← All use cases

I built Keito as a vibe coded side project at $0.

SEO Agents X (Twitter) AI marketing agentsAI for SEO

The play

Use AI agents to write SEO pages grounded in product docs and real user data, then monitor Reddit for unanswered questions to answer directly.

How it works

  1. Feed AI agent product documentation, API reference, support questions, and analytics on actual user workflows
  2. Identify low-difficulty keywords with real search volume in new categories that lack existing answers
  3. Have agent write pages against source material rather than from scratch
  4. Set up alerts for Reddit mentions of competing tools
  5. Answer Reddit questions directly where you would have commented anyway
Result reported by Sion

$4,980.03 in July with no ads, no cold email, minimal content

Tools mentioned
Claude Code
Good for
Founders in new product categories with emerging demand

Try this first

Compile your product docs, API reference, support tickets, and user workflow analytics as source material for the agent

Breakdown by RunAgents from Sion's original post, shown below.

The original post

Sion
@Sion_Smith
I built Keito as a vibe coded side project at $0. In July it did $4,980.03. No ads, no cold email, hardly any content. Just two channels, both run by agents. Here's the playbook:Channel one is 103 pages I did not write and mostly have not read. Everyone's instinct with programmatic SEO is volume: spin up 500 near-identical pages off a spreadsheet and hope. That worked in 2019. Now it gets you wiped, because Google rewards the opposite. Their framework is E-E-A-T: experience, expertise, authoritativeness, trust. In plain terms, depth. One page that actually answers the question beats forty that circle it. Which is a problem when an agent writes the pages, because agents are good at length and terrible at depth. The fix is that the agent never writes from nothing. It writes against the Keito docs, the API reference, real support questions, and analytics showing which workflows people genuinely use. Real source material is the whole difference. Then the harder half: choosing what to write about. You want low difficulty and real volume. In a brand new category the short obvious keywords are still sitting there unclaimed. Example: someone types 'how do I bill a client for work Claude Code did.' That is a live problem with a credit card behind it, and nobody had written the page, so there was nothing to outrank. Most of the pages still die, and that is the model, not a failure. You are buying lottery tickets that cost almost nothing and pay for years. The dead pages are the price of the live ones. One more thing almost nobody is doing: llms.txt and heavy docs. People increasingly find tools by asking an assistant instead of searching. That is a different index entirely, and right now it is close to empty. Channel two is Reddit. I do not automate the posting, I automate the listening. An alert fires when someone mentions the tools we replace, and I go answer it myself. The rule: only comment where I would have commented anyway. Being the person with a real answer converts; being the person with a link gets you removed inside an hour. Both channels only work because the category is new. The pages are unwritten and the questions are unanswered. That window closes.
265 likes · 49k views·13d
View on X (Twitter)