Short, practical tips for working with AI — Claude, ChatGPT, Gemini and more.

When you open a conversation with Claude, ChatGPT, or Gemini — don't jump straight to the task. Start with: "I am a [role] working on [topic]. I need to do X." Why does this matter so much? AI models try to guess your context. Once you give it to them, they don't need to guess — and you get an answer tailored to you, not to the average user. Bad example: "Write me a marketing email" Good example: "I'm a product manager at a SaaS company. Our audience is CFOs at mid-size businesses. Write a short marketing email introducing the automated reporting feature we just launched." The difference between the two requests is a few seconds. The result? Completely different.

One of the things that separates people who use AI routinely from those who get real value from it: a personal prompt library. The idea is simple: every time you write a request that worked well — save it. Notion, Google Docs, a text file — the tool doesn't matter. What's worth saving? - Prompts for writing different types of emails - Prompts for summarizing meetings - Prompts for data analysis - Prompts for editing and improving texts - Prompts for weekly planning After a month or two, you have an arsenal that saves you hours. You're not starting from scratch with every request — you're starting from a proven starting point.

Claude, ChatGPT, Gemini — they're all "artificial intelligence," but they're not equal for every task. Just like there's a chef's knife and a paring knife — both are knives, but not interchangeable. Some useful rules of thumb: - Claude (Anthropic): Excellent for writing, long-form content, document analysis, complex instructions. Loves context and improves the more you give it. - ChatGPT (OpenAI): A versatile multi-tool, including code writing, web search, DALL·E for images. - Gemini (Google): Deep integration with Google Workspace — Docs, Sheets, Gmail. Useful if you work heavily with Google tools. The real tip: Don't marry one tool. Try a few for the same task and see who answers you better. Over time you'll develop clear intuition.

Five minutes before an important meeting — if you talked to AI beforehand — you show up completely differently. Here's the prompt I use: "I'm about to meet with [role/company] regarding [topic]. Meeting goal: [goal]. Help me with: a) What questions should I ask? b) What are my weak points in this meeting? c) What's the best pitch for this audience?" Result: you enter the meeting with sharp questions, ready for challenges, and clear on what you want to get out of it. This isn't cheating — it's smart preparation. Bonus: After the meeting, plug in the key points to AI and ask for a summary + required actions.

Before you send an important email — let AI read it. Not to rewrite it, but to catch what you're missing. The prompt: "Here's an email I wrote: [text]. The recipient is [role/context]. The goal: [what I want to happen]. a) Is the message clear? b) Is there anything that might come across wrong? c) What's the CTA — is it clear enough?" You'll be surprised how often you get a response like: "The second sentence might sound defensive" or "The CTA isn't clear, consider phrasing it as a direct question." It's not about the AI writing better than you. It's that from its perspective — it's reading the email for the first time, exactly like the recipient will.

AI can be wrong. That's well known. The problem is it's wrong with complete confidence — and no hint that something's off. The simple solution: cross-check. After getting an important answer from Claude, plug it into ChatGPT and ask: "Is this answer correct? What's missing from it? What might be wrong?" And vice versa. This works because each model was trained on different data, with different biases, and different strengths. What one misses — the other sometimes catches. Especially important for: - Factual information (dates, numbers, names) - Legal/financial/medical advice - Critical production code - Business decisions based on AI data Remember: two wrong AIs don't equal one right one. But two that agree — that's already much more reliable.