Claude

How to Write Better with Claude (Without Sounding Like a Robot)

May 2026

By Mark Reeves — author of all 47 AI Field Guides. Written and tested in a working business.

AI-generated portrait. Pen name. Real work.

The output problem isn't Claude — it's the briefing.

The content strategist had rewritten three paragraphs of Claude's output before she realized she was rewriting all of it. The brief had been nine words. That was the problem.

Claude doesn't know who you are, who you're writing for, what you sound like, or what the piece needs to do. Without that, it produces the average of everything it's read — which is competent, structured, and completely generic.

The fix isn't a better AI. It's a better briefing.

Why AI Writing Sounds Generic

When you give Claude a vague prompt — "write a blog post about project management" — it makes assumptions. It doesn't know your audience, your voice, your point of view, or your format preferences. So it defaults.

It picks the most common structure for that type of content. It uses the most neutral version of a professional voice. It writes to nobody in particular.

The result reads like it was written by committee. Which, in a sense, it was — a committee of every document Claude was trained on.

The generic output isn't a failure of the model. It's the expected output when the model has no specific information to work with. Give it specific information, and the output changes.

What Claude Actually Needs to Write in Your Voice

There are five things Claude needs before it can write anything that sounds like you.

Role context. Who are you, and why are you credible to speak on this topic? Don't assume Claude infers it from context. Tell it explicitly: "I'm a solo consultant who has run 200+ discovery calls in the SaaS space" lands differently than "I'm a consultant."

Audience. Who is this for? Not a demographic — a specific person with a specific problem. "A freelance designer who's been using Figma for three years and wants to start charging more" is useful. "Small business owners" is not.

Tone reference. Tell Claude what the tone is, not just what topic it's covering. "Practical, direct, no hedging — think someone who's done the work and doesn't have time for caveats" is a tone instruction. "Professional but approachable" is too vague to mean anything.

Constraints. What's off-limits? What format do you need? How long? One idea per paragraph? Short sentences only? These aren't preferences — they're specs. Treat them like specs.

Format. Are you writing a LinkedIn post, a newsletter section, an article intro, a product description? The same topic written for different formats produces different output. Tell Claude what container the writing lives in.

With all five in place, Claude has something to work with. Without them, it's guessing.

The Voice Reference Technique

The fastest way to get Claude writing in your voice is to show it your voice before asking it to produce anything.

Paste in a paragraph or two of your actual writing — something you've already published or sent that represents how you write at your best. Don't explain it. Don't analyze it. Just paste it with this instruction:

"Here's an example of how I write. Before you produce anything, study the rhythm, the sentence length, the level of formality, and the kind of words I use. Don't mirror it mechanically — just let it calibrate your output."

Then make your actual request.

The difference is significant. Claude is capable of picking up on things you might not even consciously know about your own writing — how often you ask rhetorical questions, where you use one-word paragraphs for emphasis, whether you tend toward active or passive constructions. Give it the sample and it adjusts.

This works for any writing task: articles, emails, social posts, proposals. The sample doesn't need to be long. Two hundred words from the right piece is enough.

Editing AI Output vs. Starting from Scratch

A mediocre first draft is faster to edit than a blank page. This is not a preference — it's a function of how editing works.

When you edit, you're responding to something. You can see what's wrong and fix it. When you write from scratch, you're generating structure, argument, and language simultaneously. That's slower, and it's the part AI actually helps with most.

Even when Claude's first draft isn't quite right, it usually has something useful: a structure you can rearrange, a sentence you can steal, a frame you hadn't thought of. Your job shifts from building to shaping. That's faster.

The trap is over-editing. If you find yourself rewriting every sentence, you're doing the work twice. When that happens, it's usually a brief problem — the draft is too far off because Claude didn't have enough to work with. Fix the brief before editing the output.

The better pattern: edit for voice and accuracy, not for style from scratch. Punch up the first sentence. Cut the parts that sound like nobody said them. Tighten. Don't retype.

The Iteration Pattern That Works

The most common mistake when iterating with Claude is vague feedback.

"Make it punchier" doesn't work well. "Make the sentences shorter" does. "Make it sound more like me" doesn't work. "Remove the hedge words — 'perhaps,' 'might,' 'could be' — and replace passive constructions with active ones" does.

Claude responds well to specific, behavioral instructions. When something is wrong, name exactly what it is and tell Claude what you want instead.

