A hosted base is easy to start and expensive to leave, so the trial is the only cheap moment to find out what it does badly. Four checks, none of which takes more than twenty minutes.
Export it, for real
Not read the documentation about export: run it, download the file, open it. You are looking for the page hierarchy, the attachments and readable text. A proprietary archive that only re-imports into the same product is the thing you would most like to know about in week one rather than year three.
Restrict one page and see what happens
Permissions are per space in most products and per page in a few. Neither is wrong, but the one you have determines your structure, and finding out after you have built the structure is the expensive order.
Put it on your own domain
For a customer-facing base this decides whether the content builds your search presence or the vendor's. Some plans put it behind an upgrade and some cannot do it at all; both are worth knowing before the content exists.
Look for a review date field
Not a modified date, which every product has and which tells you nothing about whether anybody read the page. If there is no place to record that a person confirmed an article, that record has to live somewhere else, and it still has to exist.
Questions people ask about hosted knowledge base
Is a hosted base worth it for an internal wiki?
Often the internal one can live in a tool you already pay for. The hosted purchase usually makes sense first for the customer-facing base, where the domain, the search presence and the anonymous access matter.
What about uptime?
It is rarely the deciding factor at this size. A base being down for an hour is an inconvenience; a base being wrong for a year is the failure that costs money.
How much should this cost?
Per-editor pricing is the norm and is usually fine, because the editor count stays small. Watch for per-reader pricing on internal bases, which punishes exactly the behaviour you want.