Tool
AI Text Detector
A heuristic AI detector that shows you the signals it's reading — not just a confident-sounding verdict. Useful for self-editing your own drafts. Not reliable enough to accuse anyone of AI use.
Why we built an honest detector
Most online AI detectors give you a single number and a confident verdict. They don't tell you what they're looking at, how accurate they are, or why they could be wrong. That's a problem, because real-world consequences — failing grades, lost contracts, dismissed work — depend on that output.
This tool is built differently:
- Every signal is visible. You see what specific patterns the score is based on.
- The accuracy warning is loud, not buried. Detectors are unreliable; this tool says so directly.
- Use cases are honest. Self-editing, learning what AI writing looks like, removing your own stiff drafts — yes. Grading, hiring, or accusations — no.
If you need this tool to make a high-stakes call about someone else's writing, please don't. The technology isn't there yet, and it may never be.
Frequently asked questions
How accurate is this AI detector?
No more accurate than any other free heuristic detector — meaning, expect 60-70% accuracy on raw AI text, and much lower on edited or hybrid (human + AI) text. The hard truth: no AI detector, including expensive paid tools, is reliable enough to be used as proof of AI authorship. The tool surfaces signals; it does not give verdicts that should be acted on.
Why are AI detectors so unreliable?
Three reasons. (1) Modern AI models (GPT-4, Claude 3.5) write in ways that are statistically indistinguishable from human writing on short samples. (2) Detectors flag patterns that also appear in non-native English speakers, technical writing, and formal essays — producing high false-positive rates. (3) Any edit by a human to the AI text destroys most of the signals detectors rely on. OpenAI shut down its own AI classifier in 2023 because of these problems.
What signals does this detector look at?
Five signals: (1) Presence of phrases AI models overuse like "delve into," "tapestry of," "in conclusion." (2) Burstiness — variance in sentence length, since AI tends toward uniform sentences. (3) Vocabulary diversity (type-token ratio), since AI repeats itself slightly more. (4) Punctuation density, since AI uses many em dashes and semicolons. (5) Informal markers like "yeah," "lol," and trailing ellipses, which AI almost never produces.
Can I trust this to grade student work or check freelance submissions?
No. Using any AI detector — including this one and the paid alternatives — as evidence of academic dishonesty is unethical and statistically unsound. The false-positive rate is too high. A 2023 Stanford study found AI detectors flag essays by non-native English speakers as AI-written at over 60% rates. If you are concerned about AI use in submitted work, address it through assignment design and conversation, not through detector output.
How can I write so my work does not look AI-generated?
Three habits that work: (1) Use contractions ("don't," "it's," "you're"). AIs avoid them. (2) Vary sentence length aggressively — some short, some long, some fragments. (3) Include specific personal anecdotes, real numbers, and named examples. AIs default to abstractions. The phrases this detector flags are also worth avoiding ("delve into," "navigate the complexity," etc.) regardless of how you wrote the draft.