AOS/APS Manifesto · Open standard
In the era where machines decide, the most human brand wins. And the most human is the one that lets a machine prove it.
Why this standard exists, in seven points.
The web is splitting in two.
There is the web humans see — HTML, design, persuasion — and the web agents read — protocols, structured data, verifiable claims. Your customers are starting to arrive through the second one. They don't browse. They probe.
Agents don't trust marketing. They verify it.
A human can be moved by a good slogan. An agent checks. It reads your llms.txt before your homepage and your /.well-known/ before your docs. It weighs what it can verify over what you claim. In the agentic era, the humanist truth survives: the most trustworthy brand wins — and now trust is auditable.
Being readable is not being preferred.
The industry has standards for readability: is the file there, can the agent parse it, can it call your API. That is operability, and it is becoming a commodity. Everyone will have it. It answers «can an agent use this?» — never «why should an agent choose you?».
Preference is decided by verifiable evidence.
When an agent recommends a brand, adds it to a cart, or names it in an answer, it is making a bet on your behalf — on the strength, consistency and verifiability of the evidence around you. We tested it: the same claim, signed and verifiable, moved a frontier model's trust and its willingness to recommend by more than twenty points. Provenance is not a vibe. It is a signal, and it is measurable.
So we measure two things, not one.
AOS (Agent Operability Score): can an agent operate you? APS (Agent Preference Score): why should an agent choose you? Operability is the price of entry. Preference is the moat.
Truth over theater.
This standard rewards evidence, never assertion. A claim counts only when a proof backs it. A proof's weight comes from who signs it and whether it's public — not from a number you picked. Missing data is marked missing, never invented. And a boundary — telling an agent when your claim does not apply — raises trust, because it removes the agent's risk of recommending you wrong.
Climb to the top of the pyramid.
Most brands live at Level 1 (prose) or Level 2 (schema you wrote yourself). The best reach Level 3 (the same fact agrees across independent sources). Level 4 — a cryptographically signed, independently verifiable claim — is almost empty. It costs 64 bytes and sub-millisecond math. That gap is the whole opportunity, and it is open today.
The full ladder
The experiment
points of trust from a frontier model when the same claim is signed and verifiable, vs. the unsigned version. Provenance is a measurable signal, not a vibe.
The verifiability pyramid
The bet, in one line
In the era where machines decide, the most human brand wins. And the most human is the one that lets a machine prove it.