No AI. No black boxes. Just rules.

Detect AI text without using AI

Statistical analysis, pattern matching, and rule-based correction. You can audit every decision and explain every result.

Try it free

Try it instantly, no sign-up required

or install the Chrome extension
12
Detection signals analyzed
6
Correction steps applied
0
Machine learning models used
How it works

Three steps. Full transparency at every stage.

01

Detect

Four independent analyzers score readability, vocabulary diversity, sentence burstiness, and stylometric patterns. Each returns evidence, not just a number.

→
02

Correct

Six sequential steps fix spelling, style issues, AI phrases, passive voice, vocabulary repetition, and sentence uniformity. Each change is logged.

→
03

Verify

Detection re-runs on the corrected text. If any metric got worse, those changes roll back automatically. You see exactly what changed and why.

Export results as a branded PDF detection report to share or archive.
The difference

Other detectors use the same AI they claim to detect.

We use math. Every rule is a regex, a formula, or a statistical threshold you can read and audit.

× ML-based detectors

Trained neural classifiers that can't explain their decisions
False positives on polished human writing with no recourse
Opaque models updated without notice
Detection only (no correction pipeline)
Accuracy degrades as LLMs evolve

✓ Eyes on Text

Rule-based analyzers that show every signal and threshold
Per-sentence evidence so you see exactly what was flagged
Open rules you can audit, tune, and extend
Full correction pipeline with verification
Statistical patterns are LLM-agnostic and work on any model
Tunable detection with presets for common use cases and fine-grained control for power users
Under the hood

What the detection pipeline actually measures.

Readability variance

AI text clusters in a narrow Flesch band. We measure per-paragraph variance. Humans are inconsistent; machines aren't.

Vocabulary diversity

Type-token ratio, Shannon entropy, and Zipf compliance. AI text underuses rare words and overuses common ones.

Sentence burstiness

Standard deviation of sentence lengths plus a curated list of AI-telltale phrases. Uniformity is the clearest AI signal.

Stylometric fingerprint

Burrows' Delta on function-word frequencies. A proven forensic technique adapted for AI-vs-human attribution.

Chrome extension

Right-click any text on any webpage to analyze it instantly. No copy-paste, no tab-switching — detection lives where you read.

Install extension
Pricing

Simple plans. No per-word pricing games.

Full control: Balanced Strict Lenient Academic Marketing Genre presets and threshold tuning available on Pro and Custom plans.
Monthly Annual Save 15%
Starter
Free
For trying it out. No account needed.
  • 3 analyses per day
  • Detection only
  • Up to 500 words
  • Web interface
Get started
Professional
$19 / seat / month $16 / seat / month
$192 / year
For writers, editors, and content teams.
  • 100 analyses per month
  • Detection + correction (Beta)
  • Up to 10,000 words
  • PDF detection reports
  • Genre-tuned detection presets
  • Adjustable detection thresholds
  • Correction verification scores
  • Re-run analysis
Start free trial
Enterprise
Custom
For platforms and publishers at scale.
  • Unlimited analyses per team
  • Detection + correction (Beta)
  • No per-document word limits
  • PDF detection reports
  • API access
  • Custom detection thresholds
  • Dedicated support
Contact sales

Text integrity you can
actually understand.

No neural networks. No opaque classifiers. You can trace every signal, correction, and decision back to a rule.

Try Eyes on Text free