Here’s a frustration I hear all the time: people copy prompting tips they saved back in 2023, paste them into Claude, and wonder why the answers feel flat. The tricks that worked on older models can actually make you worse now, and you’d never know it.
That’s exactly what caught my eye in a sharp post from an AI professional named Ruben Hassid. The original poster laid out a set of updated rules for prompting Claude, and I was genuinely surprised by how many old habits made the “stop doing this” list. I’ve been guilty of half of them myself, so I want to break down what the creator shared and why each rule matters.
Why old prompting advice goes stale
Models change. Fast. A trick tuned on a 2023 model can behave completely differently on today’s version, yet most of us never re-test our favorite hacks. The expert’s core point is simple: treat prompting like a skill that expires, not a rulebook you memorize once. With that mindset, here are the rules worth adopting now.
The updated prompting rules
- Never end questions with “right?” Tacking on “right?” (or worse, “maybe?”) signals the answer you want, and the author points out that AI is very good at reading those signals. It won’t push back to please you. Skip yes/no framing entirely. Instead of fishing for agreement, ask “Compare X vs Y for this [context]” and let the model actually reason.
- “Think step by step” doesn’t pull its weight anymore. The original poster argues that the old “think step-by-step” and “gather info, then plan” phrases have lost their punch on newer models. The stronger move is to assign a role in just a few words, like “…as a reliability engineer.” That single tweak reshapes the whole response.
- Never trust AI’s cited sources. This contributor is blunt about it: AI sounds confident even when it’s wrong, and cited links may not even exist. Open every single source before you rely on it. Confidence is not evidence.
- Give only 3 rules per prompt. The creator’s take is that the model stops reliably reading past three rules in one shot. If you have more, feed them in follow-up prompts. A handy second follow-up: “Check your draft against these ONE AT A TIME, then revise.” You get every rule applied instead of the first few.
- Stop giving examples. This one flipped my thinking. Rather than “Example 1: [solved case],” the author suggests giving a clear goal, pasting the real numbers and context, and naming what you’ve already ruled out, like “I didn’t change format, send time, or subject.” Then close with a line that forces precision: “Ask me for the data you need before you answer.”
- No prompting rule is ever permanent. The mind behind this post ends on the most important idea. Every hack you love was tested on some specific model. When a new model drops, re-test your tricks before you trust them. Green today, stale tomorrow.
Why it matters: the person who shared this isn’t handing you a permanent cheat sheet. He’s teaching a habit. Prompting well means re-testing what you assume works, because the ground keeps shifting under you.
How to put this to work today
You don’t need to overhaul everything at once. Pick one prompt you use often and run it through these upgrades:
- Strip out any “right?” or “does this make sense?” endings that beg for agreement.
- Swap “think step by step” for a one-line role that fits the task.
- Trim your instructions down to three rules, then layer the rest in follow-ups.
- Replace your example blocks with a goal, real context, and a request for the model to ask you for missing data.
- Click through every link it cites before you quote it anywhere.
Try the same prompt the old way and the new way, side by side. I think you’ll feel the difference in a single session, and that little test is really the whole lesson in action.
The reason I love finding posts like this is that they turn a vague feeling (“my prompts got worse”) into something you can actually fix. This LinkedIn creator packed a lot of practical value into a short post, and these six rules are the kind of thing you’ll want to keep handy.
Give the full LinkedIn post a read for the complete breakdown, then go re-test one of your own favorite prompting tricks. Delete the 2023 habits, save the new ones.