Writing
Notes on AI citation, search behavior, and what we learn scanning the web for a living.
What ChatGPT actually cites when you ask about developer tools
We ran 40 developer-tool prompts through ChatGPT for eight weeks and logged every source it named. Official docs won more often than we expected — but only when they answered the question in the first paragraph.
Why Perplexity cites differently than ChatGPT, and what that means for your docs
Perplexity leans on recency and structure in a way ChatGPT doesn't. We compared citation patterns across 200 prompts to find out what moves the needle for each.
Structured data doesn't fix bad writing, but it does help good writing get found
We added FAQPage schema to a page that was already answering the question well, and to one that wasn't. Only one of them saw any change.
A methodology note: how we run 210 prompts a week without asking permission
The exact scan cadence, retry logic, and how we handle a model that starts rate-limiting mid-week. Published because this audience will ask.
Choosing 40 prompts instead of 4,000: a case for a smaller, sharper prompt list
More prompts sounds like more signal. In practice it dilutes every recommendation across questions nobody actually asks. Here's how we size a prompt list.
Google AI Overviews cites almost nothing you'd expect from ten years of SEO
Ranking first in Google no longer means winning the AI Overview above it. We measured the overlap, and it's smaller than most teams assume.