Understanding Sentence Impact & Highlighting in Advanced Scan

Advanced scan offers more detail into just how we make our AI probability determination. We scan your text to find which sentences and phrases are disproportionately affecting your overall AI or human score. Sentences with high impact are very likely to be adding to our confidence in your AI/human probability score, while sentences with medium to low impact are proportionately affecting your score.

Some sentences in an otherwise AI or human document might not show up as highlighted. They may still have an effect on your score, but they aren’t more likely than other parts of your document. You may find that adjusting an entire document changes your score more than just adjusting the sentences. This is because our detector holistically analyzes the document, and its prediction can depend on a pattern that affects the bulk of the text.

Understanding Probabilities

GPTZero’s percentage refers to our model’s probability that your document was written by either AI or human. Probability is a calculation that predicts how likely something is to happen; in this case, how likely a document was written by a human or AI.

In other words, “6% AI” means that our software believes only 6 times out of 100 cases we’d see something similar written by AI, and 93 times out of 100 times, your document was written by a human.

Unlike our plagiarism detector, which can find an exact copy of something that is referenced or duplicated on the internet or databases, current AI detection is always probabilistic and predictive. This is because AI is constantly evolving and adapting, and so is human language.

There is a difference between AI and human writing, and we seek to make sure we preserve knowledge of that difference. However, probabilistic results, where sometimes the confidence score can be low, are also why we don’t recommend using an AI detection result as the only proof for academic punishment or discipline.

Read more on what to do if you detect AI in your classroom.

Understanding Confidence

Every score also comes with a statement about our confidence in our prediction. Confidence measures the rate at which we would incorrectly identify the text document (error rate). These can range from:

• Highly confident: our error rate is less than 2%
• Moderately confident: our error rate would be around 10%
• Low confidence: our error rate would be 14% or higher

Our confidence levels adapt and are set automatically based on maintaining the highest accuracy rate (95%+) while also keeping the lowest false positive ratio (<1%). In other words, if we aren’t confident something is AI, we generally air on the side of “human” with a low confidence score, to avoid the case where a human is falsely accused of AI.