DEV Community: Hubai Prompt

How to Write Claude Prompts: 4 Proven Tips for Better AI Results

Hubai Prompt — Mon, 02 Nov 2026 10:01:07 +0000

You know that feeling when you ask Claude for something, get back a flat, generic answer, and just end up rewriting it yourself? That happens to almost everyone.

It's not really Claude's fault, though. It's the prompt.

Most of us write Claude prompts the way we'd ask a coworker a quick question — one sentence, not much context, and hope it works out. Claude fills in whatever's missing with a guess. And a vague prompt usually gets you the safest, most average guess it can make.

Once you see the pattern, it's easy to fix. So this is a straightforward look at how to write Claude prompts that actually work — what makes a good one, a real example of a weak prompt versus a strong prompt, and a few small mistakes worth avoiding.

Why Most Claude Prompts Fall Flat

Usually it's because we write them the way we'd think out loud. Something like "write me a blog post about productivity." That's a topic, not really an instruction. Claude doesn't know who it's for, what tone to use, or how long it should be.

So it fills in those blanks itself, with whatever feels safest. That's why vague prompts tend to come back sounding a little generic, there just wasn't enough to go on.

The other common thing is rewriting a similar prompt from scratch every time. Small wording changes end up giving you slightly different results, which makes things feel inconsistent even when you're asking for basically the same thing.

What Actually Makes a Claude Prompt Work

There are really just four parts to writing a good Claude prompt.

The first is context — who you are, who it's for, what's going on. "Write a follow-up email" doesn't give Claude much. But something like "I just had a discovery call with someone who seemed interested but didn't commit, and I want to follow up without sounding pushy" gives it a real situation to work from.

The second is the goal. Not the general topic, but what you actually want made. "Write about SEO" is a topic. "Give me a 3-step plan a small business owner can follow this week to fix their page speed" is a goal.

The third is guidelines, and this is the part people skip most often. Tone, length, what to include, what to avoid. Just say it plainly — "keep it under 100 words," "no corporate language." Each one you add is one less thing Claude has to guess.

The fourth is output format. Do you want an email, a list, a short paragraph? If you don't say, Claude just picks whatever feels standard, and that's not always what you needed.

A Quick Example: Weak Prompt vs. Strong Prompt

Here's a weak Claude prompt: "Write a cold email for my consulting business."

There's not much to work with there — no audience, no offer, no tone, no length. So the result ends up generic enough to fit almost any consulting business, which means it probably won't fit yours.

Now here's the same task, done with more detail: "I run a small marketing consultancy helping local service businesses — plumbers, electricians, dentists — get more leads from Google. I need a cold email, under 100 words, for someone who's never heard of me. Open with a specific, relevant pain point, not a generic compliment. Keep the tone direct and confident, not salesy. End with a low-pressure ask, like 'worth a quick look?' instead of 'book a call now.'"

Same task, but there's a lot less room for Claude to guess wrong this time.

Common Mistakes When Writing Claude Prompts

Being vague about your audience is more common than people think. "Small business owners" covers a lot of ground — it helps to say what kind, what size, and what they actually care about.

If you skip tone entirely, you'll usually get something neutral back, which rarely sounds like how you'd actually talk.

Asking for too much in one prompt — five different things at once — tends to give you five okay things instead of one good one. It's usually better to split it up.

And it helps to mention what to avoid, not just what to include. Something like "no clichés, no corporate phrasing" often makes more difference than the actual request itself.

Habits That Help You Write Better Claude Prompts

If a prompt worked well once, save it. You can usually reuse it with small tweaks instead of starting over each time.

It also helps to share a few examples of your own writing before asking for something new — gives Claude something real to match instead of guessing your style.

And if the first draft isn't quite right, it's worth going back and fixing the prompt instead of just editing the output by hand. That way the fix carries over the next time you use it too.

FAQ: How to Write Claude Prompts

What's the best way to write a Claude prompt?

Give it context, a specific goal, clear guidelines on tone and format, and the output shape you want. Those four things cover most of it.

Why does Claude keep giving me generic answers?

Usually because the prompt leaves gaps, and Claude fills them with its safest, most average guess. Adding specifics tends to fix that.

How long should a good prompt be?

However long it takes to remove the ambiguity. Simple tasks might need just one clear sentence. More complex ones usually need a bit more context and detail.

Should I write a new prompt every time, or reuse ones that worked?

Reuse what's worked before. If it got you a good result once, it's usually worth adjusting slightly rather than starting from scratch.

What's the biggest mistake people make with Claude prompts?

Being vague about the goal and skipping guidelines. Those two things tend to matter the most.

Can I tell Claude what tone to use?

Yes, and it helps a lot. Naming a tone — conversational, direct, warm, formal — is a simple addition that makes a real difference.

Wrapping Up

A flat, generic answer usually just means one piece was missing — context, goal, guidelines, or format.

So next time that happens, it's worth going back to the prompt instead of rewriting the output yourself. Once you get the hang of how to write Claude prompts this way, it stops feeling like guesswork — that's usually where the actual fix is.

If you'd rather not build prompts from scratch, there are other guides on the blog for cold email, proposals, and content planning that are ready to copy and adjust.

AI Blog Writing: How to Produce Good SEO Content That Ranks on 1st page

Hubai Prompt — Mon, 02 Nov 2026 09:31:33 +0000

You published the post two weeks ago. AI wrote most of it, it reads fine, and it's sitting on page four of Google, getting nothing.

