The same six components from §1 don't deserve equal weight on every task. Which ones matter most shifts with the type of work:
| Task type | Weight these components |
|---|---|
| Analysis | Rules/criteria — state explicitly what to evaluate against, and ask Claude to show its reasoning |
| Research | Background data + output format — pair with whichever feature currently handles real-time information retrieval (web search, as it's presently named) when current information is needed (see Module 1's feature-selection judgment), and ask for sourced, attributable claims |
| Drafting | Examples + tone — a strong example of voice or structure does more work than a longer instruction paragraph |
| Brainstorming | Volume, with judgment deferred — ask for breadth first; don't ask for "the best idea" up front, since premature filtering cuts off the exploration the task actually needs |
Recognizing which task type you're in — and which components to lean on because of it — is what "adapting prompting strategy to the task" means in practice.