JumpFly just published a cautionary tale that should make every SEO and content marketer pause. Its client EarthMed agreed to test AI-assisted blog content on an established regional site. The result: 31 posts helped crash organic search performance across the entire domain.
The experiment that backfired
EarthMed’s site was not a content weakling. It holds 35,000 indexed pages and, until July 2025, more than 500 blog posts hand-written by an industry expert. At the client’s request, JumpFly began publishing AI-assisted posts in mid-July 2025, roughly one every nine days.
This was not lazy copy-paste output. Each post used highly targeted custom prompts, then a genuine subject expert reworked it for about 90 minutes. Some months saw four new AI-assisted posts and four AI-rewritten posts go live.
When Google launched its December 2025 core update on December 11, blog organic clicks fell 30.7% over the next 28 days compared with the previous period. By August 2026, the blog had lost 67.8% of its clicks; its share of total site traffic fell from 45.4% in October 2025 to 24%.
The damage was domain-wide
This decline was not limited to the blog. The homepage lost 77.2% of clicks year over year, and branded clicks to the homepage slipped as well. Google also stopped indexing the last eight blog posts by May 2026, a clear signal it did not value the pages enough to show them.
Why Google demoted a polished AI test
Google does not officially ban AI-written content, but it draws a hard line when automation is used to manufacture pages at scale. As JumpFly notes, Google’s scaled content abuse policy warns against using generative AI to create many pages without adding user value. Its quality raters are instructed to treat low-value AI content as lowest quality, and that feedback informs the algorithms.
Search and AI engines increasingly punish synthetic content. JumpFly points to Google’s SAFE and S-CTS systems, which target synthetic text and video, plus EU moves toward watermarking AI-generated text. Public tools such as Copyleaks, ZeroGPT and GPTZero already make AI text easier to detect, and Google’s own detection is far stronger.
The practical bar is E-E-A-T: experience, expertise, authoritativeness and trustworthiness.
Signals to watch before it’s too late
- Blog share of total site clicks declining
- Branded clicks to the homepage dropping
- Google refusing to index recent posts
- Impressions collapsing after core updates
What to do now
EarthMed has removed all AI-assisted posts and returned to expert-written publishing, but recovery is not quick. JumpFly estimates recovery from a content-quality or spam hit can take six to 18 months or longer once Google recrawls and reassesses the site.
The takeaway for performance marketers is blunt: use AI for research, keyword mapping and outlines, not for writing or polishing published pages. If organic search and AI visibility matter to your business, a short-term content win is not worth a domain-wide demotion.
Source: JumpFly, Inc.
