Technology

Powered by proprietary detection engine90.6% accuracy on the RAID benchmark

Our core detection engine is trained on a large corpus of human and AI-generated text. It achieved 90.6% accuracy on the RAID benchmark, which is a comprehensive evaluation that tests detectors across multiple AI generators, domains, and adversarial attacks.

90.6%
RAID Accuracy
20+
Languages
<5s
Analysis Time

How It Works

From raw text to a detailed verdict in four steps.

1

Text Preprocessing

Input text is tokenized and segmented into sentences. Language is automatically detected for multilingual support.

2

Classification Engine

Each sentence is passed through our internal detection engine, which analyzes writing patterns to distinguish AI-generated text from human writing.

3

Score Aggregation

Per-sentence scores are weighted and combined into a final suspicion score (0.0–1.0) and a document-level verdict: likely AI, possibly AI, uncertain, or likely human.

4

Sentence-Level Results

Per-sentence scores are returned alongside the document-level verdict, letting you see exactly which parts of the text triggered detection.

Our detection engine

At the core of our detection engine is a fine-tuned classifier — a state-of-the-art language model trained specifically to distinguish AI-generated text from human writing:

  • Deep contextual understanding — separates content and positional signals for richer representation of language patterns
  • Precise token-level analysis — captures subtle writing cues critical for detecting AI patterns, even after light editing or paraphrasing
  • Trained on diverse AI output — fine-tuned on text from GPT-4, ChatGPT, Claude, Gemini, Llama, Mistral, and other models across multiple domains

This approach gives our detector the ability to identify AI-generated text reliably while maintaining low false positive rates for human writers.

Our Principles

The values that guide how we build and improve our detection technology.

Accuracy First

We prioritize minimizing false positives. Mislabeling human-written text as AI-generated can have serious consequences, so precision comes before recall in our detection pipeline.

Privacy by Design

Submitted texts are not stored permanently or used for training. We process content for analysis and return results — your writing stays yours.

Transparent Confidence

AI detection is probabilistic, not absolute. We communicate confidence levels clearly so you can make informed decisions about the content you review.

Continuously Validated

We regularly benchmark our models against established datasets and real-world adversarial examples. As AI writing evolves, our detection keeps pace.

About RAID

RAID (Robust AI Detection) is an established benchmark for evaluating AI text detectors. It tests across:

  • Multiple generators — GPT-4, ChatGPT 5.4, Claude Opus 4.6, Gemini 3, Llama, Mistral, and others
  • Multiple domains — news, Wikipedia, academic writing, creative fiction
  • Adversarial attacks — paraphrasing, word substitution, and other evasion techniques

Scoring 90.6% on RAID means our detector reliably identifies AI-generated text even under challenging, real-world conditions.

See It in Action

Paste any text and get an instant AI detection analysis powered by our fine-tuned classifier. Completely free.

Free to use — no account needed.