Brain Dumps

My coding agent runs a cheap model to save money. The trouble is the moments a cheap model quietly gets wrong. So I kept a smarter one on the bench — until the extension that was supposed to fetch it kept dying on long sessions. Here's what I built instead.

Twelve weeks of replacing every part of my SEO workflow with AI. This is the full system — keyword research, brief generation, schema, internal linking, monitoring, citation tracking. The pieces, how they connect, and the parts that still need a human.

In the App Router you don't hand-write meta tags — you export a metadata object or a generateMetadata function and Next.js handles the head. Here's how the whole system works, from static objects to dynamic OG images.

Title under ~60 chars, keyword-first. Meta description ~120–160 chars, written to earn the click. Google rewrites descriptions a lot — so write for the times it doesn't.

You don't need to duplicate everything. X/Twitter falls back to Open Graph when twitter: tags are missing — so implement OG as the foundation, then add two Twitter-specific tags and call it done.

Ahrefs costs €400/month. Here's the workflow I actually use — Claude Code orchestrating DataForSEO's pay-per-call API, clustering the output, and spitting out a brief that doesn't read like a SaaS demo.

Duplicate title tags come in two flavours — two elements on one page fighting each other, or the same title copied across dozens of pages. Both hurt rankings. Here's how to find them and fix them for good.

The real signs of AI writing — delve, em-dash clusters, rule-of-three, copula avoidance — plus the AI-code tells nobody talks about, and how to score and fix text.

Google's June 2026 update calls llms.txt 'fine' to use — but says it won't help (or hurt) Search. Ahrefs found 97% of llms.txt files get zero bot requests. The real story.

By mid-2025, ~35% of new websites were AI-generated. But the data on whether everything 'looks the same' is more surprising than the meme. Here's what's real and what isn't.

1200x630 px, 1.91:1 ratio — that's the baseline. But the safe zone, the file-size ceiling, and the width/height attributes crawlers need are what most guides skip. Here's all of it.