Primary Retrieval Role

What primary retrieval role means

Every page has a structural role in the retrieval ecosystem: the job it does for the reader. ContentGrapher classifies this role before running the concept analysis, because the required concept structure depends on what the page is for.

A page trying to do two jobs at once produces a diluted signal. An AI system retrieving "how to do X" will prefer a focused guide over a page that mixes explanation, step-by-step instructions, and a product pitch.

The five roles

Explain — defines or explains what something is. The reader is building understanding. Required concepts are definitions, relationships, mechanisms, and examples. Example: "What is transformer architecture?"

Guide — procedural how-to execution. The reader is completing a task. Required concepts are steps, prerequisites, decision points, and failure modes. Example: "How to fine-tune a language model."

Compare — X vs. Y, alternatives, trade-offs. The reader is evaluating options. Required concepts are evaluation criteria, trade-offs, use-case fit, and when each option applies. Example: "RAG vs. fine-tuning: when to use each."

Evaluate — is X right for me, best X for Y. The reader is assessing fit for their situation. Required concepts are fit criteria, constraints, decision signals, and alternatives for non-fits. Example: "Is vector search right for my use case?"

Convert — transactional, drives a next step. The reader is taking action. Required concepts are the value proposition, trust signals, objection handling, and a clear action path.

How the roles connect

The five roles map a reader's progression through a topic: from understanding what something is (Explain), to doing it (Guide), to comparing options (Compare), to assessing fit (Evaluate), to taking action (Convert). This ordering is why scope decisions matter across pages — a concept that belongs to a later stage in this progression often becomes adjacent on an earlier-stage page, because it requires a different kind of explanation than the page's primary role allows.

A well-structured content set routes the reader through these stages via explicit links between focused pages, rather than trying to serve multiple stages on one page.

How to diagnose a mixed-role page

The clearest signal of a mixed-role page is in the boundary classifications: look for adjacent or excluded concepts that belong to a different structural role. If your explain-page has several adjacent concepts that are procedural steps (guide-role content), those are scope problems, not coverage gaps.

A low coverage score alongside an accurate PRR classification usually indicates content gaps within that role, not role mixing. Role mixing and content gaps are different problems. Role mixing shows up in the boundary classifications. Content gaps show up in the writing brief.

What to do with a mixed-role page

Separate the roles into dedicated pages and link between them. A user who understands the concept (explain page) should be able to follow a link to the guide. A user who completes the guide should be able to link to the evaluation page.

If separation is not practical, decide which role is primary and demote the secondary role to a brief mention with a link to the dedicated page. Do not try to serve both roles equally.

How confident is the classification

ContentGrapher produces a confidence score for each PRR classification, visible in the analysis header. A high-confidence classification means the page's concept structure clearly pointed to one role. A low-confidence classification means the content is less clearly structured around a single role — often a sign that role mixing is already present.

If confidence is low, inspect your concept list for adjacent or excluded concepts that belong to a different structural role. That is usually where the ambiguity is coming from.

If the classification seems wrong

First, check the primary topic (anchor) that ContentGrapher detected — an incorrect anchor produces an incorrect PRR. The anchor is editable in the analysis header. Correcting it and re-analyzing will produce a new classification.

If the anchor is correct but the PRR still seems wrong, the classification reflects the structural role ContentGrapher detected from the actual content mix, not the intended purpose. The concept distribution, not the title or intent, drives the classification.

To change the PRR, change the content structure: reduce or remove content that belongs to a different structural role, then re-analyze. See Boundary Classifications to understand which concepts are creating the role conflict.