Prompts and topic clusters
The two objects that decide what RankX AI measures. How a prompt differs from a keyword, what a cluster joins together, and how to structure both.
Two objects decide what RankX AI measures about you. A prompt is a question asked of an AI assistant. A topic cluster is the subject a prompt, a keyword and a piece of content all belong to.
Get the clusters roughly right and everything becomes readable by subject. Get the prompts right and you are measuring the questions your buyers actually ask.
A prompt is not a keyword
They look similar and they behave completely differently, and the difference is the most useful thing on this page.
| A keyword | A prompt | |
|---|---|---|
| Asked of | An AI assistant | |
| Shape | What someone types into a search box | What someone asks a person |
| Length | Short | Longer, and constrained |
| Measured by | Position, to depth 20 | Whether your brand is named in the answer |
| Charged | Per keyword, per check | Per prompt, per assistant, per run |
| Widens | Rank tracking and AI Overviews | AI visibility and AI Answer Citations |
"CRM software" is a keyword. "Best CRM for a small law firm in Manchester" is a prompt.
Putting a keyword into a prompt panel measures very little: an assistant asked a bare keyword returns a vague overview naming the four best-known brands in the category, on every assistant, every run, forever. That is not a measurement of you.
What makes a prompt measure something
It is a question a buyer would actually ask, in the words they would use.
It does not name your brand. A prompt that names you is a question about you, and the assistant will name you in the answer. Whether you come up unprompted is the entire point.
It has constraints. Segment, geography, use case, price band, integration. Constraints are what make the answer specific enough to be about a real competitive set.
It asks one thing. A compound question produces a compound answer and an ambiguous verdict.
It is one you want to win, not one you already win. A panel tuned to look good tells you nothing.
Each prompt also carries a buyer stage: researching, comparing, or ready to buy. A panel that is all one stage tells you about one moment in a purchase.
A topic cluster is the join
A cluster is a subject your business wants to be known for, and it is the single structure that keywords, prompts and content all hang from.
That is its whole value: it makes three different measurements comparable. "We rank well on this subject, the assistants never name us on it, and we have published nothing about it" is a sentence you can only construct when one structure spans all three.
Get them roughly right rather than exactly right. Clusters are editable, renaming one loses nothing, and the cost of not having them is far higher than the cost of imperfect ones.
What a good cluster looks like
A subject, not a keyword. "AI visibility tracking" is a cluster; "best ai visibility tracking tool" is a keyword inside it.
A subject you sell into. A cluster you have no offer for produces prompts you cannot win and briefs nobody should write.
Distinguishable from its siblings. Two clusters that would take the same keywords and the same prompts are one cluster with two names, and having both makes every by-cluster reading useless.
Named the way you would say it aloud. The generated set skews to industry vocabulary, because it is generated from industry pages.
Structuring the two together
A workable shape for a new Website:
- Six to ten clusters covering what you actually sell. Fewer than six is usually one cluster doing several jobs; more than ten is usually a taxonomy rather than a strategy.
- Two to four prompts per cluster, spread across buyer stages, weighted to the stage where you should be winning.
- Keywords attached to clusters at the point you save them, not later. An unclustered keyword is one no brief, prompt or reading will ever find.
- Prune after the first re-check. Every active prompt is a recurring cost multiplied by assistants and by cadence, and this is the cheapest moment to cut.
Why cluster structure is not automated
There is no tool that creates a cluster, on any surface including MCP. Attaching a keyword to an existing cluster is a tool, because that is bookkeeping.
Deciding what the clusters are is a decision about how the business describes itself, and it is the one place where generating "something reasonable" produces lasting mess: every keyword, prompt and article filed under an invented cluster inherits the invention.
Where to go next
- AI Prompts, for managing the panel.
- Topic clusters, for the screen.
- Keyword research, for filling them.
Glossary
Where terms are defined. The general AI search vocabulary lives in the RankX AI glossary; the terms specific to this product are defined here.
Websites, workspaces and accounts
RankX AI's object model in one page. What a Website is, what a Client Workspace is, what belongs to which, and the vocabulary used throughout these docs.