Prompting Resources
A curated directory of official docs, research, and guides. See our full research process →
New here? Start with these
Learn Prompting — Introduction
Why this mattersThe most approachable starting point if you've never written a structured prompt before — no assumed background.
Google Workspace with Gemini Prompt Guide⬇ Download
Why this mattersOrganized by job role rather than by technique — useful if you want to see prompts for your actual day job, not abstract examples.
Prompt engineering overview
Why this mattersStates the three preconditions for prompt engineering (success criteria, a way to test against them, a first draft) before anything else — worth reading even if you don't use Claude.
Prompt Engineering Guide
Why this mattersThe most comprehensive single reference for prompting techniques across models — good for looking something up, less good for a first read start-to-finish.
Official docs
Best practices for prompt engineering (blog)
Why this mattersNon-technical framing; good link for general-audience readers.
DeepSeek API documentation
Why this mattersModel IDs churn fast; V4 generation released April 2026.
Define success criteria and build evaluations
Why this mattersSource for the 'success criteria' scoring dimension.
Gemini Enterprise Agent Platform - prompting strategies
Why this mattersBest published prompt-debugging checklist.
GPT-5.2 prompting guide⬇ Download
Why this mattersCurrent flagship guidance; model remains prompt-sensitive and steerable.
Llama 4 model cards and prompt formats
Why this mattersChat template tokens matter for raw endpoints.
Model guidance / prompt guidance
Why this mattersHub page. Markdown available by appending .md; full index at llms.txt.
Prompt design strategies
Why this mattersSource of the instruction-placement rule (put your question after long context, not before) — the single most-overlooked piece of model-specific guidance we found.
Prompt engineering guide
Why this mattersDistinctive advice: iterate fast rather than perfect the prompt.
Prompt engineering how-to guide
Why this mattersMOVED: docs relocated from llama.com/docs to developer.meta.com/ai/docs.
Prompt Engineering whitepaper (Boonstra, Sept 2024)
Why this mattersLocal copy already in project knowledge.
Prompt guidance
Why this mattersDocuments how prompting differs by GPT-5.x reasoning-effort setting — the one thing most other guides miss about current OpenAI models.
Prompting best practices (living reference)
Why this mattersModel-specific guidance first, then cross-model techniques, then migration notes.
What's new in Gemini 3.5 Flash
Why this mattersUpdated 2026-07-30. Source of the instructions-after-context rule and verbosity note.
Research
A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT
Why this mattersPattern taxonomy; useful for template library naming.
A Survey of Context Engineering for Large Language Models⬇ Download
Why this mattersFormal definition of context engineering as a superset of prompt engineering.
Chain-of-Thought Prompting Elicits Reasoning in LLMs⬇ Download
Why this mattersSTALE-BY-CONTEXT: predates reasoning models. Pair with Wharton Report 2.
Context Engineering: From Prompts to Corporate Multi-Agent Architecture⬇ Download
Why this mattersFive context quality criteria: relevance, sufficiency, isolation, economy, provenance. Scoring rubric source.
GEPA: Reflective Prompt Evolution
Why this mattersScore-plus-explanation feedback metric. Beats MIPROv2 by >10%, GRPO by up to 20% with 35x fewer rollouts.
Large Language Models Are Human-Level Prompt Engineers (APE)
Why this mattersEarly automatic prompt optimisation.
Prompting Science Report 1: Prompt Engineering is Complicated and Contingent⬇ Download
Why this mattersPoliteness had minimal aggregate effect; removing formatting instructions caused drops. SSRN 5165270.
Prompting Science Report 2: The Decreasing Value of Chain of Thought⬇ Download
Why this mattersReasoning models gain marginally; 20-80% time cost. SSRN 5285532.
Prompting Science Report 4: Expert Personas Don't Improve Factual Accuracy⬇ Download
Why this mattersAudits vendor persona advice directly.
Prompting Science Reports
Why this mattersThe source for most of what didn't survive testing on this site — persona prompting, chain-of-thought on reasoning models, politeness. Read this before trusting any prompting advice on social media.
Self-Consistency Improves Chain of Thought Reasoning
Why this mattersSTALE-BY-CONTEXT.
Structured Context Engineering for File-Native Agentic Systems
Why this mattersThe source for "format doesn't matter, model tier and context retrieval do" — 9,649 experiments across 11 models and 4 formats.
The Prompt Report
Why this mattersThe field's canonical technique taxonomy — 58 techniques cataloged and named. Useful as a reference, not a page-turner.
The Prompt Report - plain-language summary
Why this mattersLink this beside the paper for non-technical readers.
When "A Helpful Assistant" Is Not Really Helpful
Why this mattersThe peer-reviewed source for "assigning a persona doesn't improve accuracy" — 162 roles, 2,410 questions, tested directly.
Community
One Useful Thing
Why this mattersEthan Mollick's newsletter — a reliable second opinion on whether a new prompting claim is real or hype, from someone who runs the actual experiments.
Simon Willison's Weblog
Why this mattersThe best ongoing, practitioner-level commentary on what's actually changing in this field week to week, not just what vendors announce.
The lethal trifecta for AI agents
Why this mattersReference for the prompt-injection hygiene feature.
Tutorial
Interactive prompt engineering tutorial⬇ Download
Why this mattersBest hands-on free course.
Prompt engineering tutorial (Google Sheets)
Why this mattersLighter-weight spreadsheet version. File > Download.
Prompting capabilities cookbook
Why this mattersFour capabilities: classification, summarization, personalization, evaluation.
Tool
Best prompt management tools 2026 field guide
Why this mattersVendor-authored; useful for consolidation facts, treat claims as marketing.
Metaprompt notebook (a prompt that writes prompts)
Why this mattersClosest published cousin to Prompt Builder's own architecture.
Prompt optimization market overview 2026
Why this mattersSource for PrompTessor's 6-dimension 0-100 scoring - direct competitor benchmark.