Does Turnitin Detect AI? Let’s Investigate

Who hasn’t wanted to cut corners when the deadline is too close for comfort, right? Today, with chatbots like ChatGPT, it’s oh-so-easy: just describe what you want to get in a prompt, and AI will do the rest for you.

Over half of students – 59% – reported using AI at least once a month, according to the Time for Class 2024 report. This trend hasn’t gone unnoticed by educators, making AI content scans as common as plagiarism checks.

So, if you’re planning to use AI for your next paper, one question may be on your mind: Can Turnitin detect AI writing?

How Does Turnitin Detect AI, Exactly?

As of writing, Turnitin runs both a plagiarism and an AI scan on the submitted text by default. So, you should assume that your paper will get checked for AI content if your school, college, or university uses Turnitin. (Unless they opted out of the feature like Vanderbilt did in August 2023, of course.)

Now, how does Turnitin AI detection work? Here’s the rough explanation based on the Turnitin guide on its AI writing detection capabilities for educators:

• The paper is submitted to the Turnitin platform.
• Its text is broken down into multiple segments, each roughly a hundred words or five to ten sentences long.
• Segments are overlapped to provide context for each segment.
• Turnitin’s AI detection model scores every sentence on a scale between 0 and 1 (0 for purely human-written content, 1 for entirely AI-generated sentences).
• The tool then calculates the share of the likely AI-generated content in the submitted text. This score shows the share of the text likely to be generated by AI.
• The score is displayed in the Similarity Report, and the AI Writing Report highlights sentences likely generated by AI.

N.B. Turnitin can identify the AI content paraphrased with AI tools as well.

What AI Detector Does Turnitin Use? What to Know About Its Detection Methods

Turnitin doesn’t use a third-party AI detector; the company has built its own AI checker and integrated it into the platform’s workflows. So, the AI score is calculated by default whenever you check a paper with Turnitin.

Now, how can Turnitin (or any other AI checker) determine whether the sentence was written by a human or generated by AI? Well, human writing tends to be somewhat random and idiosyncratic. AI models like GPT-4, in turn, generate text by identifying the most likely next word in a sequence. So, their output follows a highly predictable, repetitive structure.

As for specific large language models (LLMs), Turnitin can detect content generated by:

• GPT-3
• GPT-3.5
• GPT-4
• GPT-4o
• Gemini (Pro)
• LLaMA

Turnitin’s AI checker works only on long-form text. So, if your submission is under 300 words, the tool will simply refuse to run an AI scan on it.

N.B. The tool doesn’t make the AI Writing Report automatically available to students. So, you can find out how your paper performed only if your instructor shares the PDF report with you directly.

What Does Turnitin Check For?

Turnitin provides a Similarity Report for every paper submitted to it. But Turnitin is adamant about one thing: it doesn’t detect plagiarism in the text.

So, what does Turnitin detect? It compares the submitted paper against all the other files in its database and looks for matches. That database includes publicly available content, articles from periodicals and publications, and other students’ works submitted to Turnitin in the past. Quotes and references can be excluded from the results.

In addition to the Similarity Report, every submission is accompanied by the AI Writing Report. This report shows the share of text likely generated or paraphrased by AI writing tools.

The AI Writing Report differentiates between AI-generated and AI-paraphrased content: the former is highlighted in cyan, and the latter uses purple.

The AI Writing Report becomes available only if Turnitin’s AI score exceeds 20%. That’s done to prevent false positives, according to the company.

N.B. Turnitin can detect AI content only within papers written in English, Spanish, and Japanese. AI paraphrasing detection is available only in English. In either case, the submitted text has to be long enough to ensure prediction accuracy.

How Good Is Turnitin AI Detection?

Turnitin itself maintains that its false positive rate is below 1%. In this context, false positives refer to human-written content getting flagged as AI-generated text. That can happen when AI checkers deal with texts written by non-native English speakers, and researchers proved that AI detectors were indeed biased against non-native speakers.

So, if you’re wondering, “How accurate is Turnitin at detecting human content?”, the answer is over 99%. But that’s only part of the picture.

In fact, Turnitin admits that up to 15% of AI-generated content may get mistakenly labeled as human-written by its AI checker. That means that if the displayed AI score is 30%, the actual one can be as high as 45%.

Besides that, Turnitin acknowledges that whenever the AI score is below 20%, the risk of false positives skyrockets – and the company doesn’t know why. That’s why scores below 20% are not displayed. (They’re replaced by an asterisk instead.)

How to Use Turnitin AI Checker: Your Short Guide

First, let’s start with a disclaimer: Turnitin doesn’t sell individual subscriptions or licenses. So, if you’re wondering, “Can anyone use Turnitin?”, the answer is no. Only educators and instructors have full access to the platform’s features. Students may be allowed to submit files to it, but this differs from one school to another.

