The Fleet Audit
Once a week we check every site we manage for the signals that actually make a person findable and trusted by Google and AI — the things a website builder never looks at. One pass, every site, every signal. Here is exactly what we check and how it runs.
A pretty website isn’t the job
A website builder hands you a nice-looking page and calls it done. But a nice page that Google can’t read, that has no structured identity, and that no other profile points to is invisible — you are still outranked by people who are worse at the work but better at being found. Findability is a system, not a template. The audit is how we verify that system is actually in place, on every site, continuously.
What we check
Eleven signals today, grouped the way they build on each other — from “does it even load” up to “is it a recognized entity wired into a network of proof.” The list grows; the method doesn’t.
How it runs: audit once, dispatch many
The naive way is one scheduled job per check — which re-crawls all 200 sites for every signal. We don’t do that. One read-only pass checks everything, writes the gaps, and hands them to cheap AI agents that fix only what’s missing.
It runs in Anthropic’s cloud, not on anyone’s laptop, on the cheapest model that can do each step — deterministic scripts for the mechanical parts, a mid-size model for judgment, the flagship only for genuinely new work. Keeping ~200 sites in spec costs single-digit dollars a week.
The method, in the open
We don’t hide how this works — transparency is the proof. This is the literal check registry from our audit script; add a line, gain a signal:
CHECKS = { "homepage_up": chk_homepage_up, # reachable + valid HTTPS "has_title": chk_has_title, # legible "meta_description": chk_meta_description, # legible "indexable": chk_indexable, # not blocking Google "og_image": chk_og_image, # shareable "entity_schema": chk_entity_schema, # Person/Org schema present "sameas_links": chk_sameas_links, # profiles declared to Google/AI "hub_post": chk_hub_post, # has the hub page "hub_has_wheel": chk_hub_has_wheel, # the footprint wheel is on it "hub_links_master": chk_hub_links_master, # rolls up to the master teaching "home_links_master": chk_home_links_master, # homepage references the hub }And this is the playbook a worker agent follows to fix a gap — the same discipline every time: fetch the homepage; skip parked/template/non-person sites; extract the real social profiles (never invent a follower count); build the footprint wheel; publish via a revocable API token, never a raw password; verify the live page links back to the master. Honest about what it can’t do, too — which is exactly what makes the rest trustworthy.
