AI Content Experiment: Publishing 200 Articles in 30 Days Generated $2,800/Month Before Deindexing

In March 2024, Derek Morrison acquired a dormant health supplements review site for $8,500 through Flippa. The site had 47 existing articles, 1,200 monthly visitors, and generated $340/month from Amazon Associates commissions. Derek's thesis: scale content production using GPT-4 to capture long-tail supplement queries before competitors could react.

Within 30 days, Derek published 200 AI-generated articles. Traffic surged to 18,400 monthly visitors by day 45. Revenue peaked at $2,800/month on day 52. Then Google's March 2024 core update hit. By day 67, the site lost 94% of organic traffic and was effectively deindexed from search results.

This case study dissects Derek's rapid-scale experiment, the revenue mechanics that briefly worked, the algorithmic signals that triggered the penalty, and the post-collapse recovery strategy that salvaged 30% of the original asset value.

The Acquisition Thesis: Speed Over Quality

Derek's acquisition criteria filtered for sites with established domain authority but stagnant content production. The target site—SupplementScout.io—had a Domain Rating of 38 (Ahrefs), 820 referring domains, and a publication gap of 14 months. Previous owner had abandoned the project after a traffic plateau at 1,500 monthly visitors.

Derek's hypothesis: the domain retained enough trust signals to absorb a massive content injection without immediate algorithmic suspicion. He projected 90 days of runway before Google's quality filters would catch up to the publishing velocity.

Acquisition details:

• Purchase price: $8,500 (25x monthly profit multiple)
• Traffic baseline: 1,200 monthly organic visitors
• Revenue baseline: $340/month (Amazon Associates, 2.8% conversion rate)
• Existing content: 47 articles averaging 1,800 words
• Domain metrics: DR 38, 820 referring domains, 6-year domain age

Derek ran due diligence through Semrush and confirmed no previous manual actions or algorithmic penalties. The domain's backlink profile showed natural link acquisition patterns—mostly editorial links from health blogs and supplement forums. No PBN footprints detected.

The Content Production System: GPT-4 Plus Thin Human Oversight

Derek built a production pipeline using GPT-4 API paired with a Python script that automated article structure generation. The workflow:

1. Keyword extraction: Scraped 2,400 long-tail supplement queries from Ahrefs (search volume 50-500/month, keyword difficulty under 20)
2. Prompt engineering: Created a 1,200-word prompt template that instructed GPT-4 to write product comparison articles in a "neutral review" voice
3. Batch generation: Processed 200 articles in 6 days (33 articles/day average)
4. Human QA: Hired a $15/hour VA to verify factual accuracy, add Amazon affiliate links, and check for obvious AI markers (8 minutes per article)
5. Publishing cadence: Uploaded 7-10 articles daily to avoid triggering spam filters with bulk publishing

Each article followed this structure:

• 1,600-2,200 words (mean: 1,847 words)
• 5-7 H2 subheadings with H3 breakouts
• 3-5 Amazon product embeds per article
• FAQ section (4-5 questions, AI-generated)
• Internal links to existing cornerstone content (2-3 per article)

Derek did not use AI detection tools or humanization services. The articles were published as-is after the VA's light editing pass. He assumed high publishing velocity would outweigh quality deficits during the critical first 60 days.

The Traffic Surge: 18,400 Monthly Visitors in 45 Days

By day 15, the site's indexed page count jumped from 51 to 187 pages. Google Search Console showed 142 impressions per new article within the first 72 hours—significantly faster than the site's historical indexing speed (7-10 days per article).

Traffic growth trajectory:

• Day 1-15: 1,200 → 4,600 monthly visitors (283% increase)
• Day 16-30: 4,600 → 11,200 monthly visitors (143% increase)
• Day 31-45: 11,200 → 18,400 monthly visitors (64% increase)
• Peak day 52: 21,100 monthly visitors (17.5x baseline)

Top-performing articles captured featured snippets for queries like "best magnesium glycinate vs citrate" and "ashwagandha KSM-66 vs Sensoril comparison." These snippets drove 34% of total traffic during the peak period.

Revenue scaled proportionally:

• Day 1-15: $340 → $820/month
• Day 16-30: $820 → $1,650/month
• Day 31-45: $1,650 → $2,400/month
• Peak day 52: $2,800/month (8.2x baseline)

Amazon conversion rate held steady at 2.6-2.9%, indicating the traffic quality was legitimate. Bounce rate increased from 62% to 71%, but time-on-page metrics remained within acceptable ranges (1:42 average, down from 2:18 baseline).

The Algorithmic Penalty: March 2024 Core Update Deindexing

On March 5, 2024, Google launched its March 2024 core update, which explicitly targeted AI-generated content at scale. The update notes mentioned "reducing low-quality, unoriginal content" and "rewarding content with unique value."

Derek's site was hit on day 63:

• Day 63: Traffic dropped from 19,800 to 11,200 monthly visitors (-43% in 24 hours)
• Day 67: Traffic collapsed to 1,100 monthly visitors (-94% from peak)
• Day 70: 147 of 200 new articles removed from Google index
• Day 75: Manual review confirmed no manual action—this was an algorithmic devaluation

Google Search Console showed 89% of the deindexed articles flagged with "Crawled - currently not indexed" status. The remaining 11% retained indexing but ranked on page 4-7 for target keywords (effectively zero traffic).

