Prompt engineering is the process of crafting and refining inputs—known as prompts—to effectively guide large language models (LLMs) such as GPT-4, Copilot, and Gemini in generating accurate and relevant responses. It acts as a bridge between human intent and machine understanding, ensuring that AI systems deliver meaningful and useful outputs. By designing prompts with clear instructions, contextual information, or examples, users can influence the model’s behavior across a wide range of tasks without altering the model’s underlying parameters.

Prompts can take various forms, including questions, instructions, documents (such as PDFs, Excel spreadsheets, and Word files), and other types of files.

Effective prompt engineering follows key principles to enhance AI performance. Providing detailed and specific prompts helps minimize ambiguity, while incorporating examples guide the model toward the desired output. Breaking down complex queries into step-by-step instructions improves clarity and accuracy. Specifying structured formats—such as bullet points or numbered lists—encourages well-organized responses. Continuous testing and refinement of prompts further optimize AI interactions, making the system more effective for specialized applications.

Interactions with prompts guide a pre-trained LLM during the session by influencing its generated responses, but they do not retrain or permanently change the model’s underlying characteristics. 

Prompting only temporarily influences the outputs within the session — it shapes how the model behaves while you interact with it, but once the session ends, the model "forgets" unless you design memory into it.  The model’s core knowledge stays the same. Prompting is like steering a car, not rebuilding its engine.  

In the context of pre-trained models, a "session" refers to the period of interaction between the user and the model, beginning when a conversation starts and ending when it is closed or reset. The concept of a session is crucial because the comparison of the following prompting techniques is only valid when each technique is applied in a separate, individual session. If you wish to use multiple prompting techniques within the same session, you must carefully design the prompts to achieve the desired outcomes. 

If you tune or fine-tune the model itself, you create a retrained version of the LLM. However, in this section, we focus on prompting techniques to customize responses and achieve better results using pre-trained models.

To obtain expected responses from AI, prompts must be carefully designed and appropriately adapted to the situation. Prompts can influence the model positively in some cases and negatively in others. Adding more prompts does not necessarily lead to better outcomes. This is why a variety of prompting techniques are employed to guide the model toward producing more accurate and relevant answers.