Experiment ● Running

Experiment: AI-Generated vs Hand-Written Content

Can Google detect and penalize AI-generated content? Does it rank differently from human-written content on the same topic?

Hypothesis

In 2026, Google cannot reliably distinguish high-quality AI-generated content from high-quality human-written content on the same topic when both are accurate, useful, and properly structured. If this hypothesis is correct, the ranking difference between variants will be within noise levels (<5% impressions difference at comparable positions).

Why This Matters

Google's Helpful Content system was designed in part to devalue "content created primarily for search engines rather than people." AI content farms producing thin, unhelpful content are an obvious target. But the question for legitimate publishers is whether genuinely useful, accurate AI-generated content faces the same penalties.

The Genbox blog posts are AI-generated (by Claude Sonnet). This experiment uses this site's own blog posts as the AI variant and compares them to a hand-written counterpart on the same topic, with identical SEO treatment.

Setup

Variable AI Variant Human Variant
Author Claude Sonnet 4.6 Human editor
Topic Same keyword target Same keyword target
Word count ~1500–2000 words ~1500–2000 words
Schema markup Article + BreadcrumbList Article + BreadcrumbList
Internal links Same count Same count
Publish date 2026-04-14 2026-04-14
AI-disclosure None (no "written by AI" label) None

What We're Measuring

  • Average ranking position — at 30, 60, 90 days post-indexing
  • Impressions trajectory — crawl frequency and discovery speed
  • CTR at matching positions — behavioural signals
  • Manual action / Helpful Content signal — any Search Console warnings

Current Status

The AI variant (three blog posts: regex cheat sheet, JSON formatting guide, AI code generators) was published on 2026-04-14. The human-written counterpart articles will be published on the same date once authored.

Results will be updated here as Search Console data comes in.

Timeline

DateActionStatus
2026-04-14AI variants publishedDone
TBDHuman variants publishedPending
TBD + 30dFirst snapshotPending
TBD + 90dFinal analysisPending
← Back to all experiments