Feature
Research Report
The AI Transformation 100: Concrete Ideas for Fixing How We Work
Prompt engineering has evolved from clever phrasing into a structured discipline that defines how large language models (LLMs) think, reason and act.
Today’s techniques, which range from simple zero-shot instructions to complex program-of-thought architectures, determine not just accuracy but transparency, reliability and control. Let’s take a look at the types of prompt engineering and how they can be used to mold AI reasoning.
Table of Contents
- 1. Zero-Shot, One-Shot & Few-Shot Prompting
- 2. Chain-of-Thought Prompting
- 3. Tree-of-Thought Prompting
- 4. ReAct Prompting (Reason + Act)
- 5. Self-Consistency Prompting
- 6. Contextual and Instructional Prompting
- 7. Prompt Chaining
- 8. Persona Prompting
- 9. Program-of-Thought Prompting
- Where Retrieval-Augmented Generation (RAG) Fits In
- Prompting as Programming
1. Zero-Shot, One-Shot & Few-Shot Prompting
At the simplest level, prompt engineering is about how much context a model gets. Zero-shot, one-shot and few-shot prompting are about how much context you provide to “learn the task” on the fly.
Zero-Shot Prompting
Zero-shot prompting relies entirely on pretraining: you ask for a summary or translation with no examples.
“Accuracy depends on how well the model’s internal knowledge already aligns with your intent,” explained Derek Ashmore, agentic AI enablement principal at Asperitas.
One-Shot Prompting
One-shot prompting introduces a single example — “here’s the pattern I want” — which clarifies tone and format. For instance:
The Prompt:
Rewrite this sentence in a punchier, more conversational style.
Example:
Original: Organizations must evaluate the risks associated with artificial intelligence adoption.
Rewrite: AI adoption isn’t risk-free, and companies need to know what they’re walking into.
Now rewrite this: Enterprises should consider implementing governance frameworks before deploying autonomous AI agents.
Few-Shot Prompting
Few-shot prompting adds several examples to teach nuance and consistency, especially for domain-specific tasks.
“This usually improves accuracy, especially for tasks like sentiment tagging, formatting or code generation,” said Shrinath Sanjay Thube, an IEEE member and AI practitioner.
Related Article: Prompt Engineering: Techniques, Examples and Best Practices
2. Chain-of-Thought Prompting
To make models more transparent,chain-of-thought (CoT) prompting asks them to show their reasoning step by step.
“Chain-of-thought prompting makes an AI ‘show its work,’ much like a math teacher asking students to write out each step,” Ashmore said. By revealing intermediate logic, developers can spot flaws and understand how the answer was derived.
Thube said this structure is key to trust, because these methods force the model to explain itself. “You can see each step in the reasoning, which helps build trust, especially in high-stakes tasks like compliance reviews or financial forecasting."
3. Tree-of-Thought Prompting
Tree-of-thought (ToT) extends CoT prompting by running several reasoning branches in parallel. Rather than following one line of thought, the model evaluates several possible chains, compares their outcomes and selects the most consistent or optimal result.
This approach mimics human brainstorming — testing different paths before deciding. By weighing multiple options, tree-of-thought prompting often improves factual accuracy and reduces hallucinations.
4. ReAct Prompting (Reason + Act)
For complex workflows requiring models to reason and interact with tools, Reason + Act (ReAct) prompting alternates between planning and execution. This approach prompts the model’s reasoning so it alternates between thinking and requesting actions from its environment, such as retrieving data, calling an API or running a tool.
The model reasons, proposes an action, receives the output and then reasons again — an iterative feedback loop ideal for research or troubleshooting. ReAct lets the model alternate between thinking and acting. For example, it can reason about a question, run a web search or tool call and then reflect again with new information.
“That loop makes it better at multi-step or open-ended tasks,” Thube noted.
5. Self-Consistency Prompting
While chain-of-thought prompting focuses on transparency, self-consistency prompting enhances accuracy. It generates multiple reasoning paths for the same query and selects the answer most consistent across them. Ashmore called it a “lightweight form” of ensemble reasoning that helps smooth out randomness and improve accuracy on hard problems.
Self-consistency runs multiple reasoning paths and picks the most agreed-upon answer, which Thube explained cuts down noise and increases reliability in complex decisions like diagnostics or planning.
