How AI-generated content performs in Google Search: A 16-month experiment

Source
ai-contentgoogle-searchcontent-strategyseo-experimente-e-a-tranking-signalscontent-scale

Summary

A 16-month experiment publishing 2,000 unedited AI articles across 20 zero-authority domains found strong early traction (71% indexation in 36 days, 122k impressions in month 1) followed by near-total ranking collapse by month 3, with only 3% of pages remaining in the top 100. Without backlinks, E-E-A-T signals, or unique insights, Google indexes AI content readily but stops surfacing it. A March 2026 follow-up found that adding fresh AI content lifted impressions on older stagnant pages by 17-19x, suggesting a potential crawl-signal use case.

Key Insight

Key findings with numbers:

  • Month 1: 71% of pages indexed within 36 days; 122,102 impressions, 244 clicks; 28% of ranking URLs already in top 100
  • Month 3 collapse: top-100 pages dropped from 28% to 3% - a ~90% wipeout
  • Month 6: ~70-75% of all impressions and clicks had already occurred in the first 2.5 months; only 25-30% added over the remaining 3.5 months
  • Month 16: 20% top-100 pages (after August 2025 spam update gave a temporary lift); cumulative 1,092,079 impressions and 1,381 clicks total across 20 sites for 16 months - that is roughly 69 clicks per site per month at steady state
  • YMYL penalty is real: Finance domain retained only 9/100 indexed pages; Health retained 14/100 - much lower than other niches

Non-obvious takeaways:

  • Google’s “testing period” behavior is well-documented in theory, but this experiment quantifies the drop precisely: the window is roughly 2-3 months for unassisted AI content
  • The fresh-content crawl-signal finding is actionable and distinct from ranking: publishing new AI content (even low-quality) can wake up dormant pages that already have some trust
  • The experiment deliberately used no internal linking, no backlinks, no human editing - so results represent a true floor, not typical performance
  • The August 2025 spam update paradoxically improved some AI content sites marginally, suggesting the update was more nuanced than a blanket AI-content penalty