Rules Anthropic hides in their own docs
2026-10-02T13:52:40Z openrouter_free google/gemma-4-31b-it
The email outlines 10 essential prompting rules for Claude 4.8, emphasizing a shift from loose prompts to specific, positive, and structured instructions to avoid flat outputs and ensure the model utilizes tools like web search effectively.
The CoolDeep AI newsletter highlights a shift in prompting strategy for Claude 4.8, noting that the model has moved away from guessing user intent and now follows instructions literally. To optimize results, the author distills Anthropic's 31-page prompting guide into ten actionable rules: 1) Define the specific output format rather than just the task; 2) Set strict length constraints; 3) Use positive instructions instead of negative ones; 4) Use action-oriented verbs; 5) Explicitly command the use of web search for verification; 6) Provide writing samples to match a specific voice; 7) Use the phrase "Go beyond the basics" for creative tasks; 8) Explicitly request the model to "Think before answering" to enable reasoning; 9) Create "skills" for repetitive prompts; and 10) Be exhaustive in defining output, order, length, tone, and format. The email also includes promotional content for AI-related investment briefings and the CoolDeep AI daily alert service.