Webinar
Jun
3
From AI Investment to CX Results: What Enterprise Leaders Need to Know
Move beyond experiments. See how top enterprises scale AI for CX results.
Register
Webinar
Jun
4
Content Strategy Leaders Live: Managing Risk, Compliance & AI in Financial Services
Learn how financial services leaders are modernizing content systems without disrupting trust, compliance or experience.
Register
Webinar
Jun
16
From Legacy to Launch-Ready: How Gainbridge Made Its Website a Marketing-Led Growth Engine
Join in to learn how a D2C annuity brand gave marketing full website ownership — without slowing down or risking compliance.
Register
Webinar
On demand
Content Strategy Leaders Live: Managing Scale, Safety & AI in Manufacturing
How manufacturing leaders update content systems to handle product complexity and scale while integrating AI safely.
Watch Now
Webinar
On demand
AI With Intent: How CX Leaders Decide Where Automation Belongs (And Where It Doesn't)
Senior CX leaders from REI & Hyatt reveal their framework for using AI to boost retention.
Watch Now
Webinar
On demand
From Content Sprawl to Competitive Advantage: A Knowledge Operations Roadmap
How automation, AI and a centralized content platform turn knowledge fragmentation into a scalable competitive advantage.
Watch Now
6. Contextual and Instructional Prompting
Not every challenge requires deep reasoning. Sometimes the goal is tone, structure or domain accuracy — cases suited to contextual or instructional prompting.
“Contextual prompting shapes tone, format or perspective within a single prompt, so the model’s response better aligns with the goal,” Ashmore said. This might mean specifying a legal tone, summarizing for executives or adapting to brand voice.
Instructional prompts guide the model’s behavior explicitly (“Act as a technical writer…”), ensuring outputs match expectations without additional examples.
7. Prompt Chaining
For longer or more complex workflows, prompt chaining breaks the process into manageable stages. Chaining breaks a complex task into sequential steps (analyze, plan, then write) so each stage builds on the last.
Each link produces structured output that feeds the next, improving coherence and traceability — especially valuable for enterprise use cases like report generation or compliance reviews.
8. Persona Prompting
Persona prompting assigns a role or identity to the model — scientist, teacher, project manager — to shape perspective and language. The technique blends creativity with control, letting developers fine-tune responses for empathy, authority or tone.
By defining intent (“You are a support agent for a cloud provider”), models can respond with domain-relevant language and style, maintaining brand alignment while remaining contextually grounded.
9. Program-of-Thought Prompting
Program-of-thought prompting has models generate executable code to solve reasoning problems. Looking ahead, Ashmore said prompt engineering is becoming increasingly programmatic.
“We’re already seeing this shift in frameworks like Claude-Flow, CrewAI and OpenAI Assistants, where prompts function more like functions or policies inside an agent workflow."
Thube said prompting is shifting from art to architecture. “We’re moving toward reusable, structured patterns — almost like function calls — notably in agent systems that need consistency and control."
Where Retrieval-Augmented Generation (RAG) Fits In
Retrieval-augmented generation (RAG) sits adjacent to prompt engineering. Where prompt engineering is about how you structure the instruction given to an AI model, RAG is about giving that model outside, often proprietary, information to use before it answers.
“RAG is used when the model needs additional information beyond what’s already in training data,” said Ashmore. It expands the model’s context by retrieving relevant documents or facts at runtime, grounding the response in current, authoritative knowledge and reducing the risk of hallucination.
RAG is ideal when responses need to be grounded in current or internal data, Thube agreed. For tasks relying on accuracy — financial analysis, product documentation, policy compliance — RAG ensures every response reflects verifiable facts.
Related Article: CTO’s Guide to Strategic AI Prompting: 20+ Prompts to Master Today
Prompting as Programming
Both analysts see convergence between software development and prompt design, with Ashmore predicting that prompting is becoming programming.
“Developers who understand both will define the next generation of agentic systems,” he said.
In this emerging paradigm, prompts are no longer static commands — they’re components in a modular reasoning architecture. Whether chaining, retrieving or reasoning through code, prompt engineers now design the cognitive workflows that power agentic AI.