Skip to main content
A number tool answers in figures, so you never page through thousands of reviews or do the sums yourself. Each is a GET under /companies/{id}/numbers/, for the target or a confirmed rival, and most compare the company with each of its confirmed rivals. A report’s exhibit saves one tool’s question, and the report asks it again each time it’s read.

The answer

Every tool answers in one shape:

The tools

Each route’s parameters are in the API reference.

As a file

Every tool takes format=csv: its rows under its columns, a number a cell. With part=figures as well, its named figures come as the file instead, a row each, with the label, value, unit, company, how many it rests on and the interval. A file holds counts and shares, never a review’s words, and is named for the company, the tool and the day.
part without format=csv is a 400: the JSON holds both. An MCP tool answers JSON only.

Read a figure as it’s meant

  1. A window is the reviews’ own days. A pull of the newest 250 can be three months or three years, so basis.sample gives the first and last day read, and a window that starts before them says the days before aren’t covered.
  2. Only through confirmed links. A review counts only through a link a person confirmed, and a rival only once a person confirmed it. Everything else is counted in basis.left_out.
  3. Under 30 is small. A figure from fewer than 30 reviews is small, comes with its interval, and the caveats say a gap means little.
  4. Missing is never zero. A figure the source withholds stays null; a month with no reviews read is null, not 0.
  5. Cite it, don’t copy it. In a report, put the figure in braces from its exhibit, and the report fills it.