2 dials shape the tree and guide topical authority mapping
Breadth sets the parent branches, 4 to 10. Each parent must cover a distinct division of the seed topic, and no 2 parents may overlap in scope. Push breadth to 10 on a narrow seed and the branches start splitting hairs. 6 is the default for a reason.
Depth is 2 or 3. At 2 levels you get parents and children, and every child comes back with an empty children array. At 3, each child gets 2 to 4 grandchildren, which are the narrowest and most specific pages in the map.
Do the arithmetic before you pick. Breadth 6 at depth 2 gives you a root plus 6 parents plus roughly 24 children: about 31 nodes, readable in one screen. Breadth 8 at depth 3 gives 8 parents, around 32 children, and up to 96 grandchildren. That is over 130 nodes, and a 130-page map is a 2-year commitment nobody finishes.
The instruction the generator follows is to go wider before it goes deeper. A parent that only makes sense with a single child was the wrong parent.
Seed topic and audience do the rest. The audience field changes which divisions get treated as top-level. "Payroll compliance" for operations leads at 50 to 200 person companies produces different parents than the same seed for accountants, because one group cares about process and the other about statute.
Every node carries 4 things in the topical map
- Name. The topic in plain words, not a page title.
- Target keyword. Narrower than its parent's, and unique across the entire map. 2 nodes with the same keyword would be 2 pages competing with each other.
- Node type. Pillar for the root and parents, cluster for children, support for grandchildren. The tree colors each level accordingly.
- Scope. One line saying what the node covers and where it stops.
The scope line is the part that earns its place. Sibling collision is the failure mode of every topical map: 2 nodes that read as different topics until you brief them and find you have commissioned the same page twice. A scope line that says where a node stops is a boundary you can check a draft against.
From a 3-level run on "payroll compliance", one branch:
Wage and hour rules [pillar]
keyword: wage and hour compliance
scope: Federal and state rules governing hours worked
and what must be paid for them. Stops before tax
withholding, which is its own branch.
Overtime classification [cluster]
keyword: exempt vs non-exempt classification
scope: Tests that decide who is owed overtime. Not
how overtime is calculated.
Salary threshold tests by state [support]
keyword: state salary threshold overtime
scope: States setting a higher threshold than the
federal floor, and the current figures.
Note the second half of each scope line. That is the anti-collision device.
Exporting the generated map
Copy as Markdown gives you the whole tree as a nested bullet list, node name bolded, keyword in backticks, node type in italics, scope after a dash. Download .md writes the same thing to a file named after the root.
Nested bullets are the format most tools accept. Any mind-mapping app that imports markdown will render the indentation as branches, and the file drops into a doc or a repo without conversion. Branches are collapsible in the tool, so a large map is navigable before you export it.
Because a language model builds the tree, the page asks you to pass a human check before the run.
No search volume, and no pretending otherwise
Every keyword in the map is a targeting suggestion. There are no volumes, no difficulty scores, no CPC figures, and no traffic estimates, because there is no keyword API attached to this tool.
Any such number would be invented. A model asked for a monthly volume will supply one, formatted convincingly, derived from nothing. A map decorated with 200 fabricated volume figures is more dangerous than one with none, because you will sort by that column and build a year of work on top of it.
The honest workflow is 2 steps. Build the map here, where the value is the structure and the scope boundaries. Then export the keyword column into a tool with a real data source and validate demand before you commit to the branches. Some nodes will turn out to have no audience at all. That is a normal outcome and finding it costs you one export.
What a map is not
It is not a cluster plan. There are no internal link directions here, and no hub-and-spoke assignments. Hierarchy and link graph are related but not identical, and a node's parent is not automatically the page it should link to.
It is not an audit either. The map does not know what you have already published, so it will contain nodes you covered years ago. Diff it against your own sitemap before treating any of it as a to-do list.
And it is not a keyword list. Nodes are topics with boundaries. Several will need more than one page; a few will collapse into one.