AI-Generated Content Dominates on Some Sites

We analyzed 3,900 articles from the blogs of 246 Fortune 500 companies. Our AI detector examined the text of reviews to find how much content was likely generated by humans or AI.

Summary of Key Findings:

1. Approximately 10.86% of articles posted on the official blogs of Fortune 500 companies are potentially AI-generated. 
2. Sites have a wide variation, from almost no AI-generated content to more than half of content possibly being generated by AI.
3. Shorter entries were 5.7x more likely to be suspected of using generative AI.

Study Method and Data:

1. The text from 3,900 blog entries that were posted on the official sites of Fortune 500 companies were rated by Originality.AI’s AI detector as human or AI generated.
2. We tracked the number of words in each article, and the percentage of a company’s entries that were likely AI-generated.
3. The word count of the articles was analyzed along with whether the review was suspected to be written by a human or AI. The relationship between word counts and AI-generated text was analyzed. 

AI in Blog Articles Study Findings

Almost 11%  of articles at official blog sites are suspected of being AI generated

The articles on the blogs of Fortune 500 companies are mostly human written. Using our AI detector, we found that 10.86% of articles reviewed in November 2023 were likely to be AI-generated. However, there was a lot of variation in how much potential AI content was found on different companies’ sites, ranging fully from zero to 100%. A closer examination was made of the 20 companies for which there were at least 20 different articles in the dataset, to a high of 72 articles. Those individual sites ranged from a low of 1.56% to a high of 63.16.%.

Reference our accuracy study for more details on Originality’s AI Detector - https://originality.ai/ai-content-detection-accuracy/

Higher Word Counts Less Likely to Have AI-Generated Content

A strong negative association was found for the number of words in the article and its odds of rating a high chance of being written by generative AI. The longer the article, the more likely it appeared to be written by humans. The shorter the article, the more likely it appeared to be written by AI. Dividing the dataset in half at the median 3,723 words created a stark difference, with 18.8% of the shorter articles examined appearing to be AI-generated, compared to only 3.3% of the longer articles. Shorter blog entries were 5.7x more likely to rate as AI-generated than the longer ones.

This finding has a certain resonance with an Originality.AI study on Sports Illustrated articles, where shorter, less substantive sections of articles, like the introduction, were more likely to be potentially written by AI as compared to the longer bodies and actual reviews in articles. This finding suggests that generative AI be more likely used for shorter discussions, while longer articles, which might have more details and be more in-depth, remain the creation of human authors.

Summary

AI-generated content might be in wide use for some Fortune 500 blogs, while others appear to use none. Questions that remain to be answered include if the writing staff or the audience is aware of the AI content on these sites? Should they be? Is a feature article less interesting if it is generated by AI? Do readers care if shorter reporting is AI written? The controversy around Sports Illustrated’s use of AI-generated articles suggests that transparency, or at least not creating names, photographs, and biographies for non-existent authors, might be more acceptable to most people.