You may notice that in some instances the Overall Score and the sentence level scores don’t align. In this article, we explain the reasons this occurs and how best to interpret the results from the different checker platforms.

With any statistical or probabilistic analysis the more data the increase in the likelihood that prediction is accurate, the same goes for the AI content checkers which are trained on very large datasets of human-written and AI-written content.

The fundamental principle is straightforward: the larger the volume of data available for examination, the greater the likelihood of generating precise predictions. When the checkers parse the content to determine the probability it was generated by AI they analyze the content in a variety of ways, the whole document, paragraphs, and finally sentences.

The challenge comes for the prediction models when analyzing individual sentences as there is a lot less data to parse which decreases the confidence levels in their predictive models. These models are improving but it is important to take the sentence-level prediction as a guide rather than a fact. The overall score is the most reliable indicator that the document has been generated by AI.

Despite the limitations at the sentence level, this remains valuable as a useful indicator. It offers actionable insights that can guide you in refining the content to reduce the likelihood of being flagged as AI-generated. By pinpointing sentences with a higher probability of AI generation, you can focus your efforts more effectively, enhancing the authenticity of the content.

With our unique platform, you can conveniently check between checkers to see if there is consistency in the sentence-level results thus providing confidence in the prediction.

We recommend reworking the sentences that have a higher likelihood they were generated by AI and then re-check the whole document.