Thanks for the explanation. I can only repeat that I don’t think this is solvable. Unlike Signal, which uses E2E this is not possible here, so whatever promises Kagi makes will always have to stay promises as they’ll need to know the content of your search and for what account it is (as your settings influence the results) to give the right reply. At least that’s how I understand it.
1st Allow users to share the token by which they are identified. 2nd Implement a 3rd party website that offers a 'free tier' via token roulette.
That will generate noise in search results, making it hard to pin a particular search on someone. Optimally users would be able to decide how much 'noise generating searches' they allow.
Off course that would only work if paying users would be volume / rate limited.
In regard to selling the data, I've seen businesses where they sell user info just as a side project to balance the sheets. It will b an issue only if one is caught. Somewhere there I looked around the website but I couldn't even find where is the HQ and under which jurisdiction they operate. The privacy policy had lots of technical details, but not legal one.
It would be interesting if like Signal they can validate an user but provenly cannot associate the account with any actions. This might be an issue for billing and tiered plans though.