Academic text analysis
We're building an AI detector for academic writing.
We're working on a tool that shows which parts of Turkish and English academic texts read like AI writing. The analysis tool is currently in closed testing.
What we need right now: your texts
We're collecting academic writing to test and improve the detector: texts you wrote yourself, had AI edit, or had AI write from scratch.
- No assigned tasks, send it the way you actually work
- We don't ask for your name or email
- One submission takes about a minute
What Scribmetric looks at
AI pattern analysis
Splits the text into 60–160 word passages and scores how closely each one resembles AI writing.
Academic language
Marks formulaic phrases. Spelling, punctuation and tone suggestions are in progress.
Structure and coherence
Will check whether the introduction, method, results and discussion each do their job.
Citation review
Will check whether in-text citations and the reference list match.
Three steps
- 1
Paste or upload
Right here, no account needed.
- 2
Review the result
Click a marked passage to see its score and explanation.
- 3
Improve your text
Know which parts deserve another look.
The results view
Your text is marked passage by passage on the left; the selected passage is explained on the right.
This study surveyed the academic reading habits of 412 undergraduate students at two public universities. Data were collected during the spring 2023 term.
The growing availability of digital sources is reshaping how undergraduates engage with academic texts. This study investigates students' reading frequency, source preferences and self-regulation skills, and evaluates the relationships among these variables within a unified analytical framework.
The results show that most students read fewer than five hours of academic material per week and mostly leave their reading until exam periods.
Passage 2 / 3 · 37 words
Strong AI-like pattern
This passage closely resembles the AI-written texts the model saw during training.
Model score: 0.99
Turkish and English, each on its own terms
The model was trained on Turkish and English academic writing together, but it uses a separate threshold for each language. We don't judge a Turkish text by English standards.
English detectors often mistake writing by non-native speakers for AI. That's why we included English papers by Turkish academics in training; the false-alarm rate for that group is currently below 1%.
Would you send us a text?
We need real writing to improve the detector: texts you wrote yourself, had AI edit, or had AI write from scratch. We don't ask for your name or email.