I've been using LLMs daily for about 2 years (chatgpt, claude, perplexity, gemini). my prompts have gotten a lot better but honestly it's mostly because I changed how I input them, not because I learned some framework.
here's what actually moved the needle for me, ranked:
- saving good prompt templates in Obsidian (free)
when I get a great output from an AI, I save the prompt that generated it. I have a folder of prompts organized by task type (email drafting, code review, analysis, brainstorming). when I need to do something similar I pull up the template and modify it. this alone improved my results a lot.
- using the projects/custom instructions features
claude projects and chatgpt custom instructions let you set persistent context. I have my writing style, my role, my common tasks, and key preferences loaded in. every conversation starts with the AI already knowing who I am and what I do. the difference between starting from zero and starting from context is massive.
- Perplexity for research before prompting ($20/mo)
instead of asking chatgpt or claude to research something and risk hallucinations, I use perplexity first to get sourced information. then I feed that verified info into claude for synthesis, analysis, or drafting. the combo of perplexity for facts and claude for thinking is better than using either alone.
[link] [comments]