AI detection
How AI Text Detectors Work, and How Accurate They Are
AI detectors estimate how likely text is to be machine-written. Here's how they work, why they get it wrong, and how to read a score responsibly.
· 4 min read
Check AI-Generated Text on Your iPhone
AI Humanizer includes an AI Detector for iPhone and iPad. Paste text and it estimates how likely it is that the text was written by AI, shown as a percentage with a plain label such as “Likely AI” or “Likely Human”.
The result is an estimate, not proof of who wrote something.
How to check text for AI on iPhone
Open AI Humanizer and go to the Detector tab.
Paste the text into the Text to Analyze box.
The app needs a minimum amount of text before it can run a check.
Tap Detect AI Content.
Read the Overall AI Probability.
Tap Copy Results to share them, or Reset to check another text. Detections are saved in History.
Checking a rewrite
After you humanize a text, the result screen has a Check AI Detection button. It runs the detector on the rewrite straight away, without copying anything between screens.
What the result means
Overall AI Probability
The detector's estimate, as a percentage, that the text was generated by AI. The remainder is shown as the human share, so 99.8% AI appears alongside 0.2% human.
The label
A plain-language summary of the score, such as “Likely AI” or “Likely Human”. “Likely” is the important word.
What it doesn't tell you
Who wrote the text, how it was written or whether rules were broken. A high score means the text shares patterns the detector associates with AI writing, nothing more.
False positives and false negatives
Every AI detector makes two kinds of mistakes. A false positive flags human writing as AI. A false negative misses text that AI did write. Both happen with every detector, including the one in this app.
Human writing is more likely to be flagged when it is:
- very formal, formulaic or full of stock phrases
- short, such as a sentence or two
- written by someone working in their second language
AI writing is missed more often when it has been edited by a person or rewritten, or when it is short.
What the research says
Researchers found that several widely used GPT detectors consistently misclassified essays by non-native English writers as AI-generated, while identifying native writers' samples accurately (Liang et al., 2023, “GPT detectors are biased against non-native English writers”).
OpenAI withdrew its own AI text classifier in July 2023, citing its low rate of accuracy. At launch it correctly identified 26% of AI-written text, wrongly labelled human-written text as AI 9% of the time, and was unreliable on texts under 1,000 characters (Decrypt, July 2023).
Important limitations
- The score is an estimate. The app doesn't publish an accuracy figure for its detector.
- Different detectors can give the same text very different scores.
- Short texts give less reliable results than longer passages.
- A score can't show who wrote a text or how much of it was edited.
Also in AI Humanizer
Guides to AI detection
FAQ
AI Detector questions
Is the AI Detector accurate?
It gives an estimate, not a verdict. Every AI detector makes mistakes in both directions, and results depend on the length and kind of text. The app doesn't publish an accuracy figure, so treat the percentage as one signal among others.
Why does my own writing show as AI?
Detectors look for statistical patterns that are common in AI text, and some human writing has them too. Formal or formulaic writing, short passages and writing by non-native English speakers are flagged more often. A high score doesn't mean you used AI.
How much text do I need to check?
The app asks for a minimum amount of text before it runs a check. Longer passages generally give more stable results than a sentence or two.
Can I check text I've just humanized?
Yes. On the Humanizer's result screen, tap Check AI Detection to run the detector on the rewrite without copying it over.
Can the AI Detector prove that someone used ChatGPT?
No. An AI detection score can't prove who or what wrote a text, and it shouldn't be the only basis for decisions about grades, jobs or accusations.
Does the AI Detector work on iPad?
Yes. The whole app, including the AI Detector, runs on iPad with iPadOS 16 or later, as well as on iPhone with iOS 16 or later.
Check text for signs of AI on iPhone
Download AI Humanizer to use the AI Detector on your iPhone or iPad.
Download on the App StoreFree to download, with optional in-app purchases.