Step by step.
- Find a split cell: one with a count like 1/3 or 2/3 on it.
- Click it and read the answers side by side. Your name is highlighted green where it appears, competitors red.
- Look for what differed. It is usually one specific fact, one specific source the engine cited, or one specific phrase in the question that sent it somewhere else.
- Open Diagnosis for that question. The plan there is built from exactly that difference: which source to fix, or which page to write.
A one-in-three mention looks like a near win and feels like one. It is recorded as a split, and the split is often the most interesting thing on the map, because it means the engine can name you and usually chooses not to.
The reason is how buyers behave. A buyer asks once. If the engine names your competitor in two of three asks, two of three buyers are sent there. So the map scores on the majority, and a 1/3 stays a split until the majority moves.
The reason it is good news is what you find when you read the three answers. The one that named you and the two that did not differ in something specific: a fact one answer had and the others did not, a source two answers cited that says the wrong thing, a phrase in the question that steered the engine. That difference is the diagnosis, and it is what the plan on the Diagnosis tab is built from.
What the screen is telling you.
- A cell is scored on the majority. Named in two of three is a win; named in one of three is a split. The threshold is the same on every engine and every month, so the trend is comparable.
- Cells that split or changed hands are asked again until they are at least five deep, so a split on the map is a real split, not a small sample.
- The count is a share, not a rank. 1/3 means one answer in three named you anywhere in its list, not that you were third.
People ask.
- So a split is bad news?
- It is the best news on the map after a win. The engine can name you. It has the sources to do it. Something in two of the three answers pulled it toward someone else, and that something is nearly always fixable: a wrong fact, a missing page, a directory that says the wrong thing.
- Why not count any mention as a win?
- Because a buyer asks once. If two of three asks send them to a competitor, that is where two of three buyers go. Counting the one-in-three as a win would make the map say you are visible when most buyers are not seeing you.
- How many asks are behind a cell?
- Up to eight per engine per month. Cells that split or changed hands are re-asked until they are at least five deep. The confidence label under the cell says how many asks agreed.