correlate with AI-generated writing. |
burstiness | number - optional | Measures variation in sentence complexity. Higher burstiness indicates more human-like writing style. |
Metric Explanation
-
Perplexity
Lower perplexity → more predictable text → commonly produced by AI models.
Higher perplexity → more human-like variation. -
Burstiness
Measures fluctuations in sentence length + structure.
Humans naturally write with higher burstiness, while AI text tends to be uniform.
Example Responses
Listing : AI-Generated Text Detection Example
{
"label": "ai_generated",
"perplexity": 12.4,
"burstiness": 0.34
}Listing : Human-Written Text Detection Example
{
"label": "human",
"perplexity": 28.7,
"burstiness": 0.71
}