Knowing what content is human-created and what is generated by AI is crucial for ensuring cyber compliance, maintaining academic integrity, and preventing copyright infringement. The Copyleaks AI detection tool supports 30+ languages and covers GPT 3, 3.4 and 5, Gemini, Claude, plus newer models as they’re released.
Our AI detector is trained to recognize human writing patterns; flagging text as potential AI when it detects deviations from these patterns and specific AI signals. One of these signals is the use of certain phrases at a higher rate than that of a human writer. No one wants to fear false positives, which can lead to untrue accusations. We’ve tested millions of texts and confidently provide a 99% accuracy rate.
Thanks to our machine learning models, we regularly test and fine-tune our AI detector, retraining it in real-time with new data and feedback. Our goal is always to be the most accurate and user-friendly tool for AI detection.
Before Launching Copyleaks With Your Team
- Establish an AI usage policy - It’s important that before AI is brought into the classroom, guidelines and rules are put in place so that faculty and educators are on the same page. Reactions to AI’s introduction into the classroom echo that of the calculator. At first, it was banned outright, but over time, rules and guidelines were put in place to establish when a calculator could be used in the classroom and when it could not. We recommend you look to address AI in a similar manner. Suggestions for how to establish these types of policies can be found here.
- Train/align faculty & students - Take time to use AI models in the classroom. Input questions and prompts and go over the responses with students, highlighting where there could be inaccuracies or where students would need to do more research without AI and show their work of doing so. Furthermore, reviewing questions and prompts within the classroom can help students learn how to utilize AI to research properly and not believe everything it generates as fact.
What to do with a positive result:
- Review AI usage policy & FAQs - Make sure you are aware and up to date with your institution’s AI usage policy and the recommendations given by Copyleaks. If you are looking for guidance on specific AI percentage thresholds, we urge you to speak to your admins and fellow faculty to understand your Academic Integrity policy.
- Understand this is one data point - As a result of analyzing millions of documents over the years, the Copyleaks AI Detector has an overall accuracy above 99% and the lowest false-positive rate of any platform, at 0.2%. Nevertheless, false positives can still happen despite best efforts to minimize them. We encourage educators to use the AI report as one data point to discuss how a student might have worked on the assignment.
- Look at the student’s writing history/analytics - Ask yourself some critical questions: Does this style match that of previous assignments from the student? Is the writing style inconsistent? Is there a high rate of phrases that are more commonly used by AI versus humans? Is the writing or vocabulary above the student’s grade level? Use the history of previous reports to assess how many times the student has submitted content that has been flagged as AI.
- Talk with the student - When it comes to having a conversation around AI and whether or not a student uses it to write their assignment or only uses it as a resource, it helps to apply some old-school methods of “show your work.” For instance, if an assignment is flagged as AI, you can meet with the student one-on- one and ask them to discuss the answers or write a synopsis of their paper on the spot. Did they touch on the points covered, or are they uncertain? If it’s the latter, chances are higher that they didn’t entirely write the assignment. As another option, if they are allowed to use AI as a resource, including writing tools like Grammarly, ask them to supply the prompt that was used within the AI model for research and discuss what prompts they used. Then, take it a step further and go over their process in doing their research beyond AI.Interpreting Copyleaks Results
A portion of text has been flagged as AI, what does this mean?
- Text which has been deemed as likely AI generated will be highlighted in purple. An overall percentage illustrating how much AI content makes up the document is displayed at the top of the page. Below this score you will see a breakdown of AI insights, which shows the relative frequency of phrase usage by AI versus humans. Note, this is one element of many that contributes to why the text might have been deemed AI.
- When AI writes a sentence, it probes all of its pre-training data to output a statistically generated sentence, which often does not resemble the patterns of human writing. It becomes more apparent when analyzed against a vast corpus of human writing.
- AI Detectors can offer a lot of insight and data to encourage essential conversations in classrooms and boardrooms alike to determine the rules and regulations around AI. When AI content is detected, take the time to investigate further. Again, the data provided by AI detectors should be used to inform the situation, create a learning opportunity, and align on expectations.
I got an AI result, but my student claims they did not use AI generators like ChatGPT or Google Gemini, what do I do?
- Certain features of writing assistants can cause your content to be flagged by the AI Detector as AI- generated. For example, Grammarly has a genAI-driven feature that rewrites your content to help improve it, shorten it, etc. As a result, this reworked content may be flagged as AI since it was rewritten or rephrased by GenAI.
- However, if your organization is using the Copyleaks Writing Assistant the resulting reworked content will not be flagged as AI. Similarly, if when using Grammarly only small adjustments (spelling, grammar errors) are made, the reworked content will not be detected as AI.
I didn’t get an AI result, but I am worried a student used a humanizer to fool Copyleaks:
- Copyleaks always errs on the side of caution; we never want to accidentally flag users as AI, and because of that we have a very low false positive rate of 0.2%. Our data scientists have built our proprietary AI model from the ground up. This differs from our competitors, who use shared models and databases to train on.
- Copyleaks also tests our AI detection against every newly released LLM. As we routinely train our model, we use content that has been altered by humanizers in order to verify that we can detect this content and keep up with the ever-changing landscape of GenAI.
For more information on AI policies, FAQs and more. Please see the attached documentation.