What is the measurement margin?
AI engines do not answer the same question the same way twice. They phrase differently, search differently and name brands in a different order. So visibility, share of voice, position and sentiment move a little on their own, even when nothing in your market changed. The measurement margin says how far a number would move just from the AI answering the same questions again. It works like the margin of an opinion poll: ask 1,000 people, then 1,000 others, and the two results differ a little even if nobody changed their mind. The margin says how precise the measurement is. It does not say how important a change is.How to read it
Every change comes with its margin, for example −5.3 ± 18.4 points.- When the change is larger than the margin, it is a movement.
- When the change is within the margin, it is not proven. The two numbers could come from the same answers asked twice. Palagus and your assistant still show where it sits, marked as within the margin.
- A gap to a competitor within the margin counts as level.
Why it is different each time
The margin is computed anew for every change. Three things decide its size.1. The number of answers
This is the main reason. A number resting on many answers is steadier than one resting on a few, so more answers give a smaller margin. A whole project usually has a narrow margin, one AI model a wider one, and a single prompt or a single day a wide one. With fewer than 10 answers on either side, there is no margin. Palagus then says there are too few answers. If there is nothing to compare with, for example no answers in the period before, there is no margin either.2. How many changes are judged at once
When Palagus looks at many changes together, such as every AI model, every competitor or every day of a timeline, a few of them would look unusual by pure chance. So each margin in such a group is wider than it would be alone.3. The value itself
For visibility, a value near 50 % has a wider margin than one near 0 % or 100 %, with the same number of answers.Example
Visibility of about 50 %, the same number of answers in both periods, one change on its own:
With 100 answers per period and a visibility near 5 %, the margin is about ± 6.8 points.
How it is computed
Palagus uses 95 % confidence, the same for every customer, and a test that fits each kind of number: Newcombe’s score interval for visibility, Welch’s test for position and sentiment, and the delta method for share of voice. When many changes are judged together, the Benjamini–Hochberg correction widens each margin.Where you see it
The margin comes with the changes in the agent’s answers, in reports, and in what your AI assistant reads through the Palagus MCP server. The charts and tables in the app do not show it.Frequently asked questions
Why is the margin different for every number?
Mostly because each number rests on a different number of answers. A smaller filter, a shorter time range or a single day means fewer answers and a wider margin.Is there a formula?
There is no single number to remember. Palagus computes each margin from the answers behind the change, with the tests named above.Does a change within the margin mean nothing happened?
It means the measurement cannot tell it apart from chance. Ask again with a longer time range or a wider filter: more answers give a smaller margin.Next steps
- Learn about Visibility
- Learn about Share of Voice
- Learn about Sentiment
- Learn about Average Position

