Control the Output Format
Claude will shape its answer any way you ask — a table, a list, one word, a formal letter, or text inside markers you can copy out reliably. This chapter also reveals the trick mentioned in Chapter 1: starting Claude’s answer yourself.
Ask for the answer inside markers
In Chapter 2 we removed the unwanted introduction by saying “skip the preamble”. Another way: tell Claude to put the poem (the same tiny three-line “haiku” from Chapter 2) inside markers. Everything outside the markers becomes easy to ignore — and easy for a program to strip away automatically:
“Speaking for Claude” — starting the answer yourself
Remember the turn-taking rule from Chapter 1? Programmers can write the beginning of Claude’s reply, and Claude simply continues from there. It cannot repeat the opening you wrote for it — so it cannot add any introduction before it. If you start Claude’s answer with the opening marker “<haiku>”, the very first thing Claude produces is the poem itself:
Structured answers for programs
The same trick forces tidy, machine-readable answers. Ask for the haiku as a structured list of named lines (a format programmers call JSON), pre-fill the opening bracket, and Claude delivers pure structure — no chatter at all:
Combining everything
Markers for the input, markers for the output, and multiple changing parts — all in one prompt. Here a casual office message is wrapped in <email> markers, and the answer is requested in its own marked section, rewritten in a highly formal and respectful style:
Never accept a messy answer — name the format you want. Markers keep the answer clean and easy to extract, and pre-filling the first characters of Claude’s reply forces it to start exactly where you want.