Core of giving effective feedback in Claude conversations: say "what's right and what's wrong," not just "wrong." Effective feedback has two elements: what to keep (so Claude knows what not to change) + what to change (specific direction for revision). This makes each conversation round a precise step forward rather than guessing from zero every time.
Why "wrong, redo" has poor results: Claude receiving vague negative feedback can only "re-guess what you want" — it doesn't know if tone is wrong, direction is wrong, or format is wrong, so it may repeat the same mistake. Each piece of unclear feedback is wasted time in an inefficient loop between you and Claude.
Why examples outperform descriptions: human language is limited in its ability to describe style — "conversational" means A to you but Claude interprets it as B. A concrete text example eliminates this ambiguity: Claude sees the example and directly learns the target linguistic characteristics, sentence length, and vocabulary habits. This is especially effective for tasks with specific brand voice or personal writing style requirements.
Practice exercise you can do now: find a previous Claude conversation where you felt output quality was insufficient, try giving explicit feedback using the "keep X, change Y" format, and see if the next output improves significantly. This exercise builds your intuition for "giving effective feedback" — the single most important skill for getting consistently high-quality results from Claude.