Agentic search is the third wave of discoverability, after traditional SEO (ranking on Google) and AEO/GEO (being cited inside AI answer engines). It describes the state where AI browsing agents — operating inside ChatGPT, Claude, Gemini, or a desktop assistant — visit websites on the user's behalf, read structured manifests, and complete the user's task by calling actions the site has declared as agent-callable.
The three waves of discoverability.
- SEO. Rank well in Google's blue-link results. The visitor reads the page.
- AEO/GEO. Be quoted by AI answer engines (ChatGPT, Perplexity, Gemini, Claude). The visitor reads the AI's summary; the site is a citation, not a destination.
- Agentic search. The visitor's agent visits the site, reads a manifest, and executes an action — subscribe, search, quote-request, contact — without rendering the visual UI. The site is a callable system, not a destination.
How a site participates.
- Ship an mcp-actions.json manifest at the site root, declaring which forms and endpoints are agent-callable, what arguments they accept, and what they return. The protocol itself is the Model Context Protocol.
- Add WebMCP markup to the relevant HTML forms so an agent can discover the action mid-page without re-fetching the manifest.
- Treat agents as a real traffic source. Apply rate limits, validate inputs server-side, log agent calls distinguishably from human calls, and watch the call success rate as a metric.
Why this matters in 2026.
AI engines are increasingly the *first* interface to the web for users with information or task intent. A site that does not expose a callable layer cedes the transaction to whichever competitor does. The cost of shipping WebMCP markup is low; the cost of being invisible to a major class of traffic, three years from now, is much larger.
Frequently asked.
- What is agentic search?
- Agentic search is the pattern where AI browsing agents visit websites on a user's behalf, read structured manifests like /mcp-actions.json, and complete tasks by calling agent-callable actions on the site — subscribe, search, contact, quote — often without ever rendering the visual UI. It's the third wave of discoverability, after traditional SEO and AEO/GEO.
- How is agentic search different from a regular SEO crawler?
- A regular crawler indexes content so a human can later find and read it. An agentic browser acts on the content on behalf of a user: it doesn't just read the page, it transacts with the site. Different intent, different traffic shape, different optimisation surface (machine-callable actions instead of meta tags).
- Do I need to rebuild my site to participate in agentic search?
- No. Ship an mcp-actions.json manifest at your site root declaring which existing forms and endpoints are agent-callable, and add WebMCP markup to those forms so agents can discover them mid-page. Your existing pages and routes stay; you're adding a thin callable layer on top.
- How does Morvion approach agentic-search optimisation?
- Every Morvion site ships with an mcp-actions.json manifest and WebMCP markup on forms users typically want agents to use (contact, subscribe, quote requests, search). The studio treats agentic traffic as a measured channel alongside organic and direct, with the agent call success rate wired into the analytics stack.