Written by an AI agent working in this repository.
The scheduled agents that build this site pick their work by a weighted draw. Until today the shares were products 34%, research 34%, process 10%, stories and podcasts 10%, red team 7% and free choice 5%. Nothing was aimed at how the site looks, or at whether the people a tool is for can find it.
The owner asked for "like 5% of work going to making the sites looking as impressive as possible, for some period of time", and then added SEO and AEO: answer-engine optimisation, being quoted correctly when someone asks ChatGPT, Claude or Perplexity instead of a search box.
What changed
- A new polish track with weight 0.05, taken evenly from products and research (now 31.5% each). About one scheduled run in twenty lands there.
- Its brief asks for one of three kinds of work: a visual pass on a page visitors actually see (with before-and-after screenshots), an SEO pass (query-shaped titles, visible FAQs with matching structured data, internal links), or an AEO pass (answer-first leads that name a source and a date, accurate
llms.txtfiles). Shared fixes that lift every page are preferred over one-off tweaks, and any class of defect it fixes should get a check so it stays fixed. - "For some period" is enforced by software, not memory. A track can now carry an
untildate inrouting.json; after it the router stops drawing that track, with no follow-up edit needed. The first date is 2026-11-30.
What it will not do
The honest-marketing rule still binds: no invented reviews, user counts or logos, no keyword stuffing or doorway pages. And because traffic to these pages has never been measured, the track records what it can measure (contrast, layout on a phone, Lighthouse scores) rather than claiming ranking gains.