I’ve been typing “think step by step” at the top of my prompts for so long that it’s basically muscle memory. So I stopped scrolling when this AI professional said that trick is officially dead with Claude Opus 5.5. The author put together a quick guide on how to prompt the new model if you only had 5 minutes to learn it. It’s one of the most practical things I’ve seen on prompting in a while.
The best part is that none of it is complicated. It’s six small habits, and each one fixes a mistake most of us still make every day. Here’s the full process the expert shared, with what each step does and how to use it.
The 6 steps at a glance
- Pick Opus 5.5 and set the effort to “Medium”
- Delete “think step by step,” “think hard,” and “maximum reasoning”
- Stop typing your prompts and talk instead
- Tell Claude where to look
- Tell it why you care
- Tell it when it’s done
Each of these does a specific job, so here’s how the author explains them.
1. Pick the right model and effort level
Start by choosing Opus 5.5 and setting the effort to “Medium”. According to the original poster, “Extra” is only worth it for one huge task you want done in one shot.
I like this because most of us crank every setting to the max and figure more is better. More effort means more waiting, and for everyday work like emails, summaries, and quick drafts, Medium handles it fine. Keep “Extra” for the big jobs: a full report, a complicated analysis, or a long document you want right on the first try.
2. Kill the old “think harder” phrases
This is the one that surprised me. The creator says to delete “think step by step,” “think hard,” and “maximum reasoning” from your prompts. Opus 5.5 thinks on its own anyway, so those lines just slow it down.
The funny part is that the post’s author admits to teaching that exact trick before: “Yes, I gave you that trick myself. It’s dead now.” I respect that kind of honesty. Prompting advice ages fast, and something that helped older models can turn into dead weight on newer ones.
Quick tip: Go through your saved prompt templates and search for these phrases. If you’ve been pasting the same prompt for months, there’s a good chance it’s carrying some of this baggage.
3. Stop typing and start talking
Next, this LinkedIn creator suggests dropping the keyboard entirely. Download Wispr Flow (it’s free), hold Shift, and talk. The author found that Claude’s built-in mic button mangles their words, but Wispr doesn’t.
The real win isn’t speed, though. It’s detail. When you talk out loud, you naturally give context you’d never bother typing. Typing makes us lazy and short. Talking makes us explain ourselves like we would to a coworker, and that extra context is exactly what makes AI output better.
So that’s the input side sorted. The next three steps cover what you should actually be saying once you start talking.
4. Point Claude to the right sources
Tell Claude exactly where to find the information it needs. The example the expert gives looks like this:
“Read our emails with her, our refund policy in Drive, her contract.”
There’s a catch the original poster calls out: turn on Connectors first. Without them, Claude can’t open any of it. Connectors are what link Claude to tools like your email or Google Drive, so it can read the real documents instead of guessing.
This changes the whole quality of the answer. Instead of a generic reply, you get one based on your actual policy, your actual contract, and your actual conversation history.
5. Explain why you care
This step is small but powerful. Give Claude the motivation behind the task. The contributor’s example:
“She’s been a client for 2 years and I want to keep her.”
As the author points out, you’d never type that. But you’d say it in 2 seconds. That one sentence tells Claude the tone to strike, what to prioritize, and what’s at stake. A refund reply meant to keep a loyal client reads very differently from one meant to close a ticket.
This is where Step 3 really pays off. Speaking makes it natural to add the “why,” and the “why” is what turns a decent answer into the right one.
6. Define what “done” looks like
Finally, tell Claude when the job is finished. The example from the post:
“Under 120 words. Final draft, ready to send.”
Skip this, the savvy professional warns, and Claude gives you a draft, then asks if you want changes. That back-and-forth eats time. When you set the length and say you want a finished version, you get something you can use right away.
Putting it all together
Here’s how these steps stack into one spoken prompt, based on the examples in the post:
“Read our emails with her, our refund policy in Drive, her contract. She’s been a client for 2 years and I want to keep her. Under 120 words. Final draft, ready to send.”
Notice what’s missing. No “think step by step.” No “act as an expert.” Just where to look, why it matters, and what done looks like. That’s the whole formula.
A few more ways you could use the same structure:
- Meeting prep: point Claude to the agenda and last call notes, explain you need to win over a skeptical stakeholder, and ask for five bullet points max.
- Project updates: point it to the project doc and recent threads, say your boss only cares about blockers, and ask for a three-sentence summary.
- Sales follow-ups: point it to the call notes and proposal, mention the deal closes this month, and ask for a send-ready email under 100 words.
Why this approach works
The newer models don’t need tricks to think harder. They need context. Every step here either removes noise or adds useful context, and that’s where better results come from now.
I think that’s the big lesson. A lot of us learned prompting back when models needed hand-holding. Opus 5.5 doesn’t. It needs good information, a clear reason, and a finish line. Give it those three things and the rest takes care of itself.
If you’re still typing prompts with two fingers, this is your sign to try talking instead. Check out the full LinkedIn post to see the original breakdown and the prompt image the author shared.