This is a tool designed to detect the use of ChatGPT and other generative AI tools in essays and other (reasonably long) forms of student writing. It scans the content of one or more .pdf, .docx, and .docx files for traces of AI and searches the text for mentions of ChatGPT, Claude, Gemini, Llama/Meta, MS Copilot, and Grammarly. The tool also generates CSV and Excel summary files that can be downloaded after the scan. The Excel file uses cell background colors to visualize all files that used AI without acknowledging its use (marked in red, and green otherwise). Java Script needs to be enabled for this tool to work.
After you click "Check Files", the tool will scan each file in order of selection. It will display the following information for each document:
| Variable | Description |
| AI Traces | Total count of AI artifacts found in the document. |
| ChatGPT Mentioned | Whether any variant of 'ChatGPT' was found in the document, either with or w/o whitespace, case-insensitive. If ChatGPT is cited or there is a statement of its use in the document, this will return 'Yes' and 'No' otherwise. |
| Grammarly Mentioned | Whether any variant of 'Grammarly' was found in the document, case-insensitive. If Grammarly is cited or there is a statement of its use in the document, this will return 'Yes' and 'No' otherwise. |
| Claude Mentioned | Whether any variant of 'Claude' was found in the document, case-insensitive. If Grammarly is cited or there is a statement of its use in the document, this will return 'Yes' and 'No' otherwise. |
| Gemini Mentioned | Whether any variant of 'Gemini' was found in the document, case-insensitive. If Grammarly is cited or there is a statement of its use in the document, this will return 'Yes' and 'No' otherwise. |
| Llama/Meta Mentioned | Whether any variant of 'Llama' or 'Meta' was found in the document, case-insensitive. If Llama/Meta is cited or there is a statement of its use in the document, this will return 'Yes' and 'No' otherwise. |
| Copilot Mentioned | Whether any variant of 'Copilot' was found in the document, case-insensitive. If Microsoft Copilot is cited or there is a statement of its use in the document, this will return 'Yes' and 'No' otherwise. |
Moreover, the results will be red if any AI Traces were found, but there is no acknowledgment/cite of any of the AI tools above. Here's an example:
name_of_an_essay_uploaded.pdf
AI Traces: 3
ChatGPT Mentioned: No
Grammarly Mentioned: No
Claude Mentioned: No
Gemini Mentioned: No
Llama/Meta Mentioned: No
Copilot Mentioned: No
Similarly, if there are no traces found, or if at least one of the above AI tools is mentioned in the text, the results will be green, like in this example:
name_of_a_second_essay_uploaded.pdf
AI Traces: 3
ChatGPT Mentioned: No
Grammarly Mentioned: Yes
Claude Mentioned: No
Gemini Mentioned: No
Llama/Meta Mentioned: No
Copilot Mentioned: No
The same color scheme applies to the Excel summary file.
Lastly, in very rare instances, the result is not clear. The reason is that the trace patterns are too consistent. In most cases, it means that the essay was fully written bei AI, but it may also be due to a specialty font (see limitations tab). In the web interface, such cases are neither red nor green. In the Excel summary file, such cases have a grey cell color.
Many existing tools employ elaborate machine learning algorithms to detect semantic characteristics of AI-generated content.
This tools is much simpler (and perhaps more accurate than others when it comes to student writing): AI-generated syntax is full of cues and heuristics that remain in place even after changing fonts etc.
Think of these AI traces as 'meta data' (even though most of it is actually not hidden). Student writers often simply copy and paste AI-generated content, which means that those traces remain in the text.
AIDetection.info searches for very common traces and counts their number in a document. Give it a try!
A working paper that introduces the underlying methodology and validates this tool is in the making.
When you click on the upload field, you can select more than one file. There currently is no limit to the number of documents you can upload, but since your browser is doing all the processing, it may crash if you select thousands of files at once. The scanning order is determined by the order in the folder from which you select the files.
Tip for Canvas Users: Check all assignment submissions at once
For users of Canvas (the course management software maintained by Instructure),
you can easily check all your students' submissions for AI use by doing the following:
AIDetection.info will automatically go through all files in the submissions folder. You can then download the Excel summary file to get an overview over who has used AI, and who has acknowledged its use or not.
The AI detection tool is entirely client-based, meaning that it does not exchange any sensitive data (neither uploaded files nor summary files) with the server. Everything stays within your browser environment. Hence, it should be safe to use in educational environments.
There are a few limitations to this tool. First, the acknowledgment search won't work for essays that are about AI (because it simply matches AI names). Second, only files created after November 22, 2022 (the introduction of ChatGPT) can be processed. Third, if students use a specialty font, the detector may not work. In most cases, however, students use Times New Roman, Arial, or related fonts, which have yielded very accurate results.