I can't really see a principled distinction here; both devices are getting information from online search engines and presenting it to the user.
A voice assistant can be reasonably expected to be operated by a very young child, who would be responsive and trusting to verbal commands in an adult-like voice from it.
A keyboard computer with a search engine in a browser cannot reasonably be expected to be operated by a very young child, and would not be reasonably expected to issue spoken-verbal commands in response to queries.
The information content is a relevant factor in the former case but not the latter; keyboard use of search engines is not in-scope for evaluating the child safety risk of voice assistants, though it is certainly an interesting topic as well.
They are separate concerns and considering the browser case does not nullify the necessity of considering the voice assistant case.
For example, replace “10-year-old” with “3-year-old” (a simple and effective safety test for potential threat to children) and apply that test to both devices:
A) Could a 3-year-old child be reasonably expected to operate a keyboard-browser search and result interface? Could they be placed into danger by doing so?
B) Could a 3-year-old child be reasonably expected operate a voice assistant search and result interface? Could they be placed into danger by doing so?
The answers to A and B are not the same, because the threat models (to the child) are very different when “can read and type” is not taken for granted, as is implied for a 10-year-old. This is why the US has a safety commission to consider these things: common sense can hide critical threats to children that are invisible and unrecognized by well-meaning adults.
Oh, wait.
“In what ways do they restrict the available voice commands and results to protect children?”
And I bet it will be found that their interface allows never-reviewed-by-Amazon content to be read out loud to children, because Amazon loves to externalize human costs.