Most advice on LinkedIn is making you worse at Claude. These are the 7 things I'd tell anyone starting with Claude: 1. Always start a chat with Fable-5, on High. After 2 prompts, switch to the Opus-5 model mid-chat. Find the solution to problems with Fable-5 but let Opus-5 do the job. 2. Never give Claude a long list of rules to stick to. Keep it to a maximum of 3 rules. Let it think, instead of thinking of following the rules. 3. Open a chat with this prompt: "I need [task] for [goal]. I'll know it worked when [target]. Ask me questions first, using the AskUserQuestion tool." Answer the form by giving more context in the "Something else" box, in your own words. 4. Always remember: New task, new chat. Otherwise Claude re-reads the entire chat with every prompt you send and starts hallucinating after 40 prompts. 5. Turn on Research mode for deep and complex questions. It reads 40 sources so you don't. But do not blindly trust the links. Always open the source link to double-check. 6. If you want to create an image, use ChatGPT. Or ask Claude for an HTML image, so doesn't mess up the accuracy of the text. 7. Click the citations links any AI model gives you. Claude thinks of an answer first, then finds a source that agrees (sometimes the citation is written by AI itself). If you do one thing from this, do # 3. Download my entire Claude Tips guide at This is as simple as it can get. Stop unnecessarily wasting 6 months figuring out which ones matter.