Does iThenticate Detect AI Writing? (2026 Guide)

If you're submitting a research paper to an academic journal or a thesis to your university's research office, there's a good chance iThenticate is scanning it. It's the go-to plagiarism checker for publishers, institutional repositories, and research integrity offices worldwide. But AI writing tools have spread fast, and more researchers are now asking: does iThenticate also check for AI-generated content?

The short answer is yes. iThenticate added AI writing detection in 2023, built on the same technology that powers Turnitin's AI checker. Whether you're a graduate student, a faculty member, or a researcher submitting to a journal, knowing how this detection works (and how reliable it is) matters.

Yes, iThenticate detects AI writing. Turnitin (which owns iThenticate) added an AI writing detection layer to iThenticate in 2023. It produces an AI percentage score for submitted text, flagging content likely generated by tools like ChatGPT, Claude, or Gemini. Accuracy for pure AI text sits around 85-90%, with a higher false positive rate on technical and scientific prose.

What Is iThenticate, and Who Uses It

iThenticate started as a standalone plagiarism checker for academic publishers and research institutions. Turnitin acquired it and eventually integrated both products under a shared technology stack. Today, iThenticate is the version used by journals, publishers, and university research integrity offices, while Turnitin (the student-facing product) handles classroom submissions.

The user base differs from typical Turnitin users. iThenticate serves:

Academic journals and conference proceedings
University research and graduate school offices
Government and industry research departments
Publishers vetting manuscripts before peer review

Because iThenticate users are typically researchers and graduate students submitting to journals rather than undergrads submitting homework, the stakes are higher. A positive AI detection flag on a journal submission can mean rejection or retraction.

iThenticate's AI writing detection, launched in 2023, runs on the same underlying model as Turnitin's AI detector. The system analyzes patterns in sentence structure, vocabulary choices, and what researchers call "burstiness" (the variation in sentence complexity typical of human writing). AI-generated text tends to have low burstiness: sentences cluster around a similar length and complexity. Human writing has more natural variation. The detector scores submissions on a 0-100% scale, where higher scores indicate more likely AI authorship. In independent testing on academic papers, Turnitin's detector (and by extension, iThenticate's) flags pure ChatGPT output at roughly 85-90% detection rates. False positive rates on highly technical scientific writing sit around 3-9% in controlled studies. The tool flags documents at the paragraph level, so mixed documents with some AI and some human sections receive per-paragraph scores rather than one blanket number.

Does iThenticate Detect AI Writing in Standard Reports?

Yes, and it's gotten more reliable since launch. The detection is wired into the standard iThenticate submission report. When a document gets submitted, the report shows both a similarity score (for plagiarism) and an AI writing percentage. There's no separate toggle or add-on to enable.

Journals and publishers that use iThenticate can configure how they want to handle the AI score. Some treat any AI percentage above 20% as a flag for manual review. Others use it as one data point among many. There's no universal threshold yet, partly because the research community is still debating when AI assistance crosses into research misconduct.

What the detector catches:

Full paragraphs written by ChatGPT, Claude, or similar tools
Lightly edited AI output with minimal changes
AI-written sections mixed into otherwise human writing (flagged per paragraph)

What it struggles with: heavily paraphrased AI text, short passages under 100 words, and AI output that's been substantially rewritten by a human.

How Accurate Is iThenticate AI Detection?

Accuracy depends heavily on what was submitted. For text generated entirely by a current large language model and pasted in without editing, detection rates are high, around 85-90% in Turnitin's own published testing data.

The harder problem is false positives. Technical writing, particularly in fields like chemistry, medicine, or engineering, often uses tightly structured, low-variability prose. That's the same stylistic fingerprint detectors associate with AI. Studies on academic corpora show false positive rates between 3% and 9% on human-written scientific papers, depending on the field.

For context on how iThenticate compares to other tools, see our breakdown of how accurate AI detectors are overall. The variance across tools is significant.

The key limitation: iThenticate doesn't tell you which tool generated the text, only that it appears AI-generated. It also can't reliably catch lightly paraphrased AI output that's been substantially rewritten by a human author.

How iThenticate Differs from Turnitin

Both tools share the same core detection model, but the implementation differs in a few ways.

Submission type. Turnitin handles classroom submissions from students. iThenticate handles manuscripts from researchers and professionals. This shapes how flagged results get reviewed and acted on.

Reporting format. Turnitin shows AI percentage at the document level with sentence-level highlights in newer versions. iThenticate's reports focus on paragraph-level flagging, which suits longer research manuscripts.

Who reviews the flag. A student's professor sees a Turnitin flag. An iThenticate flag goes to a journal editor or research integrity office, typically people with more technical knowledge and more cautious review processes.

