Gemini answers "Empty response" for every model, but the key test says the key works. What is going on?

When a Gemini model returns no answer at all, without any reason, Google has refused the request: its filter blocked what was sent (the topic, the notes or the stories in the instructions), and every Gemini model refuses the same request, which is why rotating to another model fails too and why a short key test passes. From version 1.9.19 BookyAI says this in plain words, with Google's reason, and stops trying other models at once. Your key and the model are fine in that case; what has to change is the wording of the topic, of the Codex notes or of the stories, or you use another engine for that book (Settings, AI engines). BookyAI also names the other cases: an answer cut by the safety filter with the categories flagged, a thinking model that spent its whole budget before writing, or a model that answered nothing at all.