Prompt Recipes: Make AI Text Casual, Formal, or Friendly

AI models default to a middle register: not quite formal, not quite casual, and not quite anything in particular. The prose comes out neutral in the way that waiting-room music is neutral. It is technically inoffensive and practically useless if you need something that matches your brand, your audience, or your own voice.

The good news: tone is one of the easier things to steer with a prompt. You do not need to rewrite the output by hand every time. You need a clear instruction, a concrete example of the register you want, and occasionally a list of what to avoid. This guide gives you three ready-to-paste prompt recipes, a before/after to show what the shift actually looks like, and enough context to adapt these for whatever job you are working on.

Why Tone Shifts Are Harder Than They Look

When you ask a model to "make this more casual," it usually does one of two things: it sprinkles in contractions and calls it a day, or it overcorrects into a tone that reads like a brand trying too hard to sound relatable. Neither is what you want.

The underlying problem is that "casual," "formal," and "friendly" are not single points on a dial. Each register has its own set of signals: sentence length, vocabulary range, punctuation choices, how often the writer uses the second person, whether the piece acknowledges the reader's situation. A good prompt to change ai writing tone has to specify at least a few of those signals, not just the label.

It also helps to give the model something to push against. Telling it what to avoid (jargon, hedging phrases, exclamation points) often works better than telling it what to add, because avoidance is easier to verify.

The Casual Register: Copy-Paste Prompt

Use this when you are writing for a consumer audience, a personal blog, a social post, or any context where a stiff tone would feel out of place.

Rewrite the following copy in a casual, conversational register. Write as if you are explaining this to a friend who is smart but not an expert. Use short sentences. Contractions are fine. Avoid jargon, hedging phrases like "it is worth noting," and words that only appear in formal reports. Keep every sentence under 25 words where possible. Do not add exclamation points or phrases that sound like a hype person. Here is the copy:

[PASTE YOUR TEXT HERE]

This is a reliable casual writing prompt ai that moves the needle without tipping into pep-rally territory. The friend-who-is-smart framing anchors vocabulary and assumed knowledge level. The 25-word sentence cap keeps rhythm tight. The explicit ban on exclamation points prevents the most common overcorrection.

The Formal Register: Copy-Paste Prompt

Use this for legal disclaimers, corporate communications, white papers, grant applications, or anywhere the reader expects institutional language.

Rewrite the following copy in a formal, professional register suitable for a business or academic audience. Use complete sentences. Avoid contractions, slang, and colloquialisms. Prefer precise vocabulary over plain vocabulary where the meaning is clearer with precision. Maintain an objective third-person perspective. Do not use rhetorical questions. Here is the copy:

[PASTE YOUR TEXT HERE]

The key instruction here is "prefer precise vocabulary where the meaning is clearer." Without that qualifier, models often swap every plain word for a longer one, which reads as pompous rather than formal. Precision is not the same as complexity.

The Friendly Register: Copy-Paste Prompt

Friendly sits between casual and formal. It is warm without being sloppy, approachable without abandoning professionalism. Think customer support emails, onboarding guides, or a brand that wants to feel human without sacrificing credibility.

Rewrite the following copy in a friendly, warm register. Acknowledge the reader's situation where it is natural to do so. Use plain language, but write in complete, well-formed sentences. Contractions are fine. Avoid corporate buzzwords, filler phrases, and passive constructions. The tone should feel like it comes from a helpful colleague, not a chatbot or a salesperson. Here is the copy:

[PASTE YOUR TEXT HERE]

The "helpful colleague, not a chatbot or a salesperson" framing is worth keeping. It gives the model two failure modes to avoid and tends to produce output that is genuinely usable rather than technically correct but hollow.

Before and After: Seeing the Difference

Here is the same paragraph run through each prompt recipe.

Source (AI default output):

It is worth noting that proper hydration is essential for maintaining cognitive function throughout the day. Individuals who consume adequate water are likely to experience improved concentration and reduced fatigue.

After the casual prompt:

Drink enough water and you will think more clearly. It sounds obvious, but most people are mildly dehydrated by mid-morning, and that is all it takes to make focusing harder.

