50 Best Claude Sonnet 4.6 Prompts That Actually Work in 2026 — For Students, Writers, and Developers

Claude Sonnet 4.6 is not the same as ChatGPT with a different name. It reasons differently, refuses differently, excels at different tasks, and responds to prompting differently. The 2.5 million people who switched from ChatGPT to Claude in February and March 2026 — the QuitGPT wave triggered by the Pentagon deal — are discovering this firsthand. Many are frustrated because their ChatGPT prompts produce mediocre Claude outputs. This guide gives you 50 prompts specifically engineered for Claude Sonnet 4.6's strengths: long document analysis, pedagogical explanation, nuanced writing, analytical depth, and code with explanation. These are not generic prompts. They are designed around how Claude specifically processes instructions, formats responses, and applies its training.

How Claude Responds to Prompting Differently From ChatGPT

• Claude is more literal: if you ask for 'a short summary,' Claude will produce a genuinely short summary. ChatGPT often produces longer responses regardless of length instructions. Use this — be specific about format, length, and structure. Claude will follow structural instructions more precisely.
• Claude thinks before answering more readily: for complex analytical tasks, starting your prompt with 'Think through this step by step before responding' produces meaningfully better output from Claude than from ChatGPT. Claude's chain-of-thought is particularly strong on multi-step reasoning.
• Claude is better at intellectual humility: for research and fact-checking tasks, Claude will more accurately flag when it is uncertain or when its information may be outdated. Use this — ask Claude to explicitly mark uncertain claims. ChatGPT is more prone to stating uncertain things confidently.
• Claude handles very long inputs better: pasting a 10,000 word document followed by your question is where Claude's advantage over ChatGPT is most measurable. The full context window is used more effectively for long-document synthesis tasks.
• Claude is less likely to just agree with you: if your business plan has a fatal flaw, Claude will tell you. ChatGPT is more likely to validate and elaborate on your approach. For critical feedback and stress-testing, Claude's reduced sycophancy is a significant advantage.

