The number is the least useful part
A star rating is a single data point among many. What a model can actually use is the prose — what customers say went well, what they complain about, and the specific language they use for both.
Review text becomes description
When an engine summarises what it is like to work with you, it is largely paraphrasing reviews. That makes review content a direct input to how you are described, not merely a trust badge on a page.
Your reviews aren’t a rating. They’re a body of text describing you, written by other people.
Ask for specifics, not for stars
A prompt asking what problem you solved and how it went produces text with usable detail. “Please leave us five stars” produces a number and a blank. The former is worth several of the latter.
Consistency across platforms matters here too
Reviews scattered across platforms describing different versions of your business create the same ambiguity as inconsistent naming. Being present in a few places well beats being thin everywhere.
Negative reviews are less damaging than you think
A consistent complaint pattern gets summarised, and a single outlier usually doesn’t. What does damage you is the pattern you have not addressed — which is why reading complaints as a set is a visibility exercise as well as a service one.
Respond, because responses are text too
A reply explaining what you did about a problem is read alongside the complaint. It is one of the few places you can add your own account of an incident to a source that isn’t yours.