You ask AI to make a paragraph more engaging. The revised version flows well, but the understated observation you liked has become an enthusiastic promise. It sounds polished, but would your business say it that way?
Generative AI can help you explore wording and produce useful drafts. Deciding which suggestions belong to your brand calls for a clear understanding of your voice.
To maintain brand voice with AI, you need to be able to describe what makes your writing recognisable and assess how those qualities come through in the result. That involves choices about how you address readers, what you explain and where you let a detail speak for itself.
What Makes Your Brand Voice Recognisable?
Descriptions such as warm, confident or premium offer a starting point. However, they leave plenty of room for interpretation.
Two hotels might both describe their voice as welcoming. One could use quiet observations about the place and its surroundings. Another might address