Analyzing 90 Days of AI-Generated Content Performance: Insights from a Data-Driven Experiment
As the use of artificial intelligence (AI) in content creation continues to expand, many content creators and SEO professionals are eager to understand how AI-generated posts perform in real-world scenarios. To shed light on this subject, I conducted a comprehensive 90-day experiment, tracking the rankings and engagement metrics of 47 AI-written articles compared to human-created posts, all within the same niche and on a site with consistent authority.
This article shares the key findings from the experiment, offering a factual perspective rooted in observed data—moving beyond theory to provide practical insights.
The Experimental Setup
- Content Creation Tools Used: Claude, GPT-4, and Jasper
- Content Types:
- Pure AI-generated posts with no human editing
- AI-generated posts with extensive human editing
- Original human-written posts
- Metrics Tracked:
- Search rankings for similar keywords
- User engagement metrics such as time on page and bounce rate
Results Summary
The results reveal some intriguing patterns:
| Content Type | Top 20 Rankings | Top 10 Rankings | Engagement Impact |
|———————————–|——————-|—————–|——————————————|
| AI-only (no edits) | 3 out of 10 | 1 out of 10 | Significantly lower engagement |
| AI + Heavy Human Editing | 7 out of 10 | 7 out of 10 | Moderate engagement, similar to human |
| Human-created | 6 out of 10 | 6 out of 10 | Better engagement metrics |
Key Observations:
-
Ranking Performance:
AI-generated content, when unedited, struggled to secure top positions, with only 30% reaching the top 20 results. However, when human editors enhanced the AI posts, rankings improved notably—over 70% achieved top 10 positions—comparable to or slightly better than fully human-written articles. -
Engagement Metrics:
Despite ranking well, AI-only posts that reached top positions experienced a significant decline in user engagement. Time on page dropped by approximately 40%, and bounce rates increased by around 28%. This suggests that while the content can rank, it may not resonate as effectively with readers. -
Perception and Audience Reception:
The decline