25 Prompts for Students

• Concept explanation with examples: 'Explain [concept] to me as if I am 16 years old. Start with a real-world analogy that requires no prior knowledge. Then explain the technical version. Then give me three practice questions at increasing difficulty levels, with full answers.' — works for physics, economics, history, programming concepts.
• Essay structure review: 'I am writing an essay arguing [thesis]. Here is my current outline: [outline]. Identify the three biggest structural weaknesses, explain why each is a problem, and suggest a specific fix for each one. Do not rewrite the essay — only address the structure.' — Claude produces sharper structural feedback than generic 'improve my essay' prompts.
• Exam flashcard generation: 'Read this passage and generate 15 flashcard-style question-answer pairs. The questions should test understanding and application, not just recall. Format as Q: [question] A: [answer] pairs. Vary question types: definition, comparison, application, and 'why does this matter' questions.' — paste any textbook chapter.
• Weak topic identification: 'Here are my last three exam scores and the topic breakdown: [scores]. Based on this pattern, identify which sub-topics are most likely responsible for the lost marks, explain what a common misconception in each is, and give me one targeted practice problem for each.' — works best when you provide specific sub-topic score breakdowns.
• Socratic dialogue learning: 'Teach me [topic] using the Socratic method. Ask me a question to find out what I already know. Based on my answer, ask a follow-up question that guides me toward the key insight. Continue until I reach the core principle myself. Only give me direct information if I am stuck after three attempts.' — radically different from passive reading.
• Literature analysis: 'Analyse the following passage from [book] through three interpretive lenses: [lens 1, e.g. feminist], [lens 2, e.g. Marxist], [lens 3, e.g. postcolonial]. For each lens, identify one specific textual detail that supports the interpretation and explain its significance. Then identify one detail that complicates each interpretation.' — produces analysis at essay-ready depth.
• Math step-by-step with explanation: 'Solve this problem: [problem]. Show every step. For each step, write a one-sentence explanation of why you are doing that step, not just what you are doing. If there are multiple valid approaches, show the most intuitive one first and briefly mention the alternative.' — Claude's mathematical explanation quality is significantly better when you ask for the 'why' at each step.
• Research question brainstorming: 'I am writing a research paper on [broad topic]. Generate 10 specific, arguable research questions — not descriptive questions but questions that require taking a position and supporting it with evidence. Flag which 3 you consider most original and explain why.' — Claude flags originality more accurately than generic brainstorming.
• Counter-argument preparation: 'Here is my thesis: [thesis]. Give me the five strongest possible arguments against this position, including the evidence a well-informed opponent would cite. Do not soften these arguments — make them as strong as possible. Then suggest one specific way to address each one in my essay.' — Claude's reduced sycophancy makes this particularly useful.
• Source evaluation: 'Evaluate the credibility and usefulness of this source for a research paper arguing [thesis]: [source citation and relevant quote]. Consider: the author's credentials and potential bias, the publication context, when it was published and whether that affects relevance, and whether the section I quoted is representative of the broader argument or taken out of context.' — Claude evaluates intellectual credibility more precisely than generic 'is this a good source?' prompts.
• Comparative analysis: 'Compare [thing A] and [thing B] across the following dimensions: [list 4–5 dimensions]. For each dimension, identify which is stronger and why, using a specific example. Conclude with: which is overall more [relevant quality], and under what specific circumstances would the conclusion be reversed?' — the reversibility question produces the most analytical depth.
• Complex concept simplification: 'Explain [complex concept] to me in three versions: Version 1 for a 10-year-old (no jargon, maximum analogies). Version 2 for an intelligent adult with no background in the field (correct terms, minimal assumed knowledge). Version 3 for someone with two years of study in this field (full technical precision, assume intermediate knowledge).' — helps you understand which level of understanding you actually have.
• Debate preparation: 'I have to debate the position that [position]. Give me: the three strongest arguments for this position with supporting evidence for each. The three strongest arguments against this position that I should expect to face. The most effective rhetorical move for the opening statement. One concession I should make proactively to appear reasonable, and why making it strategically helps my overall case.' — Claude's analytical depth on debate structure is notably strong.
• Writing feedback (specific): 'Read this paragraph and give me feedback on four things specifically: [1] sentence variety — is the rhythm monotonous? [2] word choice — any overused words or clichés? [3] logical flow — does each sentence follow from the previous one? [4] concision — any phrases that can be tightened without losing meaning? Be specific — quote exact phrases when you identify a problem.' — vague feedback prompts produce vague feedback. Specific dimensions produce actionable notes.
