Every AI visibility number needs a range
Ask ChatGPT the same question twice and you can get two different shortlists. Here is why a share without a range misleads, and how to read one that has it.
Ask an AI engine which CRM a ten-person sales team should buy, then ask again. The shortlist often changes: a brand drops out, another moves up, a third appears. That is not a bug. These engines sample their words, search the web live, and are updated without notice.
So when a tool says your brand is named in 34% of answers, the honest follow-up is: out of how many, and how sure are you? A share from 12 answers and a share from 300 answers are not the same number, even when they print the same.
What a range tells you
A range is the band your true share most likely sits in, given the answers behind it. Sway prints it beside every number, like 34% ±3. Fewer answers make a wider band. More answers make it tighter.
- If last month was 27% ±3 and this month is 34% ±3, the bands don’t overlap: something real moved.
- If last month was 19% ±4 and this month is 21% ±4, the bands overlap: you can’t tell a change from noise yet.
- If a tool shows no range at all, you can’t tell either way.
Check it yourself
A range is only useful if you can trust the answers under it. In Sway, click any number and you read every answer it came from: the engine, the date, the country, and where your brand appeared. If a number looks wrong, you can see why.
A number you can’t open is a number you have to take on faith.
What to ask any vendor
- How many answers is this share built from?
- How often is each prompt asked, and on which engines?
- Can I read the answers behind a number?
- When you say it went up, how do you know it wasn’t noise?
Our full method, including how we set the range and when we call a gain real, is on How we measure.
See what AI says about you
Sway asks your buyers’ questions on every engine, drafts the fixes, and shows which ones worked. 7 days free, no card.
Start free