Fixed size chunking is a starting point, not a plan. The strategies that hold up give each chunk a context it can be understood in, and they let a small precise chunk pull a larger passage into the prompt when it matches.
About 1 min of reading
Before this new lesson, one small question about the lesson you just finished. Recall it, do not look it up.
Small chunk, large passage.
The pair
Before you run it, say out loud what it will print. Then open the answer.
Use it
Before you run it, say out loud what it will print. Then open the answer.
Read the first example one line at a time. A line that prints shows what appears on screen; the rest set things up quietly.
A chunk that repeats the heading it belongs to can be found by a question naming that subject. Without it, the chunk says 'it' and the question has nothing to match against.
A question answered by a sentence in the middle of a section is the test for your chunking. If that question misses, the chunking is wrong, and it is the middle of a section that most splitting mistakes break.
Embed a short chunk and store a pointer to the section it came from. When the small chunk matches, put the whole section in the prompt. Finding is precise and reading has its context.
Embed each sentence and include the sentences around it in what is sent to the model. That keeps the pronoun that named the subject, without making the whole section one blurry vector.
A table, a list of options, or a set of steps is one item. Chunking it per row makes a chunk that answers nothing; chunking it as one item with a heading makes it findable.
Each one is defined here on this site. Open the entry for the full definition and the other words it points at.