AI Detector

Recently, models such as OpenAI's ChatGPT (GPT-4 / GPT-4o), Anthropic's Claude, Google's Gemini, and Meta's Llama have led to the rise of machine-generated content. This synthetic content is increasingly indistinguishable from human-written content, leading to the frequent thought: "Was this written by AI?".

Despite rapid progress, these models continue to have shortcomings such as hallucinated facts as well as consequences such as enabling cheating in language courses.

This AI checker tool provides a way of screening whether a piece of content is written by a human or a machine, helping resolve the "Is this AI-generated?" question.

The AI detector uses a machine learning system (a Transformer) similar to that used to generate AI content. Instead of generating words, the AI detector instead generates the probability it thinks each word or token in the input text is AI-generated or not. The results are visualized above for the entire text as well as for each sentence.

Yes! We've added buttons above that allow you to upload PDFs and DOCX files. First, the tool extracts the text from the files. Then, we use the text-based system to perform AI detection. The result is a PDF AI detector as well as a DOCX AI deetector.

Yes! You can install the AI Content Detector for ChatGPT extension by Sapling.ai. This extension will allow you to check for AI-generated content anywhere on the web. Select text on any webpage, then click the Detect AI button to see a complete analysis of the selected text.

When using chatbots such as ChatGPT, the Detect AI button will be embedded next to each generated result allowing you to easily run AI detection with a single click.

You will also be able to edit the analyzed text and recheck your work. This will allow you to easily fix the sections that have been flagged as AI-generated.

Accuracy must be measured on a specific test or benchmark. There are also multiple measurements of "accuracy" for detection tools. These measurements balance catching as many AI-generated texts as possible while keeping false positives low. On our internal benchmarks, Sapling catches more than 97% of AI-generated texts while keeping false positives below 3%. Please note that these benchmarks tend to use longer texts and may not be representative of your text.

Sapling's detector can have false positives. The shorter the text is, the more general it is, and the more essay-like it is, the more likely it is to result in a false positive. We are working on improving the system so that this occurs less frequently.

The free version is currently truncated to 2000 characters (roughly 400 to 500 tokens). Pro and Enterprise Sapling subscribers can paste texts of up to 50,000 characters (roughly 10,000 to 12,500 tokens). For texts longer than that, please break up the text into multiple sections, or consider using our API. If you plan to process more than 5 million characters/month, contact us to see how we can better support your use case.

Yes — you can find the API documentation here.

While language models are evolving and have their differences, they usually use a similar machine learning architecture and a similar dataset on which they're trained. Hence, even AI detectors trained on different and earlier versions of language model outputs should perform significantly better than random on other models.

That said, to get the best performance, detectors should be trained on the outputs of the latest systems. Sapling regularly retrains and finetunes its detector to keep it up-to-date with new systems, and should be performant for models from the list above.

We invite you to collect a small dataset of say a dozen examples (of say blog posts and essays) and try for yourself :-).

We've seen such tools make false claims such as that a text "passed" Sapling.ai's detector even though no check was performed by Sapling.ai. Please be careful when using such tools.