What do AI subclass categories mean?

A scan result classifies the document as ai, mixed, or human. The subclass object goes one step further and says what kind of AI use that classification refers to. The current default model is 2025-07-23-base.

Under ai, predicted_class is one of:

• pure_ai (AI Generated): text written directly by an AI language model, with no meaningful human editing.

• ai_paraphrased (AI Paraphrased): AI-generated text that was rewritten, either by a tool or by hand, to disguise its origin. A high-confidence result message reads: "We are highly confident this text was originally AI, but rewritten by AI or a human."

Under mixed, predicted_class is one of:

• concatenated (AI-Human Mix): human-written text combined with AI-generated text in the same document.

• polished (AI Polished): human-written text that was refined or edited by an AI tool. A high-confidence result message reads: "We are highly confident that this text was originally written by a human and improved using AI."

A human classification leaves subclass empty. The document-level class already covers it. Confidence scores and thresholds come back on each response. They are not fixed numbers, so the zeros below are only placeholders.

A real response includes only the object that matches the document. These are the two shapes:

{
  "subclass": {
    "ai": {
      "predicted_class": "pure_ai",
      "result_message": "…",
      "confidence_score": 0,
      "confidence_category": "high",
      "class_probabilities": {
        "pure_ai": 0,
        "ai_paraphrased": 0
      },
      "confidence_scores_raw": {
        "identity": {
          "pure_ai": 0,
          "ai_paraphrased": 0
        }
      },
      "confidence_thresholds_raw": {
        "identity": {
          "pure_ai": { "reject": 0, "low": 0, "medium": 0 },
          "ai_paraphrased": { "reject": 0, "low": 0, "medium": 0 }
        }
      }
    },
    "human": {},
    "mixed": {
      "predicted_class": "polished",
      "result_message": "…",
      "confidence_score": 0,
      "confidence_category": "high",
      "class_probabilities": {
        "polished": 0,
        "concatenated": 0
      },
      "confidence_scores_raw": {
        "identity": {
          "polished": 0,
          "concatenated": 0
        }
      },
      "confidence_thresholds_raw": {
        "identity": {
          "polished": { "reject": 0, "low": 0, "medium": 0 },
          "concatenated": { "reject": 0, "low": 0, "medium": 0 }
        }
      }
    }
  }
}

Only the subclass that applies to the document is filled in. A document classified as ai populates subclass.ai. A document classified as mixed populates subclass.mixed.

The rest of the response is documented at https://gptzero.stoplight.io/.

Detection is not infallible. For a record of how a document was written, use GPTZero Authorship, including the Google Docs Writing Report in the GPTZero browser extension.