Skip to content
Flagship · fine-tuned detection model

See exactly which sentences read as AI

Not a single blunt score. A fine-tuned three-class model reads every sentence on its own — flagging raw AI-generated text and AI-paraphrased writing — then rolls it up into one document verdict you can defend.

  • Paste or PDF/DOCX ≤ 20 MB
  • Per-sentence probability
  • Free plan included
app/ai-detector
manuscript-draft.pdfanalyzed

The rapid advancement of artificial intelligence has fundamentally transformed the landscape of academic research.99.4%

Furthermore, the integration of machine learning methodologies facilitates unprecedented analytical capabilities.99.1%

Our lab measured the effect across three cohorts between 2021 and 2023.

This paradigm shift necessitates a comprehensive reevaluation of established frameworks.94.8%

Highly AI

Composition

Raw AI-generated61%
AI-paraphrased26%
Organic human writing13%

four-stage pipeline

Extract
Detect
Similarity
Report
3-class
Human · AI-generated · AI-paraphrased
≤ 20 MB
PDF & DOCX, or paste raw text
Per sentence
Individual probability on every line
Background
Long documents run as tracked jobs

// sentence-level evidence

Evidence, sentence by sentence

Every sentence is classified on its own and highlighted by class — so a reviewer can see precisely which lines were flagged, and why, instead of trusting one opaque percentage.

AI-generated

Reads as raw output from a large language model.

AI-paraphrased

AI text revised by a paraphraser or word-spinner.

Human

Consistent with organic human writing — left unmarked.

The rapid advancement of artificial intelligence has fundamentally transformed the landscape of academic research.99.4%

Furthermore, the integration of machine learning methodologies facilitates unprecedented analytical capabilities.99.1%

Our lab measured the effect across three cohorts between 2021 and 2023.

This paradigm shift necessitates a comprehensive reevaluation of established frameworks.94.8%

Selected sentence

probability 99.2%class AI-paraphrased18 words

// document verdict

One clear verdict, banded by threshold

Per-sentence probabilities roll up into a single word-weighted document score, mapped to one of three plain-language verdicts.

0%20%60%100%
0 – 20%

Essentially Human

The document reads as organic human writing.

20 – 60%

Moderately AI

A meaningful share of sentences look AI-written.

60 – 100%

Highly AI

Most of the document is flagged as AI.

// how it runs

From upload to shareable report

Long documents run as tracked background jobs through a four-stage pipeline — leave the page and a live pill tells you the moment results are ready.

  1. 01

    Extract

    Clean text is pulled from your PDF or DOCX — or taken as pasted text — then split into sentences, skipping references, headers and tables.

    manuscript-draft.pdf

    142 sentences extracted

  2. 02

    Detect

    The fine-tuned three-class model scores each sentence — human, AI-generated or AI-paraphrased — with its own probability.

  3. 03

    Similarity

    Optionally, each passage is cross-checked against live academic databases for DOI-linked source overlap. See the Similarity check →

    • Chen et al. — Learning Analytics8.4%
    • Osei & Park — AI in Education6.1%
  4. 04

    Report

    Download the highlighted PDF and replay the full interactive result any time — every run is saved to your history automatically.

    highlighted-report.pdf

    highlights · scores · sources

    Download

// FAQ

Questions, answered

Everything about how the AI Detector reads your document.

  • Instead of one blunt percentage, every sentence is classified on its own — human, AI-generated or AI-paraphrased — and highlighted by class. The per-sentence probabilities then roll up into a word-weighted document verdict, so you can see exactly which lines drove the result.

// start detecting

Run your first document free

Paste text or drop a PDF and get a sentence-level breakdown in minutes. No card required.