Agentic Resource Discovery (ARD) is not a new ranking factor or a replacement for SEO. It is a draft specification for helping AI agents discover callable resources — tools, APIs, MCP servers, skills and other agents — through machine-readable catalogs and registries. For SEO teams, the practical move is to get your resource layer clean now: indexable tool pages, structured data, accurate sitemaps, clear verification, and a maintained inventory of anything an agent could safely call.
Direct answer: what ARD means for SEO
Agentic Resource Discovery matters for SEO because it formalizes a discovery path for machine-readable resources, not just human-readable pages. If your company publishes tools, calculators, APIs, product feeds, MCP servers or agent workflows, you should prepare the same way you prepare content for crawlability: canonical landing pages, structured descriptions, ownership verification, fresh sitemaps, and a controlled resource inventory. Content-only sites do not need to panic, but they should use ARD as a signal that AI discovery is moving toward verified resources and callable actions.
Quotable summary: ARD turns “can an AI agent find and trust this tool?” into a technical SEO problem: publish a clear resource inventory, verify ownership, and keep every supporting page, feed and schema consistent.
Key entities: Agentic Resource Discovery, ARD, ai-catalog.json, AI Catalog, agent registries, MCP servers, Google Agent Registry, Microsoft, GitHub Copilot agent finder, Hugging Face Discover Tool, Cisco AGNTCY, structured data, XML sitemaps, llms.txt, resource pages.
What changed
Search Engine Journal reported that Google, Microsoft, GitHub, Hugging Face, Cisco, Databricks, GoDaddy, NVIDIA, Salesforce, ServiceNow and Snowflake published Agentic Resource Discovery, a v0.9 open draft specification for how AI agents discover and verify tools across the web. The core idea is simple: instead of pre-wiring every agent to every possible tool, organizations can publish a catalog on their own domain and registries can index those catalogs for agent discovery.
The important distinction for SEOs: ARD is aimed at callable capabilities, not ordinary blog posts. A resource could be a tool, MCP server, API, skill, agent, dataset, product feed, calculator or workflow that another system can invoke. That makes ARD closer to technical SEO, product SEO and developer relations than to a normal content refresh.
Source context
- SEJ coverage: ARD uses catalogs and registries; catalogs are published on the owner’s domain and can include trust metadata for verification.
- Draft status: The spec is early, v0.9, and Google’s Agent Registry support was described as planned rather than fully live.
- SEO implication: Treat this as preparation for AI-agent discoverability, not as a confirmed Google Search ranking system.
How ARD works in plain SEO terms
Think of ARD as a discovery stack with three layers.
In the draft model, an organization publishes an ai-catalog.json file at a well-known location on its domain. Registries crawl those catalogs, index the resources, and answer natural-language discovery requests from agents. When an agent chooses a capability, connection happens through the resource’s own protocol; ARD is the discovery and trust layer, not the execution layer.
What SEOs should prepare now
You do not need to publish speculative files for every site today. You do need to make sure any resource that could be used by an agent is already understandable, verifiable and easy to maintain.
1. Build a resource inventory
Create a single inventory of assets that go beyond static editorial content. Include:
- Free tools, calculators, scanners, checklists and templates.
- APIs, MCP servers, agent skills, datasets and data feeds.
- Product feeds, merchant feeds, documentation hubs and integration pages.
- Authentication requirements, rate limits, terms of use and support owner.
- Canonical URL, sitemap status, schema type, last reviewed date and deprecation status.
If you already use SGOinsights resources such as the GEO Readiness Scanner, the AI Search Visibility Checklist, or the AI Search Measurement Sheet, treat them as the model: each resource needs a clear public page, a unique purpose, and enough context for a machine or human evaluator to know when it should be used.
2. Tighten resource pages before catalog files
ARD may make catalogs more important later, but the canonical page still does the reputational work. A weak resource page creates weak machine descriptions.
- Use a stable URL. Avoid campaign-only URLs for core tools and APIs.
- Lead with the use case. “Audit AI search visibility gaps” is better than “our innovative platform.”
- State inputs and outputs. Agents need to know what data the tool accepts and what it returns.
- Document limits. Include rate limits, login requirements, supported markets, freshness and known exclusions.