After the formal prompt:

Adequate hydration is a prerequisite for sustained cognitive performance. Research consistently associates sufficient water intake with improved concentration and reduced physiological fatigue.

After the friendly prompt:

Staying hydrated makes a real difference to how well you can focus, especially in the afternoon slump. If you are finding it hard to concentrate, a glass of water is a reasonable first step before anything else.

Each version is accurate. Each is the same information. The register is the whole difference, and that difference determines whether a reader keeps going or bounces.

How to Adjust These Prompts for Your Specific Context

Three variables make the biggest difference when you adapt these prompts.

Audience specificity. The more you tell the model about who will read the output, the better it calibrates. "A 45-year-old homeowner who has never refinanced before" will produce different results than "an adult consumer." Add a sentence about your reader if the generic version is not landing.

Length and format constraints. Tone prompts do not automatically preserve your formatting. If you need bullet points to stay bullets, say so. If the output length matters, set a word count.

Examples of the target voice. If you have a paragraph of your own writing in the register you want, paste it in. One sentence of instruction plus one example of the target almost always outperforms a long list of instructions alone. This is the same principle behind a ChatGPT prompt that makes copy sound like you wrote it: give the model a reference point, not just a description.

For situations where the problem is not tone but specific AI-sounding phrases and patterns, you may want to combine a tone prompt with a cleanup pass. The guide on how to write a system prompt that strips out AI tells covers that layer separately, and pairing the two approaches tends to produce cleaner results than either one alone.

Where to Go From Here

These three recipes handle most everyday tone adjustments, but they are starting points. You will find that some audiences and some content types need tighter constraints. A legal audience might need explicit guidance on passive voice usage. A Gen Z consumer audience might need notes on vocabulary range that would look strange in a formal prompt.

When a single prompt is not getting you where you need to go, the fastest path is iteration with specific notes. Run the output, identify the three sentences that are most off-register, and add a constraint targeting those patterns. After two or three rounds you usually have a prompt that works reliably for your use case.

If you want a single, comprehensive starting point that handles human-sounding copy in general, the best prompt to make AI writing sound human covers the full set of adjustments in one place. You can also adjust ai tone as one layer within it rather than as a standalone pass.

For routine work, it is worth keeping a prompt file with the version of each recipe that works for your brand. Tone is one of the few things in AI editing where a good prompt, once built, mostly stays good.

---

Frequently Asked Questions

Can I use these prompts in any AI tool, not just ChatGPT?

Yes. The prompts are written in plain language and work in ChatGPT, Claude, Gemini, and most other text-generation tools. The results will vary slightly across models, but the core instructions are model-agnostic. If you are using a system prompt interface, you can paste the recipe directly there.

What if the model keeps reverting to its default tone?

This usually happens when the model has a long input with strong stylistic patterns of its own. Try breaking the task into smaller chunks, or place your tone instruction at the bottom of the prompt, right before the text to be rewritten. Models tend to weight instructions that appear closer to the material more heavily.

How is "friendly" different from "casual"?

Casual trades precision for ease: short sentences, contractions, informal vocabulary. Friendly keeps more precision and structure but adds warmth and reader acknowledgment. Friendly is still usable in professional contexts. Casual often is not. The boundary is blurry, but a good test is whether the output would fit in a business email. Friendly usually passes; casual usually does not.

Do I need to use a different change tone ChatGPT prompt for different content types?

In practice, the same recipe works across most content types with small adjustments. The bigger variable is audience familiarity with the subject matter. A casual prompt for technical content may need an added note about handling terminology, because plain language and accurate technical vocabulary sometimes pull in opposite directions.

What should I do if the rewritten copy loses important information?

Tone prompts occasionally cause the model to drop nuance or compress meaning in ways that matter. After rewriting, do a quick factual read-through against the original. Any place where the rewrite softens a number, removes a qualifier, or merges two separate points deserves a manual fix. Think of the prompt as handling about 80 percent of the work; the last 20 percent is still yours.