Threshold decisions. Turnitin institutions set their own policies. Journal editors using iThenticate also set their own policies, but peer review standards vary widely across publications.

For students curious about how Turnitin handles AI in student papers specifically, our article on whether Turnitin detects AI writing covers the mechanics in detail.

What Gets Flagged (and What Slips Through)

The detector works best against unmodified or lightly modified AI text. It struggles with several scenarios:

AI text that's been heavily paraphrased by a human author
AI text run through a humanization tool that alters the statistical fingerprint
Very short passages (under 100 words) where there isn't enough text to establish patterns

The detector doesn't flag AI-assisted writing as a binary yes or no. A researcher who uses AI to clean up phrasing on one paragraph while writing the rest from scratch will likely get a low score. The risk zone is sections where AI wrote the bulk of the content, even if a human made minor edits afterward.

One thing worth knowing: the tool has no way to verify your intent. It measures patterns in text, not whether you typed every word yourself. Some researchers use AI for literature summaries, data description, or draft structuring and then rewrite heavily. Whether that counts as AI-generated content is a policy question for your journal or institution, not a technical one.

How to Lower Your AI Score in iThenticate

If you've used AI tools in your research workflow and want your submission to read as genuinely human-written, a few approaches work:

1. Rewrite AI sections in your own voice. The most reliable method. Take the AI draft and rebuild it sentence by sentence using your own structure and phrasing. Don't just swap synonyms; change the sentence order, add your own observations, and inject domain-specific nuance that AI typically flattens out.

2. Use a humanization tool as a starting point. Tools that rewrite at the sentence and paragraph level change the statistical patterns AI detectors look for. This produces a better base for your own editing, not a final product you paste directly.

3. Check before submitting. Run your manuscript through a detector before sending it to a journal. Catching a high AI score yourself is better than getting a flag from an editor.

4. Be transparent about AI use where policies allow. Many journals now have AI disclosure sections in their author guidelines. Acknowledging AI assistance in your methods or acknowledgments can shift the conversation from misconduct to transparency.

How NaturalRewrite Can Help

If you've drafted sections with AI assistance and need the final text to pass detection before submission, NaturalRewrite handles the humanization step cleanly.

Paste your AI-drafted sections into the tool, select Academic tone (available on Starter plans and above), and NaturalRewrite rewrites the text in a way designed to pass major AI detectors, including those built on Turnitin's model. The Academic tone mode produces polished, scholarly-sounding output that matches the register of research writing.

The built-in AI detection checker lets you verify the score before you submit anywhere. If the score is still too high, you can run another pass or edit the flagged sections manually.

The free tier gives you 5 humanizations per day with no credit card required. Starter runs $7/month if you're working through a longer manuscript in one sitting. There's no file upload or API; everything works through the web interface at naturalrewrite.com.

Frequently Asked Questions

Does iThenticate check for AI writing by default?

Yes. Any document submitted to iThenticate on an institutional license that includes AI detection automatically receives an AI writing score alongside the standard similarity report. Whether AI detection is enabled for your institution or the journal you're submitting to depends on their specific contract and settings.

What percentage triggers a flag in iThenticate?

There's no universal threshold. Individual journals and institutions set their own policies. Some treat scores above 20% as a flag for manual review; others use higher cutoffs or treat the score as a supplementary signal rather than a hard trigger.

Can iThenticate detect AI writing from Claude or Gemini, not just ChatGPT?

Yes. The detection model looks for statistical properties common to large language model text in general, not output from a specific tool. Current major LLMs including ChatGPT, Claude, and Gemini all produce detectable patterns that the system can flag.

Does iThenticate detect AI differently from how Turnitin does?

The underlying detection model is the same. The main differences are in reporting format and who reviews the results. iThenticate targets research manuscripts; Turnitin targets student papers. Both run on the same Turnitin-developed AI detection algorithm.

What happens if iThenticate flags my paper as AI-generated?

It depends on the journal or institution. Editors typically reach out for clarification before making a decision. A flag doesn't automatically mean rejection. Many publishers are still working out their AI policies. Being transparent about AI use in your methods or acknowledgments section is the safest approach.

Conclusion

iThenticate does detect AI writing, and the feature is baked into standard institutional reports. For researchers and graduate students, the most useful thing to know is this: the detector performs well on bulk AI text but produces more false positives on technical prose, and sufficiently rewritten content can reduce the score.

If you're using AI drafting tools in your research workflow and want your final manuscript to read as genuinely human, NaturalRewrite's Academic tone mode can help you get there. Try it free at naturalrewrite.com.