Rubric v2.3.0
All 38 checks behind a Qomvia score, grouped by pack. Only the packs a site qualifies for are applied, so a content site is never scored on a checkout it does not have, and every score runs to 100.
If GPTBot, ClaudeBot or PerplexityBot are disallowed, agents cannot read the site at all.
Bot management often returns a challenge page to anything without a browser fingerprint.
Agents that fetch HTML without executing JavaScript see nothing on a client-rendered site.
Agents have no credentials; a wall on the public content removes the site from answers entirely.
Agents extract the main block; when navigation dominates the HTML, the answer cites the wrong text.
Title, meta description and a canonical URL are what an answer engine attributes a citation to.
Headings are how retrieval splits a page into passages that keep their meaning.
Agents answering in German need to know which of your pages is the German one.
Freshness is a ranking input for answer engines; undated pages lose to dated competitors.
A sitemap is the fastest way for an agent platform to enumerate a site.
llms.txt is the emerging convention for pointing an agent at the canonical content.
A discoverable MCP server or agent card lets assistants call your functions directly.
An OpenAPI description or a feed is what turns a website into something a system can consume.
Organization/WebSite markup is how an agent links a domain to an entity it knows.
Agent platforms need an explicit permission signal before acting on a user's behalf.
Agents and their operators need a programmatic way to resolve problems and request access.
Agent orchestrators time out aggressively; slow origins get dropped from consideration.
Multi-megabyte HTML blows up agent context windows and parsing cost.
RFC 8288 Link headers let an agent find machine resources from the response itself.
Serving text/markdown on `Accept: text/markdown` cuts agent token cost by most of the page.
A Content-Signal line in robots.txt or a response header states what AI may do with the content.
The HTTP message signatures directory lets a site verify signed agent requests.
An Agent Skills index or WebMCP tools tell an agent what it can do here and how.
/.well-known/api-catalog lists every API with its spec, docs and status.
RFC 8414 / RFC 9728 metadata, or an auth.md, tells an agent how to authenticate.
DNS-AID records and ARD are the discovery layer above MCP and A2A.
Many sites allow the homepage but disallow /products or /cart-adjacent paths agents need to read prices.
Product JSON-LD is the only reliable way an agent learns what you sell without guessing.
Price and availability are what an agent compares; in rendered markup only, they are invisible.
GTIN/SKU/MPN is how an agent knows your product is the same product it saw elsewhere.
ItemList markup lets an agent enumerate a category instead of guessing product URLs.
A feed is the cheapest way for an agent platform to ingest your whole catalogue.
QMP is the protocol Qomvia agents buy through: discovery, order creation, merchant payment URL, attribution.
ACP is what lets ChatGPT complete a checkout against your store instead of linking out.
x402, MPP, UCP and AP2 are how an agent can pay without a human-driven card form.
Agents need a linkable, crawlable cart URL to hand a basket back to the user.
A mandatory account or OTP is where most agent purchases die.
autocomplete tokens and input names are what let an agent fill an address without guessing.
This page says what is measured. Which of these checks your site fails, the evidence we captured, what to change and the snippet to paste live in the signed-in report for a site you own.
See what the report includes