When asking an AI to count the r's in “strawberry,” which unit should the prompt name clearly?

Answer: Visible letters

Visible letters ✓ — Visible letters are the unit the user actually wants counted. An AI may internally break text into chunks, but the task is about the three written r characters in the word. Naming the unit pulls the model toward the human reading of the question.

Model text chunks — Model text chunks are not the unit being asked for here. They matter to how a language model processes text internally, but a token-like chunk can contain more than one visible character or split in unintuitive ways. Counting chunks would answer a different question from counting r letters.

Spoken syllables — Spoken syllables do not answer a written letter-counting task. Strawberry has sound units, but the r's are visual characters on the page. Mixing sound units with written letters is exactly the kind of ambiguity a good prompt should remove.