Revenue collapsed in parallel:

• Day 63: $2,800 → $1,600/month
• Day 67: $1,600 → $380/month
• Day 75: $380 → $290/month (below baseline)

Derek ran diagnostics through Surfer SEO and Originality.ai. Results confirmed 96% AI detection scores across the deindexed articles. The content showed repetitive phrasing patterns, predictable subheading structures, and lack of first-person experience markers.

The Recovery Strategy: Content Pruning Plus Manual Rewrites

Derek's recovery plan focused on salvaging domain authority rather than rescuing the AI content. The strategy:

1. Prune 180 articles: Deleted all AI-generated content with zero traffic (day 80-85)
2. 301 redirects: Redirected pruned URLs to topically relevant cornerstone articles (67 redirects total)
3. Manual rewrites: Hired a supplement researcher at $120/article to rewrite the 20 highest-traffic AI articles with original research, personal testing notes, and comparison tables (day 90-120)
4. Disavow file: Submitted disavow file for 34 spammy backlinks acquired during the AI content period (likely scraped by link farms)
5. Fresh content cadence: Published 2 manually written articles per week for 16 weeks to signal editorial standards shift

By day 150, the site stabilized at 3,800 monthly visitors—3.2x the original baseline but 79% below the peak. Revenue recovered to $720/month (2.1x baseline).

Derek listed the site on Motion Invest in November 2024 for $18,000 (25x monthly profit multiple). It sold for $15,200 after 23 days—a 79% ROI on the $8,500 acquisition, but a catastrophic loss compared to the $84,000 valuation at peak revenue (30x multiple standard for Amazon affiliate sites).

Lessons From the Experiment: Where AI Content Fails at Scale

Derek's post-mortem identified five failure points:

1. Velocity triggers algorithmic scrutiny: Publishing 7-10 articles daily on a previously dormant site created a red flag. Historical publishing cadence for the domain was 3 articles/month. The 23x acceleration likely triggered quality review filters.

2. Lack of E-E-A-T signals: The AI content contained no first-person experience, no author credentials, no original research. Every article read like a summarization of existing web content—which is exactly what GPT-4 was trained to produce.

3. Structural homogeneity: All 200 articles followed identical H2/H3 patterns, FAQ formats, and internal linking structures. Google's pattern-matching algorithms likely flagged this as template spam.

4. No humanization layer: The $15/hour VA only checked factual errors and affiliate links. No effort was made to inject personal voice, original examples, or structural variation.

5. Over-optimization for snippets: Derek's prompt engineering explicitly instructed GPT-4 to write "concise answers suitable for featured snippets." This created repetitive answer-box formatting that Google's spam classifiers associated with low-quality content farms.

The experiment succeeded financially in the short term—Derek netted a 79% ROI—but destroyed 85% of potential value by triggering an algorithmic penalty that could have been avoided with lower publishing velocity and higher editorial standards.

Alternative Approaches That Might Have Worked

Three modifications could have extended the runway before deindexing:

Lower velocity: Publishing 3-4 articles/week (vs. 7-10/day) would have maintained domain trust longer. Total content volume after 90 days: 40-50 articles instead of 200.

Hybrid human-AI workflow: Use GPT-4 for research and outlines, but require human writers to draft original sections (30-40% of each article). This breaks structural patterns and injects authentic voice markers.

Quality filters before publishing: Run all AI drafts through Originality.ai, GPTZero, or Surfer SEO's AI detector. Only publish articles scoring below 60% AI probability. Rewrite sections flagged as machine-generated.

Derek's conclusion: AI content at scale works for 30-60 days on established domains, but the algorithmic penalty destroys more value than the short-term revenue generates. The optimal strategy pairs AI efficiency with human editorial oversight—not as a cost center, but as an insurance policy against deindexing events that can wipe out 80-90% of asset value in 72 hours.

FAQ

How long did the AI content rank before getting penalized?

The AI-generated articles ranked for 52 days before the March 2024 core update triggered mass deindexing. Traffic peaked on day 52, then collapsed by 94% over the following 14 days.

What was the total cost to produce 200 AI articles?

GPT-4 API costs: $340 (200 articles × $1.70 average per article). VA editing: $400 (200 articles × 8 minutes × $15/hour). Total production cost: $740.

Did Derek try to recover the deindexed articles?

No. He deleted 180 of the 200 AI articles and redirected their URLs. Only the 20 highest-traffic articles were manually rewritten. Recovery efforts focused on salvaging domain authority rather than rescuing the AI content.

What was the final ROI after selling the site?

Derek sold for $15,200 after an $8,500 acquisition and ~$2,000 in content costs. Net profit: $4,700 over 8 months (55% ROI). However, peak valuation would have been $84,000 at $2,800/month revenue—meaning the penalty destroyed $68,800 in potential value.

Would this strategy work with GPT-4o or Claude Opus?

Unlikely. Both models produce higher-quality output than GPT-4, but they still lack first-person experience and original research. Google's spam classifiers flag content patterns, not just quality deficits. Publishing 200 articles in 30 days would still trigger velocity-based scrutiny regardless of which LLM generated the content.