• Study schedule creation: 'I have [N] days until my [exam name] covering these topics: [list topics]. My confidence levels are: [topic: high/medium/low for each]. I can study [X hours/day]. Create a day-by-day study schedule that: prioritises low-confidence topics, includes active recall sessions (not just re-reading), builds in one complete practice exam, and leaves a buffer day. Explain the reasoning behind the schedule structure.' — the reasoning explanation helps you adapt the schedule when it inevitably needs adjustment.
• Peer review simulation: 'Pretend you are an expert reviewer for a top academic journal in [field]. Review this abstract/paper excerpt with the scrutiny of a peer reviewer who is looking for reasons to reject it. Identify: [1] any logical gaps in the argument [2] missing evidence or alternative explanations not addressed [3] clarity or precision issues in key claims [4] whether the conclusion follows from the evidence presented.' — Claude is particularly good at this when you signal it should be critical, not polite.
• Concept mapping: 'Create a concept map for [topic] in text form. At the top level, list the 3–4 most fundamental ideas. Under each, list 3–4 sub-concepts that derive from or are explained by the fundamental idea. For each connection, write a one-sentence explanation of how the sub-concept relates to the parent concept. Flag any concepts that connect to multiple parent concepts.' — the hierarchical text format translates easily into a real concept map or mind map.
• Reading comprehension check: 'I am going to describe my understanding of [topic/chapter] in my own words. After I explain it, tell me: what I got right, what I got wrong or oversimplified, what I left out that is important, and what my explanation suggests about my current misconceptions. Here is my explanation: [your explanation].' — far more effective than passive re-reading.
• Application question creation: 'I have understood [concept] at a theoretical level. Generate five application questions that require me to use this concept in realistic scenarios I have not seen before. The scenarios should span at least three different contexts. After each question, give the answer in a separate section I can look at after attempting it.' — application questions separate genuine understanding from memorisation.
• Interview preparation: 'I am preparing for a [type of interview] interview for [role/programme]. Give me: the 5 most likely conceptual questions in this subject area, with model answers at the level expected of a strong candidate. For each question, also give me: one follow-up question a rigorous interviewer might ask, and what a weak candidate would say that I should avoid.' — Claude's model answers for academic and technical interviews are consistently strong.
• Time management intervention: 'I need to complete [specific deliverable] by [deadline]. Here are the component tasks: [list tasks]. Here is what I have done so far: [progress]. Identify the critical path — the sequence of tasks where delays cascade into the final deadline. Then identify which tasks I can safely cut scope on if time runs short. Finally, give me a specific 2-hour work block plan for today that makes maximum progress.' — the critical path framing produces much more useful prioritisation than generic 'help me manage my time' prompts.
• Terminology glossary: 'I am studying [subject]. Generate a glossary of the 20 most important terms I need to understand for [specific exam or topic]. For each term: define it in plain language (no jargon in the definition), give one concrete example, and write one sentence connecting it to one other term in the list.' — the connections between terms are the most valuable part of this prompt.
• Error analysis: 'I got this question wrong on an exam: [question]. My answer was: [your wrong answer]. The correct answer is: [correct answer]. Explain: why my answer was wrong (what misconception it reflects), why the correct answer is right (the underlying reasoning), what I should think about next time I see a question like this, and give me one similar practice question with the answer.' — post-exam analysis is where Claude's pedagogical clarity is most distinctive.
• Motivation and planning: 'I need to study [topic] for [amount of time] today but I am struggling to start. Give me: a 5-minute starter task that requires no energy or motivation but gets me into the material. Then a concrete 25-minute first Pomodoro plan. Then the single most important concept I should understand at the end of today's session. Keep the language direct and practical — I do not need encouragement, I need a specific plan.' — Claude's reduced filler language makes it effective for task-focused prompts when you signal you want directness.
• Past paper analysis: 'Here are [N] past exam questions from [exam name] on [topic]: [list questions]. Identify: the three concepts most frequently tested, the question formats that appear most often, any patterns in difficulty (what types of questions are worth the most marks?), and what a student who answers all of these correctly must be able to do. Then create one new question that tests the same concepts.' — this prompt is most valuable with 10+ past questions.