That's not because Google penalizes AI blog writing. It doesn't — there's no filter that demotes a page for being AI-assisted. What gets ignored is content that reads like every other AI-generated post on the same topic: technically correct, oddly generic, and missing the thing that would make a reader actually trust it.

Producing AI blog writing that turns into real SEO content that ranks isn't about finding a better prompt. It's a process — one most people skip half of, then blame the AI when the post doesn't perform.

*Why Most AI Blog Writing Never Ranks
*
Ask an AI tool to "write a blog post about X" with no other input, and you'll get exactly what that prompt deserves: broad, safe, and indistinguishable from a hundred other posts targeting the same keyword.

Google's own guidance has been consistent on this — it doesn't rank content based on whether a human or a machine wrote it. It ranks based on whether the content is genuinely useful, and whether it shows real experience with the topic rather than just restating what's already ranking.

The gap between AI content that ranks and AI content that doesn't is almost never the writing quality. It's the input. A model given a real angle, a specific reader, and actual expertise to draw on produces something very different from one given a vague topic and nothing else.

*Start With Search Intent Before You Open Any AI Tool
*
Before any drafting happens, know exactly what someone searching your target keyword actually wants. Are they trying to learn something, compare options, solve a problem, or ready to buy? Each of those needs a different post, even for the same keyword.

Pull up the current top-ranking pages for your keyword and look at what they have in common — not to copy them, but to see what Google has already decided satisfies that search. If every top result is a comparison table, a 2,000-word narrative guide probably won't outrank them no matter how well it's written.

This step takes ten minutes and it's the one most AI blog writing skips entirely, which is exactly why so much of it misses the reader's actual intent.

*Structure the Post Before You Generate a Single Sentence
*
AI tools write better when the structure already exists. Draft your H1, your H2s as searchable phrases, and a one-line note on what each section needs to cover — before asking anything to generate full paragraphs.

This does two things. It keeps the AI from wandering into filler between sections, and it means your keyword placement — in the H1, in the first hundred words, in at least one H2 — is decided upfront instead of stuffed in during a final edit.

A post built from a real outline reads like it was planned. A post generated from a single broad prompt reads like it wasn't, and both readers and Google can usually tell the difference.

*Draft With AI, Then Add What AI Can't Fake
*
Let the AI produce the first full draft from your outline — that's the part it's genuinely fast at. But the draft isn't the finished post, and this is where most AI blog writing stops when it shouldn't.

Go back through and add the parts a model can't generate on its own: a specific example from your own work, an opinion that takes a real position, a number or result that isn't generic. This is the difference between content that demonstrates experience and content that just describes a topic from the outside.

This editorial pass doesn't need to be long. Even two or three genuinely specific additions to an AI draft change how the whole post reads.

*The On-Page SEO Checklist Every AI Post Needs
*
Before publishing, confirm the basics are actually in place — AI drafts frequently miss these even when the writing itself is solid.

Keyword in the H1, in the first hundred words, and in at least one H2. Meta description under the character limit with the keyword included. At least two internal links to related content on your site, one external link to a credible outside source, and one image with descriptive alt text containing the keyword.

None of this is complicated, but it's the layer that turns a well-written post into one Google can actually understand and rank for the right query.

*Turning This Into a Repeatable Process
*
The steps above work, but doing them from scratch for every single post is exactly the kind of repetitive setup that eats the time AI was supposed to save you. The fix is the same one that matters for any AI writing workflow: a structured, reusable prompt instead of rebuilding instructions every time.

JSON Prompt Mastery for Claude includes structured prompts built around exactly this kind of workflow — outline first, draft with the right SEO structure already built in, so you're not manually re-explaining your process on every post.

If you're already batching content in monthly sessions, Batch Content Creation With AI covers how to run this entire process across a full month of posts instead of one at a time.

*Mistakes That Get AI Blog Content Ignored or Flagged
*
A few patterns show up constantly in AI blog writing that doesn't rank, and they're worth checking your own posts against directly.

Publishing at scale with no editorial pass. Google's spam policies specifically target content mass-produced primarily to manipulate rankings. One AI-drafted, human-edited post is fine. Fifty unedited posts published in a week is the pattern that gets flagged.

Targeting a keyword with no real angle. If your post says nothing a reader couldn't get from the five posts already ranking above it, there's no reason for Google to rank it above them either.

Skipping the structure step. A post generated from one broad prompt almost always reads thinner than one built from a planned outline, even when the sentence-level writing quality is identical.

*Getting AI Blog Writing to Actually Rank
*
AI isn't the reason a post fails to rank — thin, generic content is, and AI makes it easier to publish more of that faster if you skip the steps that make content genuinely useful.

Start with real search intent, build the structure before drafting, add what only you can add, and check the SEO basics before you hit publish. That process is the actual difference between AI blog writing that sits on page four and SEO content that ranks.

If you want the writing tools comparison that goes with this process, 12 Best AI Writing Tools for Professionals in 2026 breaks down which ones hold structure well enough to make this workflow fast — and JSON Prompt Mastery for Claude turns this whole process into a prompt you reuse instead of rebuild every time you sit down to write.

DEV Community: Hubai Prompt

How to Write Claude Prompts: 4 Proven Tips for Better AI Results

Top AI Writing Tools for Marketing Content Creation: 9 Tools Worth Trying in 2026

AI Blog Writing: How to Produce Good SEO Content That Ranks on 1st page