Kagi works by using a bunch of upstream indexes and mixing all their results together, applying your personalizations as a post-step. Because of this, if your pinned domain does not appear in the results we get back, then there is nothing for us to pin.
We likewise do not have infrastructure to understand whether or not your query is likely to want results from letterboxd (or any of your other personalizations for that matter) such that we could manipulate our API calls to coax more relevant results out. We can't just send your hundreds of personalizations with every API call we make, it doesn't work like that.
As such, the primary ways for us to fix this are
- For us to build our own large scale search engine instead of relying on others. This is a massive undertaking, but something we are working on nontheless.
- Using some kind of lighter weight indexing and ML (read: NOT LLMS) to predict relevant domains ahead of time.
Both of these are not simple and require whole teams of resources to address, which is why issues like these will take a long time to resolve in meaningful ways, unless the SEO of your favorite sites improves in our upstreams over time.
