Rewrite, Change the Tone, or Translate with AI (Without Changing What You Meant)

Make your own paragraph clearer, or make stiff English sound natural.

Level 1 · Chat Free · 30 minutes · Updated Oct 31, 2026

You wrote something, and it’s not quite right. Too long for your boss, too blunt for a client, too stiff for a colleague, or in the wrong language for the person who needs it. This is one of the things AI does best, and also where it quietly does damage: the new version reads better, and you don’t notice that “we aim to” became “we will,” or that a number disappeared. In the next 30 minutes you’ll rewrite a real paragraph and translate a short message, and learn a quick way to prove the meaning survived.

Remember the intern: smart, fast, first day on the job. Ask them to “make this better” and they’ll make it smoother, and sometimes say something you didn’t mean. Your job today is telling them exactly what kind of change you want, and checking what they changed.

What you’ll get

What you need

• Claude, any plan. Just a normal chat. Other tools work too; see the end.
• A paragraph of your own that needs work: an update, a request, a draft for a client. We’ll use this one throughout: a five-sentence project update for your leadership team that’s too long and too technical.
• A short message to translate (optional). Ours: a reply to a Spanish-speaking supplier confirming a new delivery date. If you don’t read the target language at all, that’s fine; Step 6 shows how to check anyway.

The usual boundary applies: don’t paste confidential details into a tool your company doesn’t allow.

The three rules

Part 1: Rewrite and translate, with proof (30 minutes)

Six steps: four for rewriting, two for translating. Copy the prompts as they are, then swap in your own text.

Step 1. Write your “must not change” list (3 minutes)

Before opening the AI, read your paragraph and write down everything that has to survive the edit:

• Numbers and dates (“launch moves to March 12”, “budget is 8% over”)
• Names of people, teams and products
• Commitments and how strong they are (“we aim to” is not “we will”)
• The one thing you want the reader to do

This takes three minutes and makes every later check fast.

Step 2. Name the change you want (5 minutes)

Open a new chat in Claude and paste:

Please rewrite the paragraph below.

Reader: our leadership team. They're busy and not technical.
What I want: shorter (under 80 words) and in plain language. Lead with what they need to decide.
Don't change: any number, date, name, or how strong a commitment is. Don't add facts or reasons that aren't in my text.
If something in my text is unclear, ask me instead of guessing.

Paragraph:
[paste your paragraph]

Step 3. Make it show its changes (4 minutes)

Now give me a table of every change: original wording | new wording | why.
Then list any sentence where the meaning might have shifted, even slightly, including commitments that got stronger or weaker.

This turns “does it feel the same?” into a list you can go through in two minutes.

Step 4. Check the meaning yourself (5 minutes)

Take your list from Step 1 and tick each item off in the new version. Then read the change table and look for three things:

• Shifted strength. “Might” became “will”, “some customers” became “customers”, “slightly over budget” became “over budget.”
• Things that vanished. A caveat, a condition, a second option.
• Things that appeared. A reason, a benefit, or a promise you never wrote.

Fix anything that fails, by hand or by asking: “Put back ‘we aim to’ in sentence 2.”

Step 5. Translate with a brief, not just a language (6 minutes)

Translation is a rewrite into another language, so the same rules apply. In a new chat:

Translate this message into Spanish.

Reader: our contact at a supplier in Mexico. We've worked together for two years; friendly but professional.
Use the formal "usted".
Keep every number, date and product name exactly as written.
After the translation, list any words or phrases where you had to choose between options, and why you chose what you did.

Message:
[paste your message]

The last line matters. It shows you where the translation involved judgment, which is exactly where meaning can slip.

Step 6. Check it by translating it back (7 minutes)

You don’t need to speak the language to catch big problems. Open a new chat, or a different AI tool, so it can’t see your original:

Translate this Spanish text into English as literally as possible. Don't smooth it out.

[paste the translation]

Put the back-translation next to your original. Same dates? Same numbers? Same request? Same level of politeness? If something changed, ask the first chat to fix that sentence and check again.

Part 2: Save your writing style (10 minutes, once)

If you rewrite a lot, you’ll get tired of saying “keep it in my voice.” Claude can save a custom style built from your own writing, and you pick it whenever you start a rewrite. Think of it as giving the intern a one-page guide to how you write.

Step 1. Collect two or three good samples

Pick things you wrote yourself and liked: an update, an email, a short report. Same kind of writing you usually rewrite.

Step 2. Create the style

Below the message box, open the tools menu and choose Use style, then Create & edit styles. Click Create custom style, add your samples as a writing example (upload a file or paste text), and create the style. Claude analyzes the samples and writes a style description.

Step 3. Read and fix the description

Open the style and read what Claude wrote about your writing. Correct anything that’s wrong, and add the rules from this tutorial:

When rewriting my text: never change numbers, dates, names, or how strong a commitment is. Don't add facts. After every rewrite, list what changed.

Step 4. Use it

Start a new chat, choose your style from Use style, and paste your text with Step 2’s brief. You’ll still check the changes (Steps 3 and 4). The style makes the draft sound like you; it doesn’t make it right.

Why you can’t skip the checking

A rewrite that reads better can still say something different. Researchers have measured that:

• Even grammar fixes can change meaning. In a 2026 study, researchers gave AI models essays written before ChatGPT existed and asked for revisions, including grammar-only edits. The edits still significantly changed what the essays meant. In a separate user study, heavy AI use led to nearly 70% more essays that took no clear position on the question (Abdulhai et al., “How LLMs Distort Our Written Language,” 2026). This is a preprint, and the 70% figure comes from one user study.
• Translation quality depends on the language. In a study of hospital discharge instructions, AI translation into Spanish and Portuguese was about as accurate as professional translation, but less accurate for Haitian Creole (American Academy of Pediatrics, 2024).

Neither problem shows up when you read only the new version. That’s why this tutorial makes you write down what must not change (Step 1), makes the AI show its changes (Step 3), and checks translations by translating them back with a fresh chat (Step 6).

Checklist

Before you use the rewrite or translation:

• Every item on my “must not change” list is still there, unchanged
• No commitment got stronger or weaker (“aim to” vs “will”)
• Nothing was added that I didn’t write: reasons, benefits, promises
• No caveat or condition disappeared
• I read the change table, not just the final text
• For translations: the back-translation in a fresh chat matches the original’s dates, numbers, request and politeness
• Anything legal, medical, safety-related or contractual goes to a professional translator

What you just learned

On the surface you edited a paragraph. Underneath, you practiced checking in its most precise form: comparing two versions against a list you wrote before the AI touched anything. You also practiced giving instructions precisely: naming the smallest edit that solves the problem.

The same method works for: turning meeting notes into a clean update, softening feedback before you send it, making a technical explanation readable for customers, adapting one announcement for three audiences, and fixing the English in a document written by a non-native colleague (with their permission). Change the brief in Step 2; keep the “must not change” list and the change table.

Using a different tool?

The prompts work anywhere. Only the buttons change.

Features and plan names change often. This table was checked in October 2026; the ChatGPT, Gemini and Copilot rows still need checking against their help pages.

Sources

• How LLMs Distort Our Written Language (Abdulhai et al., arXiv, 2026)
• Research evaluates machine translation platforms for pediatric discharge instructions (American Academy of Pediatrics, 2024)
• Configure and use styles (Claude Help Center)