Test mass-produced AI pages cheaply: 60,000 GPT-generated pages and what crawlers did
A deliberate experiment built a 60,000-page site with a small OpenAI model for about $10 and recorded that AI crawlers requested it far more than Googlebot. The author expects it to be deindexed.
Evidence: The source shows its work: steps, screenshots, code or data you can inspect.
The business problem
Founders are tempted by programmatic SEO at near-zero cost and want to know what mass-produced AI pages actually attract.
What was tried
A Node script called gpt-4.1-nano for 300 topics across 200 countries, producing a title, meta description, five statistics, three analysis paragraphs and two FAQ items per page, with 50 parallel requests in under 30 minutes. Pages were served from Cloudflare Workers with a serverless SQLite database, and server-side tracking with user-agent and IP checks replaced client-side tracking because bots do not run JavaScript.
What was reported (mixed)
In the first 12 hours GPTBot made over 29,000 requests, and the author reports about 470 times more GPTBot requests than Googlebot and bots as about 98% of all requests, with the site found through its sitemap alone. No Google indexing or ranking data is reported, and the author predicts deindexing.
Limitations
One domain, no control site, and partial early data because tracking was added partway through. The 470x figure is stated for two different periods, human traffic is not quantified, and the deindexing prediction is an expectation, not a result. The author calls the experiment not a business and concludes that programmatic SEO for traffic alone is a dead end. The code was promised but not yet published.
What you need
An OpenAI API key, Node, and Cloudflare Workers and D1. Cost is given as about $10 for generation plus about $10 a year for the domain.
Sources
- Metehan's GEO/SEO & AI/ML (Substack) ↗ Firsthand write-up, published March 4, 2026
Source published: March 4, 2026. Last reviewed here: October 11, 2026. Spot a mistake? Tell us.
Tools in this workflow
- OpenAI API: Generates every pageTry OpenAI API
- Cloudflare: Hosting on Workers and D1Try Cloudflare
Related workflows
Set up Hermes Agent on a cheap VPS with Telegram, morning cron briefings and a reusable content research skill
A 14-day test of Hermes Agent on a small VPS, reached through Telegram, with memory files, a daily and weekly briefing, and custom skills for content research.
Chain Hermes agents through Telegram into a research, writing, editing and publishing team
A business-automation blog shows how to let several Hermes agents message each other in Telegram so a research agent can hand work to a writing agent in a content pipeline.
Build a Claude Code subagent that writes SEO and answer-engine research briefs with Perplexity and Firecrawl
A step-by-step build of a Claude Code subagent that, given a post or topic, gathers search trends and search results and writes a research brief with keywords, snippet opportunities, competitor gaps and a suggested title and slug.
Decide what to write next by checking Search Console queries and competitors with an MCP-driven SEO research agent
A newsletter writer built a coding-agent workflow that starts from his own Search Console queries, checks demand and competitors through a keyword tool, and labels each idea as write, update, expand, test or ignore.