Understanding your numbers
How ranges work
AI assistants don’t give the same answer twice. Ask the same question five times and your brand might appear in four answers. A single check can’t tell you much, so Ocita measures the spread.
What a range means
Every metric has a value and a range, such as 47% (42 to 52). The value is what was measured. The range shows where it could reasonably sit, given how much the answers vary.
- A narrow range means a steady number.
- A wide range means fewer prompts, fewer repeats, or more disagreement between answers.
A 90% range will sometimes miss. It covers variation in the sample, not mistakes in matching names or changes in how an engine works.
How it’s calculated
Each range is a 90% interval from a two-stage bootstrap, repeated 2,000 times: draw prompts at random, then draw answers within each drawn prompt, and recalculate the metric each time. The range runs from the 5th to the 95th percentile of those results.
Drawing prompts first matters. Answers to one prompt resemble each other, so treating every answer as independent would make numbers look more certain than they are. How Ocita measures has the full method and a worked example.
Smaller ranges in the app
Two small summaries use a simpler 95% Wilson interval, because they rest on only a handful of answers: the rate for one prompt on one engine, and the four-week band on This week. Neither one decides a Real gain or Real drop.
How to get narrower ranges
- Track more prompts. This helps most, because the range reflects how much results depend on which prompts you track.
- Use more repeats: 5 a week on Standard, 3 on Founding.
- Give it time. The baseline for changes pools up to four weeks.
When the collection method changes
Each answer records how it was collected: through the engine’s API, or from what its app shows. The two can name different brands for the same prompt, so an engine’s numbers never mix them. Within a week, an answer collected another way (a backup through the API when the usual source failed) is shown on Answers with its label and marked as not counted. If an engine’s method changes, its comparison starts again: you’ll see New baseline where a Real gain or Real drop would show, until there’s a week collected the new way to compare against.
On charts
Charts draw the range as a shaded band around the line, never as a bare line.