◇ Could this help me?
Yes. The mechanism is explained correctly and it is the part most people have never been told: a document with visual support is processed as both text and images, so the cost has little to do with how many words are in it, and that single fact explains most surprise usage spikes. The four fixes follow logically from it and are ordered sensibly, with conversion first because it is the largest saving and costs nothing. General practice, with two honest qualifications: the specific multipliers are approximations rather than measurements, and the description of how a project retrieves from stored documents is an implementation detail that can change, so rely on the reasoning rather than on the numbers.
Help me stop overpaying for documents. For each file I describe below, tell me whether its meaning actually lives in the text or in the layout, charts, and images. For the text-only ones, convert them to clean markdown before we work with them and tell me what was lost. For the mixed ones, tell me exactly which pages need to stay in their original format and help me split the file so only those do. Then tell me, for the way I actually use these documents, whether they belong in a project for repeated retrieval or in a single throwaway conversation, and why. Finally, flag any habit of mine that is quietly the expensive part rather than the files themselves.