Lesson Content

Most users prompt Claude the same way they Google something.

A short phrase. A quick question. They get a response, take it or leave it, and move on. This works for simple tasks. For anything complex, this approach leaves enormous quality on the table.

Claude's output is highly responsive to how a prompt is structured. The following ten patterns are not tricks or exploits – they are the natural language versions of how Claude processes requests most effectively.

Pattern 1 – Chain of Thought: "Think through this step by step." Adding this phrase prompts Claude to reason explicitly before giving a conclusion. For analytical, mathematical, or multi-step problems, this produces significantly more accurate results. Claude is less likely to skip a step or reach a wrong conclusion when asked to show its work. Use it for: complex decisions, logic problems, multi-step plans, any question where the reasoning matters as much as the answer.

Pattern 2 – Role Assignment: "Act as a [specific expert]." Claude's responses shift in expertise and approach when you assign a role. "Act as a senior UX researcher reviewing this design" produces different and often more useful feedback than simply "review this design." The role tells Claude what lens to apply. Use it for: expert-level feedback, switching communication register, matching tone to audience.

Pattern 3 – The Other Side: "Now argue the opposite position." After Claude gives an answer, ask it to argue against that answer as convincingly as it can. This surfaces assumptions, reveals weaknesses, and gives you the full picture before you decide. One of the most underused prompts in any analytical workflow. Use it for: decision analysis, risk assessment, debate preparation, testing the strength of an argument.

Pattern 4 – What Am I Missing: "What am I not considering here?" After presenting your plan or analysis, ask Claude what you might be overlooking. This prompts Claude to look for gaps, blind spots, and unconsidered angles. The answer is often the most valuable part of the conversation. Use it for: planning, strategy, writing reviews, any situation where your own perspective might be incomplete.

Pattern 5 – Uncertainty Surfacing: "Tell me what you are not sure about." Claude is trained to give confident-sounding answers even when uncertain. Adding this breaks that pattern and asks Claude to flag its own confidence level explicitly. Responses with explicit uncertainty flags are more trustworthy than ones that present everything at the same confidence level. Use it for: factual research, medical or legal questions, any response you plan to act on without independent verification.

Pattern 6 – Format Specification: "Respond as a [table / bullet list / numbered steps / one paragraph]." Telling Claude exactly what format you want eliminates the mismatch between what you need and what Claude produces by default. If you need a table, say so. If you need three sentences, say so. Use it for: any output you will directly use or share.

Pattern 7 – XML Tags for Structure. When providing Claude with long documents, background context, and instructions in the same message, use XML-style tags to separate them. Example: wrap a document in <document> tags, background in <context> tags, and your request in <instructions> tags. This prevents Claude from conflating different inputs. Use it for: complex prompts with multiple distinct inputs, any message longer than three paragraphs.

Pattern 8 – Multiple Versions: "Give me three different versions of this." Rather than iterating one response through revisions, asking for multiple versions upfront gives you a range to evaluate immediately. Specify the differentiator – three different tones, three different angles, three different lengths. Use it for: writing, email drafts, headlines, any creative task where you are uncertain of the right approach.

Pattern 9 – Self-Check: "Now review your response and identify any errors or improvements." After Claude produces a response, asking it to review its own work catches errors and gaps Claude can identify itself. Not foolproof, but catches obvious issues before you do. Use it for: any output you will publish, send, or present.

Pattern 10 – Constraints Improve Quality: "Explain this in under 100 words" / "Use only information I have provided." Constraints force clarity. Asking for a short response produces a focused one. Asking Claude to use only provided information prevents hallucination on factual questions. Use it for: summaries, explanations for non-expert audiences, factual responses where you want to control the information source.