What this covers
Most “AI writing” tutorials show you a one-liner prompt and a polished result, skipping the part where the first draft is overwritten three times and still sounds generic. This walkthrough gives you the actual prompt structure that produces shippable prose on the first try: audience + thesis + evidence + voice sample + revision loop. The same pattern works for blog posts, essays, scripts, and short reports.
Who this is for
Writers, content marketers, founders writing their own copy, anyone shipping prose under deadline. Especially useful for people whose drafts technically read fine but get edits like “feels generic” or “could be any AI.”
When to reach for it
Drafting essays, blog posts, newsletter issues, video scripts, internal memos, launch copy. Anywhere voice and structure both matter. Skip for pure brainstorms — Claude is better at executing a brief than inventing one.
Before you start
- Name the audience precisely. “Developers” is too vague. “Senior backend engineers evaluating a new database” is workable.
- Write the thesis as one sentence. If you cannot, you do not have a thesis yet — and Claude cannot save you.
- Collect 5 evidence points: data, quotes, anecdotes, examples. Without these, the draft will be opinion-flavored air.
- Find a voice sample: a piece of your prior writing or an author you want to sound like. Paste 2-3 paragraphs.
Step by step
- State audience, tone, and goal in one block:
<audience>
Senior product managers at SaaS companies, 5+ years experience,
skeptical of AI hype, decision-makers on tool purchases.
</audience>
<tone>
Confident, plain English, dry humor allowed, no marketing adjectives.
</tone>
<goal>
Convince readers that AI prompt libraries are worth building internally,
not buying from a vendor.
</goal>- Provide the thesis as one tight sentence. Example: “Internal prompt libraries beat vendor tools because they encode tacit team knowledge that vendors literally cannot ship.”
- Provide 5 bullet points of evidence. Concrete numbers, named examples, specific anecdotes. No generalities.
- Provide a voice sample — 2-3 paragraphs of writing you want the draft to mirror. This is the single highest-leverage input.
- Ask for the draft with explicit format: word count, paragraph count, intro/body/close structure, any required elements (subhead every 200 words, etc).
- Revise in two passes. First pass: structure (“move the counterargument earlier”). Second pass: line edits (“kill three adverbs in paragraph two”). Mixing passes wastes both.
First-run exercise
- Pick a piece you have already written and shipped. Reverse-engineer the prompt: write the audience, thesis, evidence, voice, and format that would have produced it.
- Run the prompt with Claude. Compare the draft to your real piece.
- Note the gap: voice, structure, evidence weighting, or all three? The gap tells you which input to strengthen for new work.
- For the next real piece, lead with whichever input was weakest in the test.
Quality check
- Read the draft out loud. AI-generated cadence shows up in awkward sentence lengths and missing rhythm. Fix as you read.
- Verify every quoted statistic, name, and claim. Claude will confidently invent the occasional source.
- Look for “AI tells”: triplets (“efficient, scalable, and powerful”), hedge phrases (“it’s important to note”), and meta-commentary (“Let’s dive in”). Strip them.
- Check that the thesis is still on the page. AI drafts tend to drift toward balanced both-sides framing.
How to reuse this workflow
- Save the audience + tone + voice sample blocks as a snippet per publication you write for. Swap the thesis and evidence each time.
- Build a “voice library” with 3-5 voice samples for different contexts (newsletter, blog, internal memo, deck). Pick the closest match per piece.
- Keep a
clichés-to-kill.mdlist of phrases Claude overuses in your work. Paste it into Custom Instructions. - Re-run a test piece quarterly. Model voice shifts subtly with each update.
Recommended workflow
Audience + tone + goal → one-sentence thesis → five evidence bullets → 2-3 paragraph voice sample → first draft with explicit format → structural revision → line edits → read aloud → publish.
Common mistakes
- No thesis. Without a spine the draft can technically be “good writing” and still say nothing.
- Asking for “good writing” without naming the audience, tone, or length. Claude defaults to generic blog voice.
- Skipping the voice sample. Voice is the input with the highest leverage and it is the one most people skip.
- Letting Claude invent evidence. Anything specific that Claude generates without a source should be treated as a placeholder until you verify.
- Editing structure and lines in the same pass. You will rewrite the same paragraph four times.
- Publishing without reading aloud. AI cadence is detectable; reading aloud catches it.
FAQ
- Will my readers know it was AI-written?: They will if you skip the voice sample and revision passes. With them, the gap closes substantially.
- How long should the prompt be?: For a 1000-word piece, expect a 400-700-word prompt. Voice sample is most of the length.
- Can Claude write in my exact voice?: Closer than you expect, given a strong voice sample and Custom Instructions. Not identical without iteration.
- What about hallucinated statistics?: Common. Treat every specific number, date, or quote as a placeholder until verified.