Skip to main content
AI engines don’t give the same answer twice. Ask the same prompt again and one reply names you while the next doesn’t. So Mentioned, Recommended, Understood, Praised and the Heralded Score are estimates, and each comes with a 95% range, a low and a high value that show how precise it is.

​
What the range means

The range covers where the figure would land if the engines were asked your same prompts again. It describes how much the engines’ answers vary. It says nothing about how the figure would change with different prompts. Your tracked prompts are the whole population it describes.

​
Why some ranges are wide

The range narrows as the figure rests on more evidence, meaning more prompts, more engines and more answers per cell. A brand tracking a handful of prompts, or a week with few answers, gets a wide range. Tracking more prompts is the surest way to tighten it.

​
How it is calculated

For all four figures, Heralded computes a Wilson interval at 95% confidence on the weighted mean. Because answers are weighted, the interval uses an effective sample size that accounts for the weighting instead of the raw count of answers. It treats each answer as an independent draw. Understood uses the same interval for its answer scores between zero and one. The range for the Heralded Score is the average of the Mentioned and Recommended bounds, placed on the calibrated scale. Share of voice carries no range.

​
Don’t compare ranges to spot a change

Two reports’ ranges can overlap while the figure really moved. Ranges that don’t overlap aren’t a test either. To tell whether a figure changed, use the paired test, which is built for that question.