- Add proof. Show examples, screenshots, methodology notes, changelog links and ownership details.
3. Make structured data boringly consistent
For resource pages, structured data should reinforce the same facts shown on the page. Depending on the asset, use SoftwareApplication, WebApplication, Product, Dataset, FAQPage, HowTo, Organization and BreadcrumbList where they accurately apply.
The mistake to avoid is schema that describes a tool differently from the page, feed or API documentation. Agent discovery systems are likely to penalize inconsistency indirectly because it creates trust and verification problems.
4. Clean up sitemaps, feeds and discovery files
ARD does not replace XML sitemaps, RSS, product feeds, API docs or llms.txt. It adds another possible discovery surface. Your preparation checklist should include:
- Resource pages return HTTP 200 and are not blocked by robots.txt or noindex.
- XML sitemaps include the canonical resource URLs.
- Feeds and docs use the same resource names, descriptions and canonical URLs.
llms.txt, if used, points to durable guides and resources rather than thin marketing pages.- Deprecated tools are redirected or marked clearly, not left as orphaned endpoints.
For more context, read our analysis of the llms.txt AI search study and the broader Microsoft Web IQ shift toward AI-native grounding and publisher controls.
5. Treat verification as part of SEO governance
ARD’s domain-hosted catalog model makes publisher identity central. That should push SEO teams to coordinate with security, developer relations and product owners.
- Confirm the resource is hosted on an owned or clearly controlled domain.
- Use HTTPS everywhere and avoid mixed canonical signals across subdomains.
- Keep Organization, sameAs, contact, support and documentation links aligned.
- Define who can approve catalog entries, API descriptions and tool deprecations.
- Log version changes so registries and agents do not discover stale capabilities.
ARD readiness checklist
List every tool, API, dataset, MCP server, feed, agent skill and template with an owner and canonical URL.
Explain purpose, audience, inputs, outputs, limits, examples, pricing and documentation from the user’s point of view.
Use accurate WebApplication, SoftwareApplication, Dataset, FAQPage, HowTo, Organization and BreadcrumbList markup where relevant.
Keep canonical resource pages in XML sitemaps; align names and descriptions across RSS, product feeds and docs.
Confirm HTTPS, domain control, publisher identity, support links, changelog and trust metadata ownership.
Track crawlability, schema validation, broken links, resource usage, AI citations and referral patterns after launch.
What not to do
- Do not call ARD a ranking factor. It is a draft agent discovery spec, not a Google Search ranking announcement.
- Do not create fake resources. A catalog full of thin “AI-ready” entries without working tools creates trust risk.
- Do not publish unmanaged endpoints. If an agent can call it, someone needs to own accuracy, abuse controls and deprecation.
- Do not abandon classic SEO. Crawlable pages, internal links, schema and sitemaps are still the foundation.
FAQ: Agentic Resource Discovery and SEO
Is Agentic Resource Discovery the same as llms.txt?
No. llms.txt is generally used to point AI systems toward important text resources. ARD is intended to help agents discover callable capabilities such as tools, APIs, MCP servers and other agents through catalogs and registries.
Should every SEO site publish an ai-catalog.json file now?
No. Most content-only sites should wait until the spec and registry ecosystem mature. Sites with real tools, APIs, feeds or agent-accessible resources should prepare an inventory and technical governance plan first.
Can ARD help a SaaS or ecommerce site?
Potentially, yes. SaaS companies with APIs, integrations, agents or MCP servers and ecommerce sites with product feeds, shopping tools and support workflows are better candidates than pure editorial sites. The near-term work is resource clarity, feed hygiene and verification.
How should SEO teams measure ARD readiness?
Measure whether each resource has a crawlable canonical page, valid schema, sitemap inclusion, aligned feed/API documentation, an owner, verification details, and a monitoring plan. Then add AI visibility tracking with prompts and citations using a template such as the AI Search Measurement Sheet.
Recommended next step
Start with an audit, not a catalog file. Use the AI Search Visibility Checklist to find gaps in crawlability, answer blocks, structured snippets and AI discovery. Then run your core tools and resource pages through the GEO Readiness Scanner and prioritize fixes where a real tool, feed or workflow could be discovered by an agent.