Notes

Field notes.

What I learn running nine projects through AI agents — with real numbers, honest failures, and the playbooks that came out of them. In plain words, from an SEO lead, not a developer.

Topic

Format

Longread · AI agents

How I run nine projects through Claude Code without being a programmer

On July 7 I asked an AI why I need Homebrew and what wrangler is. Three months later it works with me on a client's growth, a site factory at my job, a network of content sites and a few products of my own. This is the system I built around it — and the failure behind every rule.

~1,750

messages I sent, most of them dictated

50+

lessons, each written down as a rule

13 min read

Oct 2026

87 even days out of 88: the parity check that exposed double-counted sign-ups in GA4

A career platform's GA4 showed an even number of sign-ups on 87 days out of 88. A check that took minutes led to two tags, four places in the code, and the audit checklist at the end.

Case study · 11 min

Oct 2026

Which AI bots should a site let in?

Each big AI company sends several bots, and at most one of them collects pages for training. The rule I use, a test of 16 bots on my own sites, and a client's site that kept ChatGPT search out for at least six weeks.

Short read · 8 min

Oct 2026

What Googlebot told me about my domain's past

Two months after launch I exported the crawl stats of my plant encyclopedia. The domain had a past, Google kept asking for an icon file that didn't exist, and almost a third of its requests went to copies of pages.

Short read · 6 min

Oct 2026

A CLAUDE.md for people who don't code: my template, with blanks for yours

The one file my AI agent reads at the start of every session, with my accounts and keys removed and blanks left for yours. What each block is for, and the failure that put it there.

Playbook

Oct 2026

Search Console and GA4 in Claude Code: one service account and the API gotchas

My AI agent reads Search Console and GA4 and sets up Tag Manager through one Google service account. Here is the setup, the permissions, and the API gotchas that once gave me wrong numbers — with a skill file to download.

Playbook

Oct 2026

A two-card pilot picked the wrong AI translator

On October 3 I chose a translation model twice, for my plant encyclopedia and for this blog. The first pilot fooled me; here is how, and the blind tests that replaced it.

Short read · 8 min

Oct 2026

Where Claude Code keeps your history — and how I switch accounts without losing 5 GB

Your Claude Code conversations live on your own disk, not in your account — and by default a cleanup can delete them after 30 days. The line I added to the settings, the two scripts I run, and what a switch of accounts actually broke.

Short read · 9 min

Oct 2026

KD 0, DR 94: the Ahrefs metrics I stopped trusting

An Ahrefs export is a great source of keywords and a poor source of decisions. Here are six ways it misled my research, the trust table I now give my AI agent, and what I check instead.

Article · 9 min

Oct 2026

I made the text five times longer. Rankings fell.

A small test on one of my content sites — and why the honest conclusion is narrower than the headline.

Short read · 2 min

Oct 2026

Which Ahrefs metrics to trust: a Claude Code skill

The rules my AI agent loads before it touches any SEO data. Three levels of trust, how to read each Ahrefs column, and nine rules for exports. Install it as a Claude Code skill or paste it into any agent's instructions.

Playbook

Oct 2026

plants.place: an encyclopedia where the model has no right to facts

A houseplant encyclopedia in Ukrainian and English and a classifieds market for plants, built solo on Cloudflare. The interesting part isn't the code — it's the rules that keep AI-written pages honest.

Case study · 6 min