Understanding, demonstrated

Don't detect AI. Discover understanding.

Cognition has a short, adaptive conversation with each student about their work - then hands you a transcript and a clear picture of what they truly grasp. Evidence you can trust, never an accusation.

Photosynthesis·Defence Live

You wrote that chlorophyll absorbs light. Which colours does it absorb most - and how do you know?

Mostly red and blue. The leaf looks green because green light is reflected rather than absorbed.

Light absorptionStrong grasp

AI detectors guess. Cognition asks.

Detection tools put a probability on a student and call it proof. That breaks trust, punishes the wrong people, and teaches no one anything. We took the opposite path.

The detector approach

  • Guesses from writing style - and is wrong often enough to ruin trust.
  • Flags non-native speakers and neurodivergent students at higher rates.
  • Produces an accusation, then leaves you to defend a number you can’t see.
  • Tells you nothing about what the student actually learned.

The Cognition approach

  • Has a real conversation about the work - no guessing, no black box.
  • Meets every student where they are, in their own words.
  • Gives you a readable transcript and a per-concept picture you can stand behind.
  • Turns an integrity check into a moment of genuine learning.
The evidence

The research is clear: detection doesn't work

Independent studies keep reaching the same conclusion - detectors are unreliable and unfair. Meanwhile, decades of learning science show that asking students to explain themselves is one of the most powerful things you can do.

61.3%

of TOEFL essays written by non-native English speakers were wrongly flagged as AI-generated by popular detectors, while essays by native speakers were rarely misclassified.

Liang et al., "GPT detectors are biased against non-native English writers," Patterns (Stanford), 2023

Withdrawn

OpenAI shut down its own AI-writing classifier in 2023, citing a low rate of accuracy. The makers of the models could not reliably detect their own output.

OpenAI, AI Text Classifier discontinuation notice, 2023

+50%

Retrieving and explaining what you know - exactly what a Cognition conversation asks for - improves long-term retention far more than re-reading. Assessment becomes learning.

Roediger & Karpicke, "Test-Enhanced Learning," Psychological Science, 2006

A student who can explain their reasoning, defend a choice, and answer a follow-up has shown you something no detector ever could: that they actually understand.

Three steps to evidence

From assignment to insight - and never a student account in sight.

01

Create a task

Pick a mode, set the topic, choose how many questions. You get a short class code.

02

Students join with the code

No accounts, no logins. Each student has a focused, adaptive conversation about their work.

03

Read the evidence

A transcript plus a clear, per-concept view of what each student genuinely understood.

One tool, three conversations

Choose the conversation that fits your goal for the class.

Defence

Adaptive questions about an assignment the student submitted - grounded in their own work.

Revision

Supportive prompts that help a student recall, connect, and consolidate a topic.

Assessment

Escalating-difficulty questions that measure the real depth of understanding.

Questions, answered

Is this an AI detector?
No. Cognition never guesses whether work was AI-generated and never returns a verdict on cheating. It has a conversation about the work and surfaces evidence of understanding for you to interpret.
Do students need an account?
No. Students join with a short class code and a first name or nickname. We never ask for student emails, passwords, or dates of birth.
Is it fair to every student?
A conversation meets each student in their own words, so it does not penalise non-native speakers or particular writing styles the way detection tools do. You stay in control of how results are interpreted.
What happens to student work and answers?
Answers and uploaded work are used only to run the conversation and produce your summary. They are never used to train AI models, and data stays in your school’s configured region.
Evidence, never accusations
Answers never train AI
Data stays in your region
Built for the classroom

See what your students really understand.

Set up your first task in a couple of minutes. Free while we're in early access.