AI & Automation Tools I Actually Use (and Recommend). Learn More
OARS — Open Agent Readiness Standard
OARS describes what it means for a business to be useable by AI — discoverable, readable, interactable, transactable, and operable end-to-end. It's free to read, free to implement, and lives at Knov.ai — I wrote and maintain it.
Agents are becoming a new kind of visitor to your website. Most sites aren't ready for them — not because anyone did anything wrong, but because the question is new. OARS gives that question a vocabulary, a scale, and a sequence of concrete steps — the same way HTML and robots.txt did for the human web a generation ago.
Each level requires full compliance with everything beneath it.
Level 1
An agent can locate and identify your business and its offerings.
Level 2
An agent can comprehend what you do, who it's for, and how it works.
Level 3
An agent can execute defined actions — booking, quoting, querying — against your systems.
Level 4
An agent can complete commercial transactions end-to-end on a customer's behalf.
Level 5
Agents can run internal workflows — not just customer-facing actions.
The five levels are written for engineers, but the work each one represents is mostly business work — decisions about what you'll let agents see, do, and transact on your behalf. Here's the plain version.
Level 1 · Discoverable
Your site declares — in machine-readable terms — that it exists, who runs it, where you operate, what you sell, and which AI agents you're happy to receive. Most of this is small, one-time work on robots.txt, a sitemap, Schema.org markup, and an llms.txt file. No new infrastructure, no new logins.
Roughly: a couple of days for most small businesses, sometimes already partially done by a good SEO setup.
Level 2 · Readable
Beyond "this business exists," agents need structured details — services, products, prices, hours, policies, the kinds of customers you serve, the kinds you don't. This is where Schema.org gets used in earnest, FAQs become machine-readable, and a knowledge declaration spells out what you can be trusted to speak on.
Roughly: one to three weeks of content and schema work, depending on catalog size.
Level 3 · Interactable
This is the line where engineering shows up. An MCP server, an agent card, OAuth with PKCE, rate-limited and well-described tools — booking an appointment, getting a quote, querying inventory. The actions are scoped, observed, and reversible. It's the first level that makes someone responsible for an integration on your side.
Roughly: a real project — four to twelve weeks depending on how many actions you expose.
Level 4 · Transactable
The agent doesn't just request a quote — it pays, books, ships, and gets a receipt. This needs a commerce protocol (ACP, UCP, x402, or equivalent), live pricing and availability, delegated authorization with scope and spending limits, signed agent identity, no CAPTCHAs in the agent path, and audit trails an accountant would actually accept.
Roughly: a meaningful build — typically only worth doing once Levels 1–3 are stable and an agent-driven sales channel is real.
Level 5 · Operable
This is the inside view: at least one core workflow — intake, scheduling, quoting, triage, processing — is substantially operated by agents, with human oversight, documented handoff rules, and metrics that distinguish agent work from human work. Most businesses will live happily at Level 2 or 3 for years. Level 5 is for operators who want to compound.
Roughly: an ongoing program, not a project. It changes how the company is staffed.
I kept hearing the same question from businesses, agencies, and engineers: how do we make a website actually work for AI agents? Every vendor had a partial answer. Nobody was drawing the whole map.
Standards are how the web absorbs shocks. HTML let any browser read any page. robots.txt let any site say "here are the rules for crawlers." Schema.org let search engines understand what a page was about without guessing. Each of those started messy and informal, and each eventually became invisible infrastructure. Nobody at a small business has to think about which sitemap format to use anymore; the question is settled.
The agent web needs the same kind of settling. OARS is an attempt to draw the lines clearly enough that the conversation can stop being "should we do something about AI agents?" and start being "where on the scale do we want to be, and by when?"
This page is the consultant's overview. The standard itself, the tooling, and the verified directory all live at Knov.ai, where they'll stay free and open.
The v1.0 specification, with the exact requirements for each level, the verification rules, the appeals process, and the policy for regressions. The thing engineers actually read.
A no-signup scan that reads what an agent would see on your site and returns a quick Level 1–2 readout. Useful before any conversation about budgets or timelines.
The JSON Schema, example files, and a generator for oars.json — drop-in for your team to start from instead of writing it by hand.
Independent review against the standard, with a listed entry in the public directory once you pass. The directory is what an agent looks at to decide who it can trust.
OARS is not a replacement for any of these — it's the scaffolding that connects them.
SEO is about humans finding your site through search engines. Level 1 of OARS overlaps with SEO substantially; if your SEO is good, you've already done meaningful work on discoverability.
Schema.org is the markup vocabulary OARS leans on heavily for Levels 1 and 2. It's the existing tool, used more deliberately and in more places.
MCP (Model Context Protocol) is the action layer — how an agent actually invokes something on your side. OARS specifies that you have an MCP server at Level 3, and what it must minimally do, but doesn't redefine MCP itself.
llms.txt and AGENTS.md are content-navigation conventions that emerged separately. OARS includes them where they belong and clarifies what each is actually for.
The point of a standard is that you can absorb new conventions as they appear, without having to re-architect from scratch each time.
Most of these come up in the first 15 minutes of any conversation about OARS, so they're collected here.
No. OARS is an open standard, not a law and not a platform requirement. Nobody can fine you for not implementing it. The point is that it gives you — and any agent visiting your site — a shared way of describing where you are, so you can decide which level is worth aiming for and stop there.
Almost none. Levels 1 and 2 are within reach for any small business and meaningfully improve how AI represents you. Level 3 starts paying off once a customer-facing action is worth automating. Levels 4 and 5 are for businesses where agent-driven commerce or operations are real strategic bets, not the norm.
"SEO for AI" is mostly Level 1 and parts of Level 2 — getting found and described accurately by ChatGPT, Claude, Perplexity, and whatever comes next. OARS covers that, but also the parts SEO never touched: taking actions on your systems, completing transactions, and running internal workflows. The visibility piece is the entry point, not the destination.
Anyone can run the automated checks and see where they stand — those parts are deterministic. Formal verification (the kind that gets you into the public directory) is a hybrid: automated checks plus a human reviewer at Knov.ai. Level 3 and up include live demonstrations against the actual systems being claimed, not just paperwork.
Yes, in roughly the same way HTML and Schema.org change — versioned, slowly, with backward compatibility taken seriously. v1.0 is what's published now. The cadence is conservative on purpose; standards that move every quarter are useless to anyone actually running a business.
I wrote OARS and maintain it. Knov.ai is where the standard, scanner, schema, and directory live — that's the open and free side of the project. This site is the consulting side: where I help businesses actually implement what the standard describes. The two are intentionally separate, so the open standard never depends on a single consultant being around.
Run the free scanner. It takes seconds and gives a quick read on Levels 1–2 without any signup or sales call. If the results suggest meaningful work is needed and you'd rather not project-manage it yourself, the audit is the next step.
Where to from here
Run the free scanner for a quick OARS Level 1–2 readout — takes seconds, no signup. Or if you'd rather have someone walk it through with you, the audit is a guided, fixed-price way to do that.