AI text detection for teachers
A detector can tell you where to look. It cannot tell you what a student did. Here is a way to use Litimus as a prompt for a conversation rather than a verdict.
Ground rule. Litimus’s Terms of Service ask you not to treat a score as conclusive or as the sole evidence of academic dishonesty. Turnitin’s guidance says the same about its own report. The process below is built around that.
A fair five-step process
1. Start with your policy, not the tool
A detector only matters against a rule students were told about. Check what your syllabus and institution say about AI assistance: banned, allowed with disclosure, or allowed for certain steps. If students were never told, a score is not a basis for anything.
2. Read the work yourself first
Does the piece sound like the student, given what you have seen in class and in earlier assignments? Does it engage with what you actually taught? Your reading is better evidence than any score.
3. Scan a full passage and read the reasons, not just the number
Paste in a whole section rather than a paragraph. Litimus will score from 8 words but says plainly that anything under 25 words is too short for a stable read, and longer passages give steadier results. Then read the readings and margin notes. A high score driven by a few stock transitions tells you something different from a high score driven by uniformly even sentences.
4. Compare against a baseline
Set the result next to the student’s earlier work and, where you have it, in-class writing. Ask for the draft history: Google Docs and Word both keep version history, and a document built up over several sessions looks very different from one pasted in whole.
5. Talk to the student before you conclude anything
Ask them to walk you through how they wrote it: where the ideas came from, why they structured it this way, what a particular sentence means. Assume good faith and keep notes. Turnitin’s own guidance advises educators to assume positive intent and to discuss findings openly with students.
What a Litimus report shows
- A 0–100 score shown as “% AI-likely”, with a label: AI-likely (66 and above), mixed / unclear (35 to 65) or human-likely (34 and below).
- Readings for sentence rhythm, vocabulary spread, stock phrases and repeated phrases.
- Margin notes that quote the stock transitions and repeated three-word phrases found in the text.
It does not show who wrote the text, which tool (if any) produced it, or whether AI use was allowed. The score is built from six measurable signals, listed in how AI detectors work.
Where it goes wrong
- Non-native English writers. Independent research found that widely used detectors consistently misclassified non-native English writing as AI-generated. Litimus is not immune. Take extra care with English-language learners.
- Formulaic writing. Templates, lab reports, short answers and highly structured genres can read as even and repetitive without any AI involved.
- Edited AI text. Text that was generated and then rewritten by hand can land in the middle of the range, or lower.
- Borderline scores. Anything in the mixed / unclear band is inconclusive by design.
More in why detectors produce false positives.
Plans for teaching
The free tier covers 2,000 words a day with no account, enough for a handful of spot checks. If you check work regularly, Pro is $15 a month (or $12 a month billed yearly) with 25,000 words a day, .docx and .pdf files and batch scans of up to 20 files. Team & API starts at $49 a month with five seats for departments. A $5 day pass adds 10,000 words for 24 hours. Full details are on the pricing page.
If your school already licenses Turnitin, that is the report your institution’s process will point to. See how it compares with Litimus.