Do AI Writing Detectors Actually Work?
Five different AI detectors tested, results are surprising.
1The Rise of AI Writing Detectors
AI writing detectors are designed to distinguish between human-written and AI-generated text. These tools are used in a wide range of fields, from education to job applications. But do they really work? To answer this question, we tested five different detectors.
2Pangram: High Accuracy Rate
Pangram performed well in detecting AI writing. It correctly identified both human-written and AI-generated text. It particularly flagged phrases like 'from the moment you...' as AI writing. Pangram allows four free queries per day in its free version.
3Grammarly: Convincing Results
Grammarly was successful in detecting both human and AI text. However, it couldn't fully identify AI writing; it marked Claude and Gemini texts as 68% and 66% AI, respectively. Grammarly's paid plans start at $12 per month.
4GPTZero: Preserving Human Writing
GPTZero operates with the mission of 'preserving what's human.' In tests, it correctly identified both human and AI text. The free version allows scanning of up to 10,000 words per month. GPTZero also highlighted the most AI-like sentences.
5Scribbr and Copyleaks: Mixed Results
Scribbr correctly identified human writing but mislabeled AI text. Copyleaks marked Claude text as 0% AI while Gemini text as 100% AI. Both tools showed inconsistent results in AI detection.
6Conclusion: Are AI Detectors Reliable?
The tests showed that AI writing detectors mostly work. While successful in identifying human writing, these tools were less consistent in detecting AI text. However, they clearly represent a significant step in AI text detection.
Original article: Popular Science — AI Yazı Dedektörleri Gerçekten İşe Yarıyor mu?