I go through Claude tokens way faster than I’d like to admit. One long chat, a couple of model switches, a short argument with the AI, and I hit a usage limit before the real work is done. So when I found this post from an AI professional who fit a dozen Claude tips into a “57 seconds read,” I stopped scrolling.
The author says these are the Claude tips you can’t afford to ignore. After reading them, I agree. Most take less than a minute to set up, and together they change how much you get from every prompt. Below I go through each one and add a bit of context on why it works and how to use it.
1. Use the Desktop App, Not the Browser
The first tip is to use the Claude desktop app instead of Claude in the browser, even after Cowork is gone. The big win is that you can work with your folders directly and never upload files. The original poster says you’ll also use Claude Code more, and that “there is no coming back from there.” I agree. Dragging files into a browser tab every time gets old fast.
2. Make Opus 5.5 on High Your Default
This one surprised me. The expert says your default should be Opus 5.5 on high, not Fable 5.2. According to the post, Fable 5.2 is “insanely good” but uses up your tokens in just 3-4 prompts, so you won’t be able to finish a full task with it. Let Opus 5.5 handle your daily work and keep the heavier model for rare cases.
3. Stick With One Model Per Chat
Switching models in the middle of a chat feels harmless, but it isn’t. The creator explains that Claude re-reads the entire chat to get the context again, which uses more tokens. Pick your model at the start and stay with it. My tip: if you really need another model, open a fresh chat and paste a short summary of where you left off.
4. Delete Your Old Claude Setup
This one is bold. The author suggests deleting the “Instructions for Claude” box in settings and removing old Claude skills you added earlier. The reason is that they were written for older models and might be slowing you down. Instead, give Claude one goal and end your prompts with this line:
“Before you start, restate my goal in your own words.”
I love this because it catches misunderstandings before Claude spends tokens going in the wrong direction.
5. Turn Your Memory On
The post’s author cites a test by the Claude team that compared results with and without memory. With memory, Opus 5.5 scored 28% better. It’s a quick switch: click the “+” in the chat box and turn on “Memory.”
6. Keep Your Memory Short
More memory isn’t always better, though. Before one session, Claude had 105 facts stored about the author. After one cleanup prompt, that dropped to 26. Fewer, sharper facts give Claude less noise to sort through. To try it yourself, ask Claude to review what it remembers about you and cut anything outdated or repeated.
7. Ask for HTML, Not a Document
This might be my favorite. Add two things to your prompt: “Make it an HTML page” and “for [who will read it].” Claude will code the image, doc, or any other format you need, and the creator says this leaves no room for mistakes. It makes sense, since code gives Claude exact control over layout. For example: “Make it an HTML page for my sales team summarizing Q3 results.”
8. Never Argue With Claude
We’ve all pushed back on a bad answer and gone back and forth. The LinkedIn creator says this is pointless: you burn tokens and the task still doesn’t get done. Instead, click “Edit” on the prompt you sent right before the bad answer. Change, edit, or fix it, then click “Save.” The bad response is gone, so it can’t steer the rest of the chat.
9. Drop the Rulebook
Tip nine builds on tip four. Delete your instructions box: remove the “about me” section, the “how I write” file, and your list of rules. The author argues these all force Claude to limit its creativity. Aim for one clear goal per prompt instead. I’ll admit this one made me think twice, since custom instructions take time to build. But the logic holds up: newer models do better with a clear goal than with a long list of rules.
10. One Claude Project Per Client
This one is for freelancers, consultants, and agencies. Create a separate Claude Project for each client, and inside each one turn off “Use account memory.” That way, what Claude knows about Client A doesn’t get mixed up with Client B. Each client’s work stays separate, and one client’s details never slip into another’s work.
11. Ask for the Spread
By default, Claude gives you the most generic answer. As the original poster puts it, that’s literally how it works. To get the weird, unlikely ideas, use this prompt:
“Give me 20 very different options. For each one, tell me how likely another AI would give it to me. The last two must be under 10%.”
This works well for brainstorming headlines, product names, campaign angles, or any task where you need something your competitors’ AI won’t give them.
12. Give Claude a Finish Line
Finally, tell Claude what “done” means to you. Here’s the example from the post:
“Done means: a one-page doc my boss can forward without editing.”
A clear target takes away the guesswork. You can use the same pattern anywhere, like “Done means: a five-bullet summary I can read in 30 seconds” or “Done means: an email my client can approve with one reply.”
Why These Tips Work So Well Together
Taken together, the tips follow one pattern: less clutter, clearer goals, and smarter token use. Clean up old instructions, keep memory short, stay with one model, and fix bad answers by editing instead of arguing. Each habit saves a few tokens on its own. Together, they make your Claude sessions faster and sharper.
If you only try three today, I’d start with editing instead of arguing, asking for HTML, and giving Claude a finish line. They take seconds to set up and you’ll see the difference right away.
Want to read the tips in the author’s own words? Check out the full LinkedIn post for all the details.