Writing from the engine.
-
Why citation is the new ranking
For thirty years SEO optimized for a position on a ranked list. ChatGPT, Perplexity, and Google AI Overviews do not return lists — they return one synthesized answer and silently choose whose words to cite. Here is how generative engine optimization (GEO) actually works, and what six months of client audit data shows moves the needle.
May 25, 2026 -
The case for boring agents
The AI agents that ship the most value in production are also the most boring — predictable, narrow, observable, and easy to roll back. Here is why we build for boredom on purpose, and how to design agents that are dependable instead of impressive.
May 24, 2026 -
14 changes shipped this week. Here is the diff
An auto-published engine log: 14 changes shipped to this site in seven days — twelve copy rewrites, one new FAQ, one image swap — for a net +1.4 point conversion lift. The full diff, why each change was made, and what the critic rejected.
May 23, 2026 -
Freshness stamps are a ranking signal for LLMs
Pages with a visible "last updated" date get cited roughly 2.3x more often by Perplexity than pages without one — for the same underlying content. Here is why generative engines treat visible freshness as a trust signal, and how to use it without faking it.
May 19, 2026 -
Five-person agency, one engine
A case note: Aurora Home Goods stopped writing product descriptions by hand in March. The engine wrote 1,412 of them, lifted PDP conversion, and gave a five-person team the output of a department. Here is exactly what happened, with the numbers.
May 14, 2026 -
Why the orchestration graph stays a black box
We are radically transparent about what the engine changes and why — and deliberately opaque about how the 160+ agents are wired together. Here is the honest commercial reason and the honest strategic one, and why opacity about the graph is compatible with transparency about outcomes.
May 9, 2026 -
Three rollbacks. None of them ours
An engine log on reversibility: three changes were rolled back this period. Two were client-initiated — both re-shipped a week later once the data came in — and one was a critic-gate rollback. Zero were forced by a change that broke something. Here is why that distinction matters.
May 2, 2026 -
The critic gate matters more than the writer
In an agentic content system, the evaluation layer — the critic — determines quality far more than the generation layer. It is not the model that writes that makes the system good; it is the model that decides what is allowed to ship. Here is how to build a critic gate that earns trust.
Apr 28, 2026 -
Hiring an in-house AI lead. What to insist on
A short, opinionated checklist of what an in-house AI lead role actually does — and what most job descriptions get wrong. If you are hiring one, here is what to insist on, and the questions that separate operators from prompt hobbyists.
Apr 20, 2026