A pattern that works:

"The second paragraph is too formal. Rewrite it so it sounds like I'm talking directly to one person, not addressing a room."

Another:

"The opening is burying the point. Lead with the main takeaway. Cut the setup."

These stick across a session because they're explicit. Vague direction gets vague correction. Specific direction gets specific correction.

When you're deep into a session and Claude has started to drift from the brief — outputs getting longer, tone getting more hedged — just restate the constraint directly: "You're adding qualifiers again. Remove them." It responds immediately.

When to Let Claude Lead vs. When to Outline First

Not every writing task benefits from the same approach.

Let Claude lead when the structure is obvious or doesn't matter much. Product descriptions, short-form social copy, email subject line variations, FAQ entries — these have a clear enough shape that Claude's initial structure is usually fine. Hand it a brief, review the output, edit. That's the whole workflow.

Outline first when the structure is the work. Long-form articles, sales pages, proposals where the argument has to unfold in a specific order — these need a skeleton before Claude starts writing. If you skip the outline and let Claude build the structure, you'll end up editing the argument and the prose at the same time. That's slow and frustrating.

The tell: if you're writing something where the order in which ideas appear actually matters to whether it convinces someone, outline it first. Give Claude the outline and ask it to fill in one section at a time. Review each section before moving to the next. Much easier to catch a structural problem in section two before Claude has written six more sections that depend on it.

Before and After: Same Task, Different Briefs

Here's the same writing task handled two ways.

The task: Write a product description for a practical AI guide aimed at small business owners.

Vague prompt:

"Write a product description for my AI guide for small business owners."

Output (typical):

Unlock the power of artificial intelligence for your business with this comprehensive guide designed specifically for small business owners. Learn how to leverage cutting-edge AI tools to streamline your operations, increase efficiency, and drive growth. Whether you're new to AI or looking to take your knowledge to the next level, this guide provides actionable insights and practical strategies to help you succeed in today's competitive landscape.

That paragraph was written by committee. It could describe any product in any category. It says nothing specific, makes no concrete promise, and uses "leverage," "streamline," "cutting-edge," and "competitive landscape" — four phrases that have been sanded down to zero meaning through overuse.

Specific brief:

"Write a product description for a practical AI guide aimed at small business owners who feel like they're behind on AI and don't know where to start. The guide gives them the exact prompts for their most time-consuming tasks — proposals, discovery calls, SOPs, hiring docs. Tone: direct, no hype, working-operator voice. No buzzwords. One idea per paragraph. Short sentences. Don't use the words 'leverage,' 'unlock,' 'streamline,' 'comprehensive,' or 'cutting-edge.' CTA: download for $9.99."

Output (revised):

Most AI guides tell you what the tools can do. This one tells you what to type.

The Claude for Business Owners guide is built around the tasks that actually eat your time — writing proposals, running discovery calls, documenting processes, drafting hiring materials. For each one, you get the prompts that work, the ones that don't, and a workflow you can run this week.

No theory. No "AI will transform your business." Just the prompts and the process, organized by the job you're actually trying to do.

Same model. Different brief. Different output.

Try It Before You Move On

Take something you're working on right now — an email, a LinkedIn post, a product description. Fill in this brief template and run it:

Role: [Who you are and why you're credible on this topic]

Audience: [One specific person with one specific problem — not a demographic]

Tone: [What it sounds like — give a behavioral description, not a label]

Constraints: [Format, length, banned words, required elements]

Format: [Where this lives — newsletter, product page, social post, etc.]

Voice sample: [Paste 1-2 sentences of your own writing that sound like you at your best]

Now write [your request].

If the output is noticeably better than what you've been getting, the brief was the problem all along.

Where to Go from Here

The techniques above — voice references, specific iteration instructions, knowing when to outline first — are the core of a working AI writing workflow. They're not tricks. They're the discipline that separates people who get good output from people who gave up after three tries.

This article covers the one-session voice reference approach. It doesn't cover how to build a Brand Voice Document that persists across sessions so you don't re-brief every time, how to handle voice drift when Claude gradually forgets your tone over a long document, or how to keep multiple collaborators briefing Claude consistently rather than each person developing their own style from scratch.

Related reading

• How to use Claude for business without leaking client data
• 5 prompts that show what Claude can actually do

Mark Reeves is a pen name. AI Field Guide publishes role-specific, practical guides for using AI tools in real work.