Grokzilla AI Detector

{
  "type": "json",
  "example": {
    "band": [
      0.5334,
      1
    ],
    "skill": "ai_detector",
    "method": "stylometry",
    "caveats": [
      "Detection is probabilistic and should not be used as the only evidence that a text was or was not written by a person.",
      "ai_likelihood is a stylometric score from 0 to 1, not a calibrated probability. The band widens when there is less text and is not a statistical confidence interval.",
      "A low score means these surface patterns are weak. It does not prove a person wrote the text.",
      "Weights were chosen on a small set of public-domain prose, poems, and synthetic assistant-style paragraphs. Fixture accuracy is not a field accuracy."
    ],
    "signals": {
      "burstiness": 0.3542,
      "line_count": 1,
      "verse_like": false,
      "hapax_ratio": 0.8841,
      "sentence_count": 8,
      "contraction_rate": 3.2967,
      "type_token_ratio": 0.7582,
      "stock_phrase_hits": 3,
      "average_word_length": 4.3846,
      "repeated_ngram_rate": 0,
      "stock_phrases_found": [
        "it's worth noting",
        "a testament to",
        "in conclusion"
      ],
      "sentence_length_mean": 11.375,
      "stock_phrase_per_100": 3.2967,
      "lexical_features_used": true,
      "punctuation_diversity": 4,
      "line_length_uniformity": null,
      "sentence_length_variance": 16.2344
    },
    "verdict": "possible_ai_influence",
    "sections": [
      {
        "result": "no_signal",
        "section": "rhythm",
        "evidence": [
          "8 sentences, sentence-length CV 0.35 (AI <= 0.34, human >= 0.55)",
          "rhythm is in the overlap range where human and prompted-AI text both sit"
        ],
        "section_weight": 0.1
      },
      {
        "result": "ai_sign",
        "section": "vocabulary",
        "evidence": [
          "tell-phrase weight 2.4 (2.64 per 100 words, 138-entry list)",
          "tells: it's worth noting, a testament to, in conclusion",
          "lexical diversity MTLD 105 (informational, not scored)",
          "surprisal mean 3.21, variance 2.15, spike 0.40; rare-word share 0.033; MATTR 0.834; Zipf slope deviation 0.63",
          "tell phrases are dense; possible AI influence (secondary signal)"
        ],
        "section_weight": 0.4
      },
      {
        "result": "no_signal",
        "section": "voice_structure",
        "evidence": [
          "ai cues: hedging (2 hedges)",
          "human cues: none",
          "first person, names, dialogue and sensory detail are not counted either way"
        ],
        "section_weight": 0.3
      },
      {
        "result": "no_signal",
        "section": "resolution",
        "evidence": [
          "rhythm vs vocabulary: no_overlap (no_signal / ai_sign)",
          "vocabulary vs voice_structure: no_overlap (ai_sign / no_signal)",
          "voice_structure vs rhythm: no_overlap (no_signal / no_signal)",
          "tell phrases present but not confirmed by two sections; secondary signal only"
        ],
        "section_weight": 0.2
      }
    ],
    "classifier": {
      "verdict": "uncertain",
      "model_kind": "logistic_regression",
      "threshold_ai": 0.585,
      "model_version": "2026-10-13",
      "probability_ai": 0.362,
      "threshold_human": 0.3233
    },
    "word_count": 91,
    "content_type": "opinion",
    "rubric_score": 0.4,
    "tell_phrases": [
      {
        "count": 1,
        "phrase": "it's worth noting",
        "tell_weight": 0.9
      },
      {
        "count": 1,
        "phrase": "a testament to",
        "tell_weight": 0.9
      },
      {
        "count": 1,
        "phrase": "in conclusion",
        "tell_weight": 0.6
      }
    ],
    "ai_likelihood": 0.7822,
    "possible_ai_influence": true
  }
}