How to Check AI Content Without Using Turnitin

Table of Contents

• Introduction
• Correlating AI Detectors With Turnitin
• Using Proxies For Turnitin
• Conclusion and Key Takeaways

Introduction to Detecting AI Writing with Turnitin and Other Tools

In this blog post, we will dive into the world of AI writing detection. Specifically, we will look at how different AI detection tools like Turnitin, Copyleaks, CrossCheck, ContentScale, and Originality.ai compare in their ability to detect AI-generated content. We ran an experiment generating both original essays written by a human writer and transformed versions generated using an 'undetectable' AI tool. By comparing the results across detectors, we aim to find proxies that correlate strongly with Turnitin so writers can get a sense of AI detection without needing direct access to Turnitin.

Understanding the landscape of AI detectors is crucial as more content marketers and writers leverage AI tools in their workflows. While AI-generated content ranks well on Google when done properly, running afoul of plagiarism checkers is still a risk. This post breaks new ground in directly comparing the leading detectors like Turnitin head-to-head so writers can navigate AI detection with confidence.

Overview of Our Experimental Setup

We began by using GPT-3 to generate a list of 10 popular essay topics. For each topic, we wrote a 500-600 word original essay using a human writer. Then we used an 'undetectable AI' tool to transform each essay and make an altered version. We selected 5 AI detectors to test: Turnitin, Copyleaks, CrossCheck, ContentScale, and Originality.ai. For each detector, we checked scores for both the original and transformed essay versions. By comparing correlation of scores, we identified the strongest proxies for Turnitin when testing AI content.

Key Criteria for Selecting AI Detection Tools

Our criteria for selecting detectors were: 1) free or low cost, 2) no strict limitations on word count or frequency of checks. For example, Copyleaks and CrossCheck have free tiers which can be accessed indefinitely by toggling location services or clearing browser cookies and cache. While less accurate, Originality.ai and ContentScale have completely free versions without limits. This enabled comprehensive testing across a range of detectors at no cost.

Comparing AI Detectors on Original Human-Written Essays

We first tested the original human-written essays across all detectors. Originality.ai consistently scored around 100% AI detection. However, correlation analysis showed only a 35% correlation with Turnitin scores. Both CrossCheck and ContentScale also showed low and negative correlation with Turnitin on original essays.

The clear winner for correlating with Turnitin scores was Copyleaks. While scores weren't identical, graph analysis showed Turnitin and Copyleaks followed similar peaks and valleys around 75% of the time. For writers without Turnitin access, Copyleaks provides a strong proxy on original human-written content.

The Limitations of AI Detectors on Untransformed Content

For unaltered human-written essays, most detectors beside Copyleaks showed poor correlation with Turnitin. This suggests their AI models may focus on different detection signals. However, another interpretation is that these tools are less accurate than Turnitin, with more erroneous or inflated AI detection scores. Until the AI models converge more, Copyleaks provides the strongest proxy for writers looking to predict Turnitin scores on original content. Other detectors have room for improvement in their accuracy and correlation with Turnitin benchmarks.

The Advantages of Copyleaks for Original Content Checking

While not perfect, Copyleaks provides a few key advantages:

• Strong correlation with Turnitin, allowing writers to predict Turnitin scores
• Free tier available indefinitely with some workaround
• Quick and easy online checking without software downloads For those producing original content at scale, Copyleaks is a strong option for frequent, free checks without the need for direct Turnitin access.

Checking Transformed Content Across AI Detectors

Next, we compared detector scores on the AI-transformed versions of each essay. Here, Turnitin consistently scored the transformed essays as 0% AI detection. However, correlation analysis wasn't feasible since there was no variation in Turnitin scores.

Instead, we looked at average AI scores. CrossCheck and ContentScale both averaged close to 0% AI detection on transformed essays. This suggests both tools could act as proxies for Turnitin when testing content that has been rewritten by AI.

The Viability of ContentScale and CrossCheck for Altered Content

For content that has been transformed or rewritten by AI tools, CrossCheck and ContentScale appear to align closely with Turnitin scores. While not perfect proxies, they allow writers to gauge the likelihood of passing Turnitin analysis after AI alteration. Some key advantages of ContentScale and CrossCheck:

• Very close alignment for altered content, predicting Turnitin evasion
• Free tiers available with minimal workarounds
• No software downloads necessary for online checking

Caveats When Using AI Detector Proxies

A few important caveats on using detectors as proxies:

• Correlation imperfect, scores may diverge in some cases
• Free tiers have usage limits requiring workarounds
• Models vary so proxies may decay over time
• Nuanced review still required to catch issues proxies miss Still, with these limits in mind, ContentScale and CrossCheck provide valuable free tools to estimate Turnitin evasion likelihood.

FAQ

Q: Why check AI content without Turnitin?
A: To save money and access limits by relying on free proxies that correlate with Turnitin's AI detection.

Q: What tools were tested?
A: The tools tested were Turnitin, Copyleaks, Cross Check, Content Scale, and Originality.

Q: What was the experiment process?
A: 10 popular essay topics were used to generate 20 essays - 10 original and 10 transformed via AI. These were checked by all tools to find correlations.

Q: Did any tools fully correlate?
A: No, but Copyleaks had a 75% correlation on original content while Cross Check and Content Scale correlated on transformed content.

Q: Can I rely fully on the proxy tools?
A: No, some variation is expected. But high correlations allow the proxy tools to be used as a general guide.

Q: What was the hardest tool to bypass?
A: Originality.ai consistently scored content as 100% AI, with almost no correlation to Turnitin.

Q: What proxy is best for original content?
A: Copyleaks Basic had a 75% correlation with Turnitin on original human-written content.

Q: What proxies are best for transformed content?
A: Cross Check and Content Scale consistently scored transformed content as 0% AI, aligning with Turnitin's scores.

Q: How can I access Cross Check and Content Scale for free?
A: Use browser/VPN switching to bypass limits. Cross Check allows 3 checks per session.

Q: Can this method be used for academic writing?
A: No, it is intended only for general SEO web content. Academic integrity policies still require Turnitin.