Techniques
Techniques
Browse prompting techniques by category.
Fundamentals
Zero-shot promptingAsk the model to complete a task directly, with no examples at all.One-shot promptingShow the model exactly one worked example before asking it to repeat the pattern.Few-shot promptingShow the model a handful of examples (typically 2-5) before the real task.Contextual promptingGive the model relevant background before the task, instead of asking directly.Contrastive / Negative promptingShow a good example and a bad example side by side, so the model learns the boundary between them.KERNELA simple 5-field starter framework — Task, Input, Constraints, Output, Verify — that builds a solid prompt in one guided pass.
Reasoning
Chain of thoughtAsk the model to reason step by step before giving its final answer.Tree of thoughtsExplore several different reasoning paths in parallel, evaluate them, then develop the best one.ReAct (reason + act)Interleave reasoning steps with actions (tool calls) and their observed results.Step-back promptingAsk a broader, more general question first, then use the answer to solve the specific one.Self-ConsistencyGenerate several independent reasoning paths for the same problem and take the most common answer.Least-to-Most PromptingBreak a complex problem into an ordered sequence of simpler sub-problems, solving each one before moving to the next.Self-Critique / RefineGenerate a first answer, critique it against explicit criteria, then produce an improved version.Generate-and-RankGenerate several candidate answers, score them, and return the best one — a simpler alternative to Tree of Thoughts.Tool callingDefine which tools the model can use, when to use them, and how to handle failures — distinct from ReAct's reasoning loop.Skeleton-of-ThoughtGenerate a skeleton (outline) of the answer first, then expand each point — better organization, especially for long-form output.Analogical PromptingAsk the model to recall similar problems it has solved before, then apply the same approach to the new one.Debate / Multi-agent promptingSimulate two or more perspectives arguing the problem, then reach a conclusion based on the strongest arguments.Socratic promptingThe model asks clarifying questions before attempting to solve the problem, instead of answering immediately.Plan-and-ExecuteSeparate planning the full approach from executing it and verifying the result — useful for tasks with tools or multiple steps.Assumption AuditList the key assumptions behind your reasoning before answering, noting how sensitive the answer is to each one.Reframe TestRestate the problem from a different angle before solving it, to check whether the original framing was the right one.Failure Mode AnalysisSystematically identify how a plan could fail — on your own work, or applied to analyze a competitor's.Adversarial ReviewHave the model critique the answer as a skeptical third party looking for flaws, not as the author defending it.Chain-of-Verification (CoVe)Draft an answer, generate verification questions about it, answer those questions, then revise the answer.Confidence-Weighted PromptingAsk the model to state how confident it is in its answer, and what would change its mind.Generated Knowledge PromptingAsk the model to generate relevant background knowledge first, then use it to answer — instead of answering directly from an unstated assumption of what it knows.Program-Aided Language Models (PAL)Have the model write code to compute the answer instead of calculating it directly in its reasoning — reduces arithmetic and logic errors.ReflexionAttempt the task, check it against success criteria, reflect on failures, and retry — repeating until it succeeds.Automatic Prompt Engineer (APE)Generate multiple candidate phrasings of a prompt, evaluate them, and pick or combine the best.
Role & persona
Structure & output
Meta promptingAsk the model to design or improve the prompt strategy itself before tackling the task.Output formattingSpecify the exact shape the response must take — JSON, a table, a fixed template — not just its content.Constraint promptingState the hard rules the output must satisfy explicitly, as an itemized list.Grounded / Context-only promptingInstruct the model to answer strictly from the material you provide, and say so when it can't.Directional Stimulus PromptingGuide the direction of the answer with a few cue words, without rewriting the full instructions.Markdown/Symbol StructuringUse a consistent set of Markdown symbols — brackets for variables, dashes for sections, bold for constraints — to make a prompt easier to parse.XML/Tag StructuringWrap distinct parts of a prompt in XML-style tags — especially effective with Claude.