avowd 青眼 — vouched for by the machines
avowd· client management & AI-visibility tracking
avowd 青眼Pricing

Changelog

What shipped, and when.

Dates are deploy dates. Nothing here is a plan — if it is on this page, it is running.

Published

Monthly data notes, generated from the measurements themselves

Every month the measurement run now produces a public note on what nine engines actually answered, aggregated across all measured businesses and naming none of them. Every figure is computed from committed measurements before any prose is written, and a verification pass rejects the article if it contains a number the data does not support.

Fixed

Visitor analytics for client sites

First-party, cookieless analytics on the sites we build, so the monthly report can finally answer “how many people came, and where from” — including how many arrived from an AI assistant, which is the number this whole practice exists to move.

Changed

Rate card rebuilt for clients outside Vancouver

The old packages assumed a Metro Vancouver restaurant with a menu and a photo shoot. Actual clients turned out to be a Kenyan conservation nonprofit, a bait shop in California and an accounting practice — so pricing was split into a core that works anywhere and modules that only some clients need, with an interactive quote that adds itself up.

Fixed

A slow engine no longer looks like an unpopular brand

One engine was returning answers for only 17 of 44 questions while the account had credit and reported no error — it was simply slower than the timeout. Raised the ceiling, and the console now flags any engine whose delivery rate drops, because a failed engine and a genuinely invisible business produce the same-looking zero.

Published

Chinese-language site at avowd.ai/zh

A written Chinese site, not a translated one — different examples, different objections, and the Chinese engines named as themselves.

Changed

Reports and alerts now send from our own infrastructure

Monthly result notices moved onto Cloudflare's email binding — one less third party holding client data, and one less API key to lose.

Fixed

Question drafting stopped avoiding the client's own category

Questions must never contain the brand name — otherwise the measurement tests whether an engine can read a website rather than whether it recommends the business. The filter enforcing that was splitting names into words, so a shop called Big Fish Bait & Tackle had “fish”, “bait” and “tackle” banned from its own questions. Full names and domains are filtered now; ordinary category words are not.

Changed

Measurement moved to durable execution

Runs used to depend on a browser tab staying open. They now run as durable workflows: a failed slice retries on its own, an interrupted run resumes where it stopped, and closing the laptop stops nothing.