16-month AI content experiment on Google Search — SE Ranking study results

On March 23, 2026, Search Engine Land published the results of a 16-month experiment conducted by SE Ranking: 2,000 articles fully AI-generated, with zero human editing, published across 20 brand-new domains, according to the original SE Ranking study. The outcome: after 3 months, 97% of pages had fallen out of Google's top 100. Only 3% held on. And after 16 months, no site showed meaningful recovery.

This is the first large-scale study documenting what actually happens to unedited AI content over time. The conclusions are stark.

The methodology: a clean experiment

SE Ranking's team purchased 20 fresh domains — no backlinks, no history, no authority. Each domain covered a different niche (finance, health, travel, sports, tech…). Each received 100 AI-generated articles targeting informational long-tail keywords. Zero human editing. Zero link building. Zero internal linking. Zero promotion.

Sites were submitted to Google Search Console on publish day, then left untouched for observation.

Results, month by month

71% Pages indexed within 36 days
122K Impressions in month 1
3% Pages still in top 100 after 3 months
0 Sites with meaningful recovery after 16 months

Month 1 looks promising: 71% of pages indexed, 122,000 impressions, 244 clicks. Google crawls and indexes AI content — that's not the problem. The problem is what happens next.

By month 3, the collapse is spectacular. From 80% of sites ranking for 100+ keywords, only 3% of pages remain in the top 100. Without domain authority, E-E-A-T signals, or backlinks, Google reverses the positions granted during initial indexation.

At 16 months, stagnation is total — with one interesting anomaly: the August 2025 spam update briefly pushed some pages back to 20% top-100 presence (up from 3%). But this rebound is still well below month 1 levels.

What this means for your content strategy

This experiment doesn't test whether AI can write. It tests whether AI content without human value-add can hold its ground over time. The answer is no.

Three concrete lessons:

  1. AI content indexes fast, crashes faster. Early weeks look encouraging — Google explores everything. But without trust signals (backlinks, clicks, dwell time), positions collapse. Unedited AI content is a flash-in-the-pan for SEO.
  2. Human editing isn't optional — it's the differentiator. What the experiment measures negatively is exactly what Cicero provides positively: sector expertise, cited sources, an editorial angle, a real named author. That's what Google rewards long-term.
  3. E-E-A-T is the durability signal. AI pages without an author, proprietary data, or verifiable sources have no Experience, Expertise, Authority, or Trust. They can rank briefly on long-tail terms, but they don't hold.

Our take: This study confirms what we've been saying all along. AI can be a powerful production tool — fast, scalable — but only when supervised, enriched, and signed by someone who actually exists. Publishing raw AI content at scale is like watering a garden with bleach: it grows fast, then everything burns.

Sources

  • Search Engine Land — "How AI-generated content performs in Google Search: A 16-month experiment" (March 23, 2026)
  • → SE Ranking Research Team — Study published in partnership with Search Engine Land
Alexis Dollé, founder of Cicéro
Alexis Dollé
CEO & FOUNDER

Growth and SEO content strategy specialist, I founded Cicéro to help businesses build lasting organic visibility — on Google and in AI-generated answers. Every piece of content we produce is designed to convert, not just to exist.

LinkedIn