Voice strength tells you how solid your derived voice is, and what would make it steadier.
It is not a grade. It is a list of what is missing.
Why ten samples
Below about ten, a derived voice is thin. It picks up the habits of whichever few posts it has rather than the pattern across your work, so drafts drift between sounding like you and sounding like nobody.
The panel counts your samples against ten. Getting there is the single biggest improvement available to your output quality.
The two gaps that matter more than the count
A channel with nothing from it. If you post on TikTok and Instagram but every sample is from Instagram, your voice is derived from half your work. How you write for a fifteen-second video is not how you write a carousel caption, and a voice built from one will write the other badly.
Only one format. Ten talking-head scripts give Vamos ten samples of one register. Add something written rather than spoken, or something short rather than long, and the voice stops over-fitting to one shape.
The panel flags both, by name.
What counts as a sample
Anything where the words are yours. A post, a profile Vamos reads posts from, a document, or text you paste in.
Your own posts are the best source, because they are you writing for your actual audience rather than you describing how you write.
What it does not measure
Whether the voice is good. It measures whether it is well-evidenced.
A voice derived from twelve samples of your weakest writing will score well and write badly. Pick the samples you would be happy to be judged on.
Raising it
Add samples until the count is met, then work through the gaps it names. Use Test it after each round to hear the difference rather than reading the report and hoping.
If the voice is wrong rather than thin, Improve in chat changes the report directly. Reset only when it was built from the wrong material entirely.