The method
How NotAI works
Every page asks one question: is this a human? NotAI answers with behavioral AI detection and human verification.
01 · NotAI Pixel
The signals that give agents away
A single async pixel detects AI agents and bots from how a visitor moves the mouse, scrolls, types, and navigates. Behavior only: text capture is a separate opt-in. Select a signal to compare a real person against an automated agent.
Humans move in arcs and make constant micro-corrections. Agents travel in straight lines at a fixed speed.
Real reading speeds up, slows down, and stops to re-read. Automated scrolling advances at one mechanical rate.
Typing arrives in uneven bursts and hesitations. Generated text lands all at once with no keystrokes behind it.
People return with familiar habits across visits. Automation repeats the same path, page after page, visit after visit.
Behavioral signals are one layer. NotAI also checks for automation fingerprints (CDP), headless browser tells, extension interference, WebGL anomalies, tampering, and honeytraps that normal use has no reason to touch. Read the pixel install guide.
02 · Text Monitor
Authorship as it happens
The standalone @isnotai/text-monitor package captures the pauses, corrections, and rhythm of real composition as it unfolds. Text detectors judge the finished essay and guess. NotAI verifies the writing as it happens: the evidence is the process, not the prose. See how the three approaches compare.
Keystroke analysis
Captures timing between keystrokes, the hesitations and bursts of human typing. A large paste with no supporting keystroke activity is one of several signals that feed the confidence score.
Authorship orbs
A reviewer view that shows the composition of each submission as color-coded orbs across five authorship classes: Imported, Single Drafted, Substituted, Quoted, and Original.
Session Player
Plays back a submission from first keystroke to final submit. Evidence a reviewer can replay before anyone is accused.
Works with any editor
Attaches to plain textareas, contenteditable elements, and rich-text editors. Native adapters for Quill and TinyMCE. Universal content diffing covers any other editor. Install surfaces include generic websites, custom apps, and Canvas LMS.
03 · The pipeline
How a verdict is reached
Whichever signal source you run, the path from raw behavior to a confidence score is the same four steps.
Capture
A single async script measures movement, scroll rhythm, keystroke timing, and navigation. No interaction required, nothing that blocks page render.
Profile
Signals combine into a profile of the session, so the same mechanical habits stand out when automation comes back.
Score
The profile is scored against models of human and automated behavior to weigh every signal together.
Verdict
A confidence score plus the evidence behind it: which signals fired and the full session timeline. The score informs; a human reviewer decides.
See exactly what is and is not collected in our privacy policy and AI transparency notice.
04 · Choose your path
Two questions. One platform.
Run the Pixel to verify visitors as they browse. Add Text Monitor wherever people write: enabled through the Pixel, or standalone on pages without it.
Are they browsing?
Use the NotAI Pixel
For page visits, navigation, and general bot detection. It runs without user interaction and reads behavior signals only.
- Identify bots from mechanical movements
- Flag AI agents completing coursework
- Protect forms from automated submissions
- Spot repeat automation across pages and visits
Are they writing?
Use Text Monitor
For essays, forms, comments, and reviews: any text input where authorship matters.
- Flag essays that were pasted in, not written in the editor
- Verify student assignment authorship
- Feed the reviewer UI: authorship orbs and the Session Player
05 · In your LMS
The verdict lands where you grade
A single install covers every Canvas course. Every verdict flows into the gradebook and the reviewer UI: authorship orbs, the Session Player, and a confidence score sit next to each submission, so an academic integrity conversation starts from evidence, not suspicion. Tiers are published under academic pricing.
Start verifying real work
Run NotAI on your LMS or any website in minutes. Institutions get flat annual pricing; everyone else starts at $49/month.