Agentic Resource Discovery Makes AI Search a Protocol Problem
Google and Microsoft-backed ARD turns AI agent discovery into a protocol and source architecture problem.
Agentic Resource Discovery is the clearest sign yet that AI search is becoming a protocol problem, not a content problem. Google and Microsoft are backing a draft specification that lets AI clients find tools, skills, agents, APIs, workflows, and MCP servers across the web without hardcoding every option into context.
Agentic Resource Discovery shifts AI search from pages to capabilities
The Agentic Resource Discovery specification treats the web as a searchable inventory of machine-usable resources. Google announced ARD on June 17, 2026 as an open specification for finding and verifying "tools, skills, and agents across the web" because agents now need to answer where a capability lives, which capability to use, and how to verify it before use (Google Developers Blog).
Microsoft described the same spec as a way for AI clients to find task-relevant resources automatically, developed with collaborators including GoDaddy, Google, Hugging Face, and others (Microsoft Command Line). That is the main change. Search no longer stops at finding a document a human might read. It starts selecting a resource a machine might call.
The official ARD specification names the resource set plainly: tools, Skills, MCP servers, APIs, workflows, and agents. It also defines the core separation: ARD helps an AI client find the right resource, while the resource is invoked through its own native mechanism (ARD specification).
AI agents need discoverable source architecture, not just optimized copy
ARD makes source architecture visible as a competitive layer. In the ARD model, a publisher exposes machine-readable capability records through catalogs and registries. The useful question for brands is no longer only "will a crawler index this page?" It is also "can a machine understand what this source is, who controls it, what it can do, and when to use it?"
That pressure matches the broader Machine Relations frame: brands are now competing inside machine-mediated discovery systems where legibility, retrieval, trust, and citation all matter. The practical implication is simple. If agents are moving toward formal discovery layers, vague brand pages lose value. Clear entity data, citable claims, and owned source structure gain value.
An adjacent Internet-Draft for Agent Discovery Protocol makes the same direction explicit from a different standards path. It proposes a three-layer discovery mechanism using DNS TXT and SRV records, well-known URIs for machine-readable metadata, and landing pages with communication channels, with DNS acting as a decentralized trust anchor (IETF Datatracker).
The real contest is resource selection
Agent discovery changes the visibility problem from "can the model mention us?" to "can the system select us?" That is a harder bar. Mentions can come from broad text retrieval. Selection requires a machine to resolve the resource, compare it against alternatives, trust the publisher, and decide the resource fits the task.
The ARD spec's design points toward that selection layer. It includes federated discovery, resource catalogs, registry search, machine-readable manifests, identifiers, and conformance testing. The IETF ADP draft takes a parallel path through domain-owned discovery and metadata. Different proposals, same signal: the agentic web needs structured discovery rails.
| Layer | Old search question | Agentic discovery question |
|---|---|---|
| Content | What page answers the query? | What resource can perform or support the task? |
| Entity | Who published the claim? | Which domain or publisher controls the resource? |
| Trust | Is this page credible enough to cite? | Is this capability verified enough to invoke or recommend? |
| Measurement | Did the page rank or get clicked? | Was the resource found, selected, cited, or used by an agent? |
That table is why this is more than developer plumbing. Agent discovery turns content operations, documentation, schema, product metadata, and reputation into one source system.
Machine Relations is becoming operational, not theoretical
Machine Relations becomes more measurable when agent discovery has explicit resource layers. The Machine Relations Stack already separates earned authority, entity clarity, citation architecture, distribution, and measurement. ARD and ADP give the agentic web a technical reason to care about those layers: a machine cannot select what it cannot resolve.
AuthorityTech's public work on entity SEO and knowledge graph optimization is relevant here because ARD formalizes the same pressure from the protocol side. Entity clarity is not a branding exercise when an AI client is deciding between candidate resources. It is selection infrastructure.
Machine Relations was coined by Jaxon Parrott in 2024 to describe the shift from human-mediated to machine-mediated brand discovery. ARD is one more data point that the shift is moving from content rhetoric into machine-readable systems.
What technical marketers should do next
The near-term move is to audit whether a brand's web presence is machine-selectable, not just machine-readable. That means treating source pages, docs, schema, entity profiles, product pages, and third-party proof as one architecture.
Start with four checks:
- Can a crawler identify the entity, category, product, and owner in the first 100 words?
- Are the claims written as direct, source-backed blocks a retrieval system can lift?
- Do owned pages and third-party sources point to the same entity relationships?
- Is there a clean measurement path for AI citations, AI referrals, and assistant retrieval?
This does not mean every brand needs to publish an ARD catalog tomorrow. It means the direction is now visible. Machine systems are getting more formal about discovery, verification, and selection. Brands that still publish unstructured positioning copy are building for the wrong reader.
For teams that need a first pass, the practical starting point is an AI visibility audit: measure what machines can already retrieve, where the entity chain breaks, and which source pages need to become clearer.
FAQ
What is Agentic Resource Discovery?
Agentic Resource Discovery is an open draft specification for helping AI clients find machine-usable resources such as tools, skills, agents, APIs, workflows, and MCP servers. Google announced it on June 17, 2026, and Microsoft framed it as automatic resource discovery for task-relevant AI clients (Google Developers Blog).
Why does ARD matter for AI search visibility?
ARD matters because it moves AI visibility closer to resource selection. A brand does not only need pages that rank. It needs machine-readable source architecture that lets agents understand what exists, who controls it, why it is credible, and whether it fits the task.
Is ARD the same as SEO or GEO?
No. SEO and GEO focus on ranking, retrieval, and generative answer visibility. ARD focuses on how AI clients discover resources they can use. In Machine Relations terms, ARD is closer to infrastructure for entity clarity, citation architecture, and distribution across answer surfaces.
Should every company implement ARD now?
Not yet. ARD is still a draft specification, and implementation maturity will vary by platform. The immediate action is to make owned sources clearer, more structured, and more verifiable so the brand is ready as agent discovery layers become part of normal retrieval.