The path from “user has a need” to “AI agent recommends your client” has five steps.

Most SEO operations are only addressing two of them.

Understanding all five is not exotic future-proofing. It’s a description of what the search infrastructure looks like right now, in mid-2026, for a meaningful and growing share of research journeys. The agencies that map this path for their clients — and build the content and technical work that makes each step successful — are the agencies that own the high-value, high-retention part of the market for the next several years.

Here’s the path.

Step one: Crawlability and indexing

Before an AI system can cite a page, it has to be able to read it. This is the part most SEO agencies already handle — clean site architecture, proper crawl directives, no blocking of important content behind JavaScript that AI crawlers can’t execute. It also includes the emerging standard of llms.txt: a file that tells AI systems what a site contains, how to use it, and which content is authoritative versus supplementary. llms.txt is table stakes, not a strategy — but it’s table stakes that most sites don’t yet have.

Step two: Entity recognition

AI systems navigate the web through entity relationships. A business is an entity with attributes: name, category, location, service area, authoritative sources, associated people, track record. The agent querying for “content strategy agency Dallas” isn’t running a keyword match. It’s looking for an entity that satisfies those attributes, with enough documented signals to trust it as a credible result.

Semantic clarity — consistent naming, structured schema, clear category signals — is what makes an entity recognizable. A business with inconsistent NAP across citations, no Organization or LocalBusiness schema, and author entities that are never named is hard to trust. Clean schema, consistent entity references, and documented expertise signals make an entity one an agent can reliably surface.

Step three: Credibility evaluation

Once an entity is found, the agent evaluates it. The signals it weighs are similar to traditional E-E-A-T but applied to a different decision: not “should this rank?” but “should I cite this as a trustworthy recommendation?” Generic coverage of a topic is useful for rankings. It’s not citable as a trusted recommendation. Specific, attributed, original content — with a named author, a documented track record, and a position that can be verified — is what earns the citation.

This is the step most content fails at. Not because the content is bad — because it’s interchangeable. An agent evaluating five agencies in a category and finding that four of them have indistinguishable content is not going to recommend all four. It’s going to recommend the one that said something specific enough to be remembered.

Step four: Extraction

Even credible content can be hard to use if it isn’t structured for extraction. An AI agent pulling a recommendation needs to extract a specific claim accurately. FAQPage schema makes Q&A content explicitly extractable. Article schema establishes authorship and publication date. The content that’s easy to extract from is the content that shows up in synthesized answers. This is where schema markup stops being a technical nicety and starts being a citation prerequisite.

Step five: Action

In agentic commerce — where the agent is completing a transaction rather than just providing information — the final step is taking an action: booking an appointment, requesting a quote, adding a product to a cart. Businesses with clean, accessible, structured entry points for AI-mediated actions will be more likely to be actioned by agents. Businesses whose entry points are buried in JavaScript or require human navigation will be skipped.

What this means for your SEO practice

The agencies that can map this path for clients — and explain which steps the client is currently failing at — are the agencies doing strategy, not just tactics. Steps one and two are standard SEO. Steps three through five are where content strategy, schema implementation, and editorial standards do the work.

The SEO Agency in the AI Era is the strategic frame. The path is complete. The question is how much of it you’re currently covering.

---

Jacob Clifton is the principal of Clifton Creative Agency. 25 years of professional writing, editing, and content strategy. Helped Television Without Pity reach one million readers a week. Built Gawker’s Morning After and Tribune’s Screener to one million monthly readers. He maps content infrastructure for a living and is unreasonably interested in the five-step path nobody else is explaining clearly.

The technical layer that connects steps one and two to the rest of the path is MCP. What MCP means for the SEO agency model is the next piece of this argument.

---