5 Prompting Fixes That Improve Output From ChatGPT And Claude - Forbes

To improve the output from large language models such as ChatGPT and Claude, five key prompting fixes are recommended. Firstly, users should structure prompts by clearly separating elements like roles, rules, data, and tasks to enhance clarity for the AI. Secondly, it's crucial to remove outdated or overly specific instructions that might hinder newer models, as these can negatively impact performance. Thirdly, instead of merely instructing the AI to perform a task, users should provide it with the necessary tools, such as calculators for mathematical operations, since instructions alone do not confer new capabilities to the model. Fourthly, when asking the AI to weigh options or make decisions, it's important to present both sides of any tradeoff, including costs and benefits, to enable a more balanced and informed response. Finally, for complex tasks, breaking down a single large prompt into multiple, simpler stages—like drafting, checking, and repairing—can lead to significantly better outcomes. The overarching principle behind these fixes is iterative testing and refinement of prompts to ensure they consistently deliver desired results.