How to Make Claude Write Like a Human With Better Examples

To make Claude write like a human, give it source examples you approve and explain the specific choices it should follow. Then edit exact passages for meaning, detail, and tone rather than asking it to “humanize” everything. The goal is writing that sounds appropriate to its reader and says something real.

Keep the accepted guidance after the conversation ends. fenbs can hold approved writing prompts and lessons from revisions so a connected assistant can read them before the next assignment. It gives you somewhere to preserve the choices behind the voice, rather than repeatedly requesting a vague feeling.

What examples should you give Claude?

1. Choose writing for a similar audience and purpose. A calm service email is a better reference for another service email than a personal essay you happen to enjoy.
2. Use your own writing or examples you have permission to share. Remove private details that the task does not need.
3. Mark what to borrow: a direct opening, concrete verbs, an ordinary vocabulary, or a clear explanation of a difficult choice.
4. Mark what must stay behind: the sample’s names, facts, anecdotes, and situation. Those belong to that example, not the new draft.
5. Include a rejected sentence and your replacement if you have one. Explain why the replacement fits this reader better.

Anthropic’s prompting guidance↗ (opens in a new tab) recommends relevant, varied examples for steering tone and structure. Do not treat a sample as a magic fingerprint. Explain its useful choices so Claude can apply them to new material without copying its circumstances.

How do you turn a sample into clear guidance?

Ask Claude to describe the sample’s observable features, then check that description yourself. “Warm and authentic” is an impression. “Uses contractions, names the practical problem, and asks one clear question at the end” describes choices another draft can follow.

Suppose your sample says, “The cabinet door still catches on the frame. Could you send a photo with it half open?” The useful features are the concrete problem and the precise request. A new message should copy that clarity, not invent another cabinet or repeat the wording regardless of context.

Write an email using the facts below.
Reader and purpose: [describe both].
Voice reference: [paste a short approved example].
Borrow: its direct opening, familiar words, and concrete next action.
Do not borrow: its facts, names, personal experiences, or exact sentences.
Use contractions where they sound natural. Keep necessary qualifications.
If a detail is missing, ask or flag the gap. Do not invent an anecdote.
Facts for this email: [paste confirmed facts].
Return the draft, then list any factual gaps separately.

The Claude AI overview explains the product; the sample explains your local writing standard. Keep those jobs separate. An assistant does not know why your client disliked a draft unless you identify the sentence and the reason.

Which details make writing sound more natural?

Use the details a person in this situation would care about. A delayed order might need the affected item, the new expectation, and the available choice. It rarely needs a paragraph celebrating the company’s commitment. The practical information gives the voice a reason to exist.

Prefer a specific observation over a general emotional claim when you have evidence. In a hypothetical project update, “The review found two missing captions” tells the reader more than “We made meaningful progress.” Specificity does not require colorful language. Often it means replacing a broad judgment with a useful fact.

Use ordinary sentence variety without imposing a pattern. A short sentence can state the consequence. The next can explain the reason and its condition. If every sentence is clipped or every paragraph ends with a slogan, the voice can become another formula.

How do you give feedback that changes the next draft?

Quote the passage, name the problem, and show the kind of replacement you want. “This sounds robotic” tells Claude that you disliked something but leaves the cause unclear. “This thanks the reader twice before answering; keep one acknowledgment and put the answer next” is actionable feedback.

When an edit works, save the before-and-after comparison and the reason. fenbs Lessons learned can preserve that correction for similar work, while Skills and prompts holds the reusable instruction. A lesson should explain the mistake, not just say that the draft was bad.

Should you add mistakes or personal stories?

Do not add fake typos, invented memories, or pretend customer experiences to create a human impression. Those additions can distract readers or make a factual piece misleading. If your own experience matters, provide the real details and check that the draft keeps them accurate.

Anthropic’s guidance on reducing hallucinations↗ (opens in a new tab) recommends allowing uncertainty and grounding responses in supplied material. Apply that boundary to voice work: a vivid sentence still needs a truthful basis. Natural tone should not strengthen an unsupported promise.

For a different assistant, see making ChatGPT write like a human; for a nearby voice-focused approach, see making Claude sound human. Examples from a Claude and ChatGPT coding comparison serve a different task, so do not assume a good technical response is the right voice reference for customer copy.

How to do this with fenbs

1. Create a free account and add a writing project. It is free for one person, with no credit card.
2. Save a reusable prompt in Skills and prompts. Include the audience, approved example, qualities to borrow, and details that must not be copied.
3. Record the voice rules a person approves in Decisions and rules. Add a Lesson learned when a repeated edit reveals a correction worth applying next time.
4. Connect Claude using Integrations, ask it to read the project context before drafting, and save the task’s progress with any review still needed.

fenbs does not train Claude on your writing or import your past chats. If you only need one email, a sample pasted beside the request may be enough. Keeping shared guidance becomes useful when the same voice needs to carry across sessions, people, or supported AI apps.

What should you check before using the result?

Read the draft as the intended recipient. Does it answer your question, use language appropriate to the relationship, and include the details you need? Then compare it with the facts. Check for borrowed sample details, invented experience, changed promises, and missing conditions.

Finally, read it aloud and revise anything you would struggle to say in that setting. Preserve your own judgment rather than approving every sentence that sounds smooth. The AI prompts collection has more focused ways to address other response problems without stuffing every instruction into one writing prompt.