Chapter 7: Prompt and Intent Mapping

Chapter 4.3 established intent mapping as one of GSO's five pillars and made the case for why it matters: prompts carry context, constraints, and comparisons that keywords never held, and content aligned with a literal query but not the functional intent behind it will not be retrieved effectively. This chapter is the operational practice that pillar pointed toward. It is not keyword research with longer queries, a framing this chapter deliberately avoids at every turn. It is a full research discipline: recognizing what a prompt actually carries, classifying it, grouping it with the prompts that share its underlying need, assessing whether that need is genuinely resolved, mapping existing content against it with real rigor, and testing the whole model against what generative platforms actually do.

Why Prompt Research Replaces Keyword Research

Keyword research was built for an interface that rewarded brevity. Generative prompts removed that constraint, and user behavior changed the moment it lifted: people describe situations now, not compressed topic labels. This is not a difference of degree, a longer string with more words in it. It is a difference of kind, and treating one as a stand-in for the other quietly guarantees content that answers a topic while missing the actual request underneath it.

This chapter builds the replacement research practice in sequence. Each sub-chapter depends on the one before it: the vocabulary from keywords-versus-prompts feeds the categorization work, categorization feeds clustering, clustering feeds coverage assessment, coverage assessment feeds the content mapping, and the mapping gets checked against reality through testing. The sequence matters. Skipping a step tends to produce a weaker version of every step that follows it.

Keywords vs. Prompts

A keyword is a search-box input compressed by an interface that never asked users to type more. A prompt is an uncompressed expression of the actual situation behind a request, carrying at least five dimensions a keyword cannot hold: stated constraints, implied comparisons, risk tolerance, urgency, and sophistication level.

Treating a prompt as simply the longest possible long-tail keyword discards exactly the information that would let content serve the real request. This distinction is not academic. It changes research method at the source: gathering and reading real prompt language, not extrapolating from keyword-volume data that was never built to hold this kind of detail. Chapter 7.1 establishes the vocabulary the rest of the chapter is built on.

Prompt Categories

Chapter 4.3 introduced six prompt categories: definitional, comparative, evaluative, instructional, exploratory, and predictive. This chapter takes that taxonomy into the field, covering how each category actually sounds in ordinary phrasing rather than in textbook examples.

Real prompts frequently blend categories, most commonly comparative language masking a genuinely evaluative request. “GSO vs SEO” reads as comparative, but the person typing it is often trying to decide whether to invest in GSO at all, which a pure feature comparison never resolves. Recognizing the blend, not just the surface category, is the skill Chapter 7.2 builds, and it is a continuous practice applied to every new prompt, not a taxonomy exercise performed once.

Intent Clusters

Ten differently worded prompts are frequently one question wearing ten outfits. An intent cluster groups prompts by shared underlying need rather than by surface wording, using a direct test: could one well-built asset resolve every prompt in the group.

Clusters can form along several axes at once, topic, funnel stage, task, risk, geography, and sophistication level, and getting the boundaries right in both directions matters. Over-fragmenting produces many thin, competing pages; over-merging produces one unfocused page serving no real need well. Chapter 7.3 covers both failure directions and the ongoing discipline of revising cluster boundaries as new prompt data accumulates.

Prompt Coverage

Prompt coverage measures the degree to which a site’s content can satisfy the real questions an audience asks generative systems, assessed cluster by cluster rather than keyword by keyword. A coverage gap is a missing information need, not a missing page, and a page can mention a topic thoroughly while leaving the cluster’s actual underlying need entirely unresolved.

Partial coverage, where a cluster is touched but not fully resolved, is the most common and least visible failure state, often hiding behind content that looks complete on the surface. Chapter 7.4 covers how to assess coverage honestly, given the current absence of mature GSO measurement tooling, and how to prioritize which gaps deserve attention first.

Content-to-Prompt Mapping

This is the chapter’s working deliverable: auditing existing content against real prompts at two resolutions. Topic-level alignment asks whether a page covers the right subject. Fragment-level alignment asks whether a specific passage actually resolves the prompt directly, in a form a system extracting a fragment could use without reading the rest of the page.

A page can pass the first check completely and fail the second entirely, and that gap is where most real coverage problems hide, invisible to any audit that stops at topic-level assessment. Chapter 7.5 covers building the mapping document, auditing flagship content honestly rather than assuming it’s exempt, and turning the finished map into a prioritized action list.

Testing Prompts

Everything built through this chapter is a model of reality constructed from research and judgment, and models benefit from being checked. Testing submits real prompts across generative platforms to validate the intent clusters and coverage assessment already built, using platform rotation, wording variation within clusters, and private sessions that minimize personalization’s influence.

Testing has real, stated limits: personalization means a single result reflects one session, not a stable ranking, and manual testing can only cover a small fraction of real-world prompt variation. Chapter 7.6 covers the method, what to document, and why results should actively correct the research rather than simply confirm it.

From Mapping to Architecture

The output of this chapter, intent clusters, a coverage assessment, and a fragment-level content map, is the direct input to what comes next. Chapter 8 uses this mapping to make architecture decisions: which clusters need a dedicated spoke page, where silo boundaries should sit, and which existing assets need restructuring rather than replacement.

This is why the sequence in this chapter matters as much as any individual sub-chapter’s content. A content architecture built on top of sloppy clustering or shallow, topic-level-only mapping inherits that sloppiness at every level above it. The rigor this chapter asks for is not perfectionism for its own sake; it is the foundation the next several chapters are built directly on top of.

Building Research That Actually Reflects Real Requests

Michael Rubinstein treats prompt research as the place where most GSO practices quietly fail before they ever reach content production, because the keyword-research habit is deeply grooved into practitioners with an SEO background, and the failure is invisible: the resulting content looks complete, covers real topics, and still underperforms against the prompts it was meant to serve.

ScribePress builds every content plan from this chapter’s sequence directly: real prompt language, honest classification, disciplined clustering, fragment-level mapping, and testing that corrects the model rather than just confirming it, because skipping any step in that sequence produces exactly the kind of content that looks right and isn’t.

Learn more about the work behind this framework at michael-rubinstein.com.

Put the framework to work