Skip to content

AI Detector

Detect AI-generated text using a hybrid approach: statistical analysis (sentence uniformity, vocabulary repetition, AI marker phrases, punctuation diversity) combined with an LLM judge for style classification. Supports text input or PDF/TXT file upload.

Open App

Overview

The AI Detector computes 6 statistical signals that distinguish AI-generated text from human writing, then optionally asks an LLM to analyze writing style. The final score is a weighted combination (60% LLM + 40% statistical when LLM is enabled). Verdicts range from "human" to "ai" based on the combined probability.

Statistical signals: burstiness (sentence length variance), vocabulary richness (type-token ratio), word formality (avg word length), sentence starter diversity, AI marker phrase density, punctuation diversity.

Data Source

Statistical analysis — built-in heuristics. LLM judge — configurable LLM backend for style classification.

Endpoints

POST /ai_detector/scan — Scan text for AI content

Upload text or a file (PDF/TXT) for AI detection. Uses multipart form data.

curl -X POST "https://atomgpt.org/ai_detector/scan" \
  -H "Authorization: Bearer sk-XYZ" \
  -F "text=Artificial intelligence has become an increasingly important tool in modern scientific research. It is worth noting that machine learning methods have shown remarkable potential in materials discovery. Furthermore, the integration of large language models into scientific workflows represents a groundbreaking development." \
  -F "use_llm=true"

Scan without LLM (statistical only, faster):

curl -X POST "https://atomgpt.org/ai_detector/scan" \
  -H "Authorization: Bearer sk-XYZ" \
  -F "text=So I was trying to get ALIGNN to work on my perovskite dataset and it kept crashing. Turns out I had NaN values in 3 of my structures. Took me two days to figure that out!" \
  -F "use_llm=false"

Scan a PDF file:

curl -X POST "https://atomgpt.org/ai_detector/scan" \
  -H "Authorization: Bearer sk-XYZ" \
  -F "file=@paper.pdf" \
  -F "use_llm=true"
Field Type Default Description
text string — Text to analyze (provide text or file)
file file — PDF or TXT file to analyze
use_llm string "true" Use LLM judge ("true" or "false")

Response is streamed as NDJSON (application/x-ndjson) with progressive updates:

  1. {"type": "status", "message": "Running statistical analysis..."}
  2. {"type": "stats", "data": {...}} — statistical scores
  3. {"type": "status", "message": "Running LLM analysis..."} (if use_llm)
  4. {"type": "llm", "data": {...}} — LLM verdict (if use_llm)
  5. {"type": "result", ...} — final combined result

Final result fields:

Field Description
ai_probability Combined AI probability (0–100%)
verdict "human", "likely_human", "mixed", "likely_ai", or "ai"
stat_probability Statistical-only AI probability
llm_probability LLM-only AI probability (null if disabled)
statistical.scores Per-signal breakdown (burstiness, vocabulary, word_length, sentence_starters, ai_phrases, punctuation)
llm_judge.reasoning LLM explanation
llm_judge.signals Per-signal indicators from LLM

Python Examples

import requests

response = requests.post(
    "https://atomgpt.org/ai_detector/scan",
    headers={"Authorization": "Bearer sk-XYZ"},
    data={
        "text": "It is important to note that machine learning has shown remarkable potential...",
        "use_llm": "true",
    },
)
# Parse NDJSON stream
for line in response.text.strip().split("\n"):
    if line.strip():
        import json
        chunk = json.loads(line)
        if chunk.get("type") == "result":
            print(f"Verdict: {chunk['verdict']}")
            print(f"AI probability: {chunk['ai_probability']}%")
            print(f"  Statistical: {chunk['stat_probability']}%")
            print(f"  LLM: {chunk.get('llm_probability', 'N/A')}%")

import requests

with open("paper.pdf", "rb") as f:
    response = requests.post(
        "https://atomgpt.org/ai_detector/scan",
        headers={"Authorization": "Bearer sk-XYZ"},
        files={"file": ("paper.pdf", f, "application/pdf")},
        data={"use_llm": "true"},
    )
import json
for line in response.text.strip().split("\n"):
    if line.strip():
        chunk = json.loads(line)
        if chunk.get("type") == "result":
            print(f"Verdict: {chunk['verdict']} ({chunk['ai_probability']}%)")

AGAPI Agent [WIP]

from agapi.agents import AGAPIAgent
import os

agent = AGAPIAgent(api_key=os.environ.get("AGAPI_KEY"))
response = agent.query_sync("Check if this text is AI-generated: Furthermore, it is essential to note...")
print(response)

References