Transcription
Here are five hidden Claude tricks that 99% of users have no idea about.
Number one is meta prompting. Now, instead of just writing your prompt, you can just ask Claude to write me the perfect prompt to get this result and then you just paste its answer back into your actual prompt. Your output quality, it goes up 10 times because Claude knows what Claude needs better than you do.
Number two is XML tags. Now, most people write Claude just one big block of text. But wrap your inputs in tags like context, task, and example, and Claude's outputs gets sharper instantly. This is literally how Enthropic's own engineers are prompting it.
Number three is PDF visual mode. So, when you upload a PDF under 100 pages, Claude actually sees the layout, the charts, the tables, the signatures, and not just the extracted text. So, you can just drop a contract, a dashboard, a screenshot, or a financial report and just ask, "What is wrong with this?" and it'll catch all the stuff that a normal text scanner would completely miss.
Number four is the self-critique loop. So after Claude gives you any answer, just reply to identify the three weakest parts of your last response and rewrite them. Now the second past, it is almost always twice as sharp as the first.
And number five is the voice ramble. So in Claude, you can just hold down the microphone button and just talk for any amount of minutes about a problem that you're trying to solve. You don't have to structure it because COG turns that ramble and that waffling into a clean SOP, an email, or a proposal in seconds.
If you want our weekly AI newsletter breaking down all the best AI strategies to grow your business, just comment.