15 Prompts for Writers

• Voice preservation editing: 'Edit this paragraph for clarity and concision while preserving my writing voice exactly. My voice is: [3–4 specific characteristics — e.g., 'long sentences with multiple clauses', 'informal with occasional dry humor', 'avoids passive voice', 'uses specific numbers rather than vague quantities']. Mark every change you make and explain in brackets why you made it.' — the voice description is essential. Without it, Claude edits toward its own default style.
• Character consistency check: 'Here is a character profile: [profile]. Here are three scenes featuring this character: [scenes]. Identify every moment where the character's behaviour or dialogue is inconsistent with the profile. For each inconsistency, explain exactly what is inconsistent and suggest one alternative that remains true to the character without changing the scene's function.' — Claude maintains character profile consistency across long pasted text better than any other tool.
• Dialogue improvement: 'Rewrite this dialogue so it sounds more natural while preserving what each character is communicating and revealing. Problems to fix: [1] any lines that sound like exposition dumps — characters saying things they would not say aloud [2] any 'on the nose' dialogue where characters say exactly what they mean [3] any lines where both characters sound like the same person. Show the original line, your rewrite, and a brief note explaining the change.' — three-column format with explanations teaches craft, not just fixes the passage.
• Structural analysis: 'Analyse the structure of this piece of writing: [paste writing]. Identify: the inciting moment (when does the reader know what the piece is about?), the structural centre of gravity (where does the piece spend the most attention?), whether the ending earns the opening, and one structural choice that is working well and one that is not.' — Claude is unusually good at identifying structural centres of gravity in creative and analytical writing.
• Readers' questions simulation: 'Read this draft and generate the 8 questions a thoughtful, sceptical reader would have while reading it — questions that the current text does not satisfactorily answer. Include: 'why should I care?' questions, 'how do you know?' questions, and 'what about [obvious counter-example]?' questions. Be the reader who is interested but not yet convinced.' — the 'not yet convinced' framing produces more useful questions than 'confused reader' framing.
• Tone calibration: 'Rewrite this paragraph in three versions: [Version A: more authoritative and confident], [Version B: more conversational and accessible], [Version C: more measured and academic]. Keep the core information identical across all three. After each version, write one sentence identifying what was sacrificed in the tone shift.' — the sacrifice identification teaches the tradeoffs explicitly.
• Opening line generation: 'The topic of this piece is [topic]. The target reader is [reader]. The piece's argument or central insight is [insight]. Generate 10 possible opening lines. Vary the approach: try a specific scene, a provocative claim, a striking statistic, a question, a piece of dialogue, and an unexpected juxtaposition. Do not repeat approaches. After all 10, identify which three you consider strongest and why.' — Claude generates more varied opening approaches when you name the approaches explicitly.
• Plot hole identification: 'Here is the outline of my story: [outline]. Identify every logical gap or plot hole — moments where a character's behaviour does not follow from their established motivations, where the plot requires coincidences that strain credibility, or where earlier story elements are forgotten or contradicted later. For each problem, rate its severity (minor/moderate/major) and suggest one way to resolve it without restructuring the whole plot.' — severity rating helps prioritise which holes to fix.
• Show vs tell conversion: 'This passage tells us information rather than showing it: [passage]. Rewrite it to show the same information through specific sensory details, action, and dialogue. Do not include any sentence that explicitly states what the original passage told us. After your rewrite, explain which specific showing technique you used for each told piece of information.' — the technique identification is what makes this educational, not just a fix.
• Research gap identification: 'I am writing a [type of piece] on [topic] arguing [thesis]. Here are the sources I am planning to use: [sources]. Identify: what important perspectives or counter-evidence am I missing? What would a well-informed reader in this field notice is absent? What assumption in my thesis would a critic most likely challenge, and what source would they cite?' — Claude's ability to identify missing perspectives is distinctive and useful for research writing.
• Word choice precision: 'In this paragraph, I have used the word '[word]' three times. Suggest five alternative words or phrases that preserve the exact meaning in each specific context. For each suggestion, note if it changes the connotation (not just the denotation) in any way — more formal, more negative, more specific, etc.' — precision about denotation vs connotation produces more useful suggestions.
• Argument stress testing: 'Here is my central argument: [argument]. Play devil's advocate as aggressively as possible. Find the weakest link in the chain of reasoning. Find the assumption that, if wrong, invalidates the entire argument. Find the strongest counter-example. Then: given these weaknesses, is the argument salvageable? What single change would most strengthen it?' — Claude takes the devil's advocate role more seriously when you signal you want an aggressive challenge.
• Ending evaluation: 'Here is my piece: [piece]. Evaluate the ending on three criteria: [1] Does it resolve or complicate the piece's central tension — and is that the right choice for this piece? [2] Does the last sentence earn the weight of being the last sentence? [3] Does the reader leave with an emotion or idea that matches what you intended? Then suggest one alternative ending approach that would serve the piece differently.' — the alternative approach is valuable even if you do not use it — it reveals what your ending is and is not doing.
• Clarity audit: 'Read this piece and identify every sentence that requires the reader to do extra work to understand it. For each, explain exactly what makes it unclear (ambiguous pronoun reference, complex nested clause, jargon without definition, implicit assumption not stated) and suggest a simpler alternative. Also flag any sentence that is clear but longer than it needs to be.' — distinguishing clarity problems from length problems produces more precise feedback.
• Pitch refinement: 'Here is my pitch for [project]: [pitch]. Evaluate it against these criteria: Does the hook communicate the core value in the first sentence? Is the target reader unambiguous? Is the differentiation from similar existing [projects/books/products] clear? Is there one sentence too many? Rewrite the pitch to be 20% shorter without losing any essential information, and explain every cut you made.' — the 20% shorter constraint forces genuine prioritisation rather than superficial edits.

