Updated prompt engineering guide covering the latest techniques for GPT-5 and Claude 4 including structured outputs, thinking mode control, and multimodal prompting.
Key Takeaways
- 1Reasoning models now handle chain-of-thought internally — focus prompts on problem definition, not reasoning instructions
- 2Structured output (JSON schema) is natively supported by most frontier models — use it for any programmatic use case
- 3Multimodal prompting works with current models — describe what to focus on rather than asking open-ended questions
- 4Few-shot examples remain effective but fewer are needed — one or two often suffice
- 5Use system prompts for role and constraints, user messages for task-specific instructions
Prompt engineering has evolved significantly with current frontier models. Here are the techniques that work in 2026.
What Changed
Modern reasoning models (OpenAI's o-series, Claude Sonnet 4.6 with extended thinking, Gemini 2.5 Pro) process prompts differently than earlier models. They can reason step-by-step internally before answering, accept multimodal inputs, and handle context windows of one million tokens or more. The old techniques need updates.
Structured Output Enforcement
Always use JSON mode or schema enforcement when you need programmatically usable output. Most frontier models now support native structured output — specify the exact JSON schema and the model fills it. This eliminates parsing errors and makes outputs directly consumable by your application.
Chain-of-Thought Is Now Built In
Motif II In-Ear Headphones
Shop the new Motif II In-Ear headphones — premium audio engineered for clarity and comfort.
With reasoning models like OpenAI o3 and Claude Sonnet 4.6 extended thinking, you no longer need to instruct the model to "think step by step" — it does that internally. Instead, focus your prompt on defining the problem clearly and specifying what a good answer looks like. Let the model handle the reasoning chain.
Multimodal Prompting
Current models accept images, audio, and in some cases video alongside text. When prompting with images, describe what you want the model to focus on rather than leaving it open-ended. "Analyze the chart on the left and compare it to the table on the right" is better than "What is in this image?"
Few-Shot Examples Still Work
For specific output formats or domain-specific tasks, providing 2–3 examples in the prompt remains effective. The difference now is that models learn the pattern faster — one example often suffices where three were needed before.
System Prompts for Consistency
Use the system prompt to define role, constraints, and output format. Keep task-specific instructions in the user message. This separation helps models maintain consistency across long conversations.
Temperature and Sampling
For factual tasks, use low temperature (0–0.3). For creative tasks, higher temperature (0.7–1.0) helps. Most production applications benefit from temperature 0 with structured outputs — determinism is more valuable than variety when you are building a product.
For Beginners
Prompt engineering is how you write instructions for AI models to get better results. In 2026, the biggest change is that models now reason internally — you do not need to tell them to "think step by step" anymore. Focus on being clear about what you want and what a good answer looks like.
For Practitioners
The shift from instruction-based reasoning to built-in reasoning changes how you should approach prompt optimization. With models like OpenAI o3 and Claude Sonnet 4.6 extended thinking, the reasoning budget is the model's to allocate. Your job is to define the success criteria clearly. For production systems, prefer structured outputs with JSON schema enforcement over freeform text — this gives you type safety and eliminates the fragility of regex-based parsing.
Motif II In-Ear Headphones
Shop the new Motif II In-Ear headphones — premium audio engineered for clarity and comfort.
Frequently Asked Questions
Quick answers about this story
Dr. Michael Lee
AI Tools Analyst & Editorial Researcher
AI tools analyst and editorial researcher at OneStep AI, covering artificial intelligence, productivity tools, and emerging technology for consumers and small businesses.
Reviewed by the OneStep AI editorial team for factual accuracy and clarity.
Our AI expertise: OneStep AI is an AI discovery and education platform. Our editorial team researches and tests the tools we cover so you get practical, evidence-based guidance — not marketing copy. This article was independently researched and fact-checked before publication.
Affiliate disclosure: Some links in this article may be affiliate links. If you click one and make a purchase, OneStep AI may earn a small commission at no extra cost to you. This never influences which tools we recommend — our reviews are based on independent research and hands-on testing.