Even if you have access to Turnitin, though, you won’t see the AI score after you submit your assignment. However, your instructor will be able to view it, along with the AI Writing Report.

Here’s how to use Turnitin AI detection if you’re an instructor:

• Ensure the file meets Turnitin requirements (size: 100MB, text length: from 300 to 30,000 words, file formats: .docx, .pdf, .rtf, .txt).
• Submit a paper in a supported language.
• Confirm your submission and let the tool run scans on the text.
• Once the results are in, go to the AI Writing tab to review the full report with highlighted AI content.

N.B. The AI score is also displayed in the Similarity Report view. You can find the indicator in the sidebar. If the AI score is below 20%, the platform will display an asterisk instead of a number.

So, Can Turnitin Detect ChatGPT?

Yes, Turnitin’s AI content scan can detect ChatGPT output. What’s more, it’s trained to detect content generated by various generations and versions of ChatGPT:

• GPT-3
• GPT-3.5
• GPT-4
• GPT-4o

Turnitin’s limitations for detecting ChatGPT content are the same as for the output of other AI models. Texts have to be over 300 words long to be scanned, and only three languages are currently supported (English, Spanish, and Japanese). The risk of false positives, while low, still exists – and false negatives persist, too.

“But does Turnitin detect ChatGPT if I rewrite some parts of it?” Well, that depends on whether you leave the AI-generated parts completely untouched. If so, your rewritten content is unlikely to get flagged, but those parts straight out of ChatGPT are. So, depending on how much of the text you’ve rewritten, your final AI score might still be relatively high.

Your best bet to avoid ChatGPT detection is rewriting and editing the whole AI-generated content – without relying on other AI tools, that is. Turnitin will detect content from AI paraphrasers, summarizers, and humanizers. So, yes, there’s no other way – you have to do some legwork yourself, whatever happens.

If you want to make sure your ChatGPT content won’t get flagged by Turnitin, you’ll also need to check how your rewriting and editing efforts impact the score. But since Turnitin doesn’t show students the AI score and doesn’t sell individual subscriptions, you’ll need to use other AI checkers.

Luckily, there’s no shortage of them: your choices include ZeroGPT, Copyleaks, QueText, and many more. If you already have a Grammarly subscription, the tool also offers an AI detection tool to its paying customers.

How Does Turnitin Detect ChatGPT?

Turnitin uses the same algorithm to detect content generated by ChatGPT as the output of any other AI writing tool. The text is broken down into snippets, and Turnitin scores each sentence on a scale from 0 to 1. The overall score is calculated as the share of sentences with high AI-generated scores.

Like with other AI models, Turnitin ChatGPT detection relies on identifying patterns in the text that indicate the text is machine-generated. ChatGPT content has several common characteristics, such as:

• Consistently long sentence structures
• Abundance of em dashes
• Extremely predictable word choice
• Repetitive vocabulary that could easily be diversified by synonyms

Keep in mind that Sam Altman, OpenAI’s CEO, ponders the idea of embedding metadata markers into AI-generated content. Invisible to the human eye but machine-readable, these markers would be akin to a watermark that instantly communicates to the detector that this is AI content.

These markers aren’t implemented in ChatGPT yet, but they may become the industry standard in the upcoming years. So, keep your eyes peeled.

How Well Does Turnitin Detect Chat GPT?

As we stated above, Turnitin itself reports that its false positives are under 1% but admits it might miss up to 15% of AI-generated output. So, how good is Turnitin at detecting ChatGPT? According to the company itself, it’s pretty good.

Yet, that doesn’t mean Turnitin can’t be fooled. Yes, you’re bound to land in hot water if you submit a fully ChatGPT-generated assignment without making any effort of your own. But if you spend some time editing it, Turnitin is unlikely to flag your content as AI-generated.

Here are four tips for your editing process:

• Add your personal voice to the text. For example, if you’ve never used ‘fascinated’ in your life and prefer to say ‘amazed,’ use Ctrl+G to replace those words.
• Remove repetition. Read the content once and identify the common words and phrases. Then, open a thesaurus and replace them with synonyms.
• Add variety to sentences. Break up long sentences, add transition phrases, and vary paragraph length. If acceptable, consider replacing overly formalized language with somewhat informal phrases.
• Lay down the specifics. ChatGPT is notoriously bad at specifics. So, add relevant quotes and citations, write concrete examples, or add an anecdote to the introduction.

‍

Best pick of the month

EssayPro.com

Save money on your essay order with this promo code to get 15% off your order!

Use promo code now!

Frequently Asked Questions