10 Prompts for Developers

• Code explanation with reasoning: 'Explain what this code does: [code]. Structure your explanation as: [1] a one-sentence plain-language summary of the function's purpose [2] a step-by-step walkthrough of the logic, explaining WHY each step is done (not just what it does) [3] any edge cases or inputs that would cause unexpected behaviour [4] one thing that could be improved and why.' — the 'why' at each step is what separates useful explanation from description.
• Bug diagnosis: 'This code is supposed to [intended behaviour] but instead it [actual behaviour]. Here is the code: [code]. Do not fix it yet. First, identify all possible causes for this discrepancy in order of likelihood. For each cause, describe exactly what test or check I could do to confirm whether it is the actual cause. Only after I confirm the cause should you suggest a fix.' — the diagnosis-before-fix structure builds debugging skills rather than just solving the immediate problem.
• Security review: 'Review this code specifically for security vulnerabilities: [code]. For each vulnerability: name the vulnerability type (SQL injection, IDOR, improper input validation, etc.), point to the specific line where it exists, explain exactly how an attacker could exploit it, and provide the corrected code. Do not mention functionality issues — only security concerns.' — the specific vulnerability type naming makes the output actionable and educational.
• Code review simulation: 'Perform a code review on this pull request as a senior engineer at a company that cares about code quality, readability, and long-term maintainability. Comment on: naming and readability, logic and potential bugs, performance concerns for edge cases, test coverage gaps (even if tests are not shown), and anything that will make this code hard to maintain in 6 months. Be direct — this is an internal review, not a polite external critique.' — the 'direct internal review' framing is important. Without it, Claude code reviews are often too gentle.
• Architecture consultation: 'I need to build [system description]. Expected scale: [N users, N requests/day]. Budget constraints: [if any]. Give me: the simplest architecture that meets these requirements (not the most sophisticated), the two or three components that are most likely to become bottlenecks as scale increases, any technology choices that seem attractive but create hidden complexity I should avoid, and the one architectural decision I need to make correctly now because it is hard to change later.' — the 'simplest that meets requirements' constraint is critical. Without it, Claude over-engineers.
• Test case generation: 'Here is a function: [function]. Generate a comprehensive test suite. Include: happy path tests (expected inputs, expected outputs), boundary condition tests (empty input, maximum size input, exactly-at-boundary values), error condition tests (invalid types, null/undefined, out-of-range values), and at least one test for a subtle edge case that a developer might miss. For each test, write a one-line comment explaining what behaviour it is specifically testing.' — the comment requirement makes the test suite self-documenting.
• Refactoring guidance: 'This code works correctly but is difficult to maintain: [code]. Do not rewrite it yet. First: identify the three most significant maintainability problems and explain why each one will cause pain as the codebase grows. Then suggest a refactoring order — which problem to address first and why. Only after I confirm the approach should you produce the refactored code.' — the 'do not rewrite yet' instruction produces much better refactoring outcomes by establishing agreement on the approach first.
• API design review: 'Here is my API design: [design]. Review it for: [1] RESTful principle adherence — any routes that are not following REST conventions and why it matters [2] naming consistency — any endpoint names, parameter names, or response field names that are inconsistent with each other [3] error response design — are error responses informative without revealing security information? [4] one endpoint that will be confusing to API consumers and how to clarify it.' — API review prompts need specific dimensions or Claude produces vague general feedback.
• Documentation generation: 'Generate documentation for this function: [function]. Include: a one-paragraph plain-language description of what the function does and when to use it, a parameters table (name, type, required/optional, description), a return value description, three code examples ranging from simple to complex, and a 'common mistakes' section listing two things users frequently get wrong.' — the 'common mistakes' section is the most valuable part and gets omitted in generic documentation prompts.
• Learning path for a new technology: 'I am a [current skill level] developer wanting to learn [technology]. I already know: [existing skills]. Create a learning path that: starts from where I am (not from absolute zero), has clear milestones I can verify with a small project at each stage, lists the best free resource for each milestone (not just 'read the docs'), and estimates realistic time for each stage. Do not recommend paid courses unless there is no comparable free alternative.' — the 'starts from where I am' instruction, combined with providing your existing skills, produces a much more targeted and efficient learning path than generic 'how to learn X' guides.