In early February 2026 I set out to stand up a small portfolio of content sites at once, on one stack, with an AI agent doing most of the writing under my direction.

Within about a week there were five: aimarketingpicks.com (AI marketing tools), remoteworkpicks.com (remote work gear and software), soflotimes.com (South Florida local), anonark.com (privacy and offline AI, the blog for a product idea), and quicksummit.net (my automation consultancy, rebuilt in Astro). By February 11 they held 66 articles between them.

This is what the following weeks taught me.

The setup

What worked

One template, cloned. Same stack everywhere meant one fix applied to every site. When I found an SEO problem on one, I fixed it on three in the same pass.

Near-zero running cost. Static hosting on the free tier meant the only real cost was domains.

Volume was trivial. The content engine added articles every day without me touching a keyboard. That turned out to be a problem.

Lesson 1: Published is not indexed

On February 11, the four new sites had 64 articles live and zero pages indexed.

The technical basics checked out: properties verified, sitemaps returning 200, robots.txt clean, pages server-rendered as static HTML, canonicals correct, no stray noindex.

Four days later the first impressions showed up: about 200 impressions and 7 clicks across the four sites combined. aimarketingpicks.com had 6 pages indexed and 75 not indexed, 72 of those stuck in "Discovered, currently not indexed."

My read was sitemap bloat. The template shipped tag pages, pagination, search and archive pages in the sitemap, so Google found a pile of thin URLs on a brand-new domain with no links. I filtered the sitemaps to real content pages and noindexed the thin templates on the three sites that had the problem.

On a new domain, every URL you hand Google is a vote on how much of your site is worth crawling, so hand it fewer, better ones.

Lesson 2: Automation without guardrails eats your work

I'd added a timestamped YouTube embed by hand to one article. A few days later it was gone. A scheduled content job had overwritten the file with a fresh version, because it treated the blog folder as empty. The file had never been committed to Git, so there was nothing to restore from.

Fixes, same day: a content registry tracking every article across all five sites, jobs that read it before writing and never overwrite an existing file, and Git commits for the articles.

If an automated system can write to your site, assume it will eventually write over something you care about. Build the guardrail before you need it.

Lesson 3: Silent failures are the default

Twice in the first two weeks, the jobs simply stopped. Once the scheduler wasn't firing overnight. Once the agent sessions running the jobs filled their context window and stopped responding. Nothing errored loudly; the output just didn't appear.

Now I check that a job ran, not just whether something got published. "No new post" and "the job is dead" look identical from the outside.

Lesson 4: Content alone doesn't build authority

By mid-February every site still had zero backlinks. That was the biggest growth blocker on the list, and no amount of extra publishing fixes it. New domains with no links get crawled slowly and trusted slowly.

Lesson 5: Volume is the easy part

Speed was never the constraint. Any AI setup can produce more articles than a new site can earn crawl budget for. The real question is whether each article deserves to be indexed: is it sourced, is it specific, does it say anything the top ten results don't?

In February I measured articles published when I should have measured articles worth publishing.

What I'd do differently

Update, October 2026

Lesson 5 turned out to be the big one.

In July 2026 I rebuilt the content system from scratch as an autonomous pipeline: research, brief, draft, quality gate, publish, deploy. Before reusing the February backlog, I ran the old pre-written articles through the new pre-publish gate. It rejected all 68 of them, for unsourced claims, derivative takes, and stale facts. The topics were mostly fine, but the articles weren't.

Those topics now feed back into the pipeline to be rewritten with real sources. Since July it has published about 70 articles across six sites, and each one had to pass the same gate the February content failed.

Should you do this

If you want to try a small portfolio of your own: