Partly. Turnitin says its English AI writing detector can detect content from Grok-4.1, the only Grok version on its published model list, according to its FAQ updated on 3 November 2026. The list is for English submissions, and the report does not say which model wrote a passage.
What Turnitin states
- The detector gives each sentence a score between 0 and 1. Sentences above a threshold are labelled AI, and the document score is the percentage of sentences above it.
- The model list covers GPT, Claude, Gemini, DeepSeek, Llama and Mistral models as well as Grok-4.1. Turnitin says it is continuously updating the model to detect additional models.
- Its Arabic launch blog from 2026 also names Grok 4.1 Fast.
- AI paraphraser and bypasser detection is, in Turnitin's words, only available for English submissions.
Limits Turnitin states
- English needs at least 300 words of prose in a long-form format; the maximum is 30,000 words. Lists, headers, references, tables, formulas and code are skipped, and the model does not reliably detect non-prose or code.
- It processes long-form English, Spanish, Japanese, Modern Arabic, Dutch, German and Turkish. Other languages are not processed.
- Scores from 1% to 19% are not shown, because results are less reliable and false positives are more likely.
- In documents of a few hundred words the prediction is mostly all or nothing.
- Turnitin may miss some AI text, and says the percentage should not be the sole basis for action. It does not make a determination of misconduct.
- Its false-positive statements are vendor claims, not independent results. It names text without much structural variation, text that repeats itself and paraphrased text with no new ideas as likely false positives.
What this means for Grok text
Short posts, replies and anything under 300 words fall below Turnitin's own minimum. A newer Grok version than the one listed, or text that was edited by hand, may not be covered. Only instructors and administrators see the result, so a writer cannot check before submitting.
What an independent second check adds
- A different model reading the same text, so one tool's blind spot is not the only evidence.
- Highlighted passages and a writing-style similarity for Grok next to ChatGPT, Claude and other families. This is a signal about style, not proof of authorship.
- A lower minimum of 100 words, on pasted text.
Run the text on GrokDetect; for PDF and Word files, Fileverity's detector takes those too. The facts page gives the false-positive rate: 0.5% of human-written texts in seven languages were wrongly flagged. Other Grok versions are covered in Grok versions, and the watermark route in Grok watermark status.
Sources: Turnitin, "AI writing detection capabilities FAQs" (guides.turnitin.com, updated 3 November 2026) and its AI writing detection release notes (updated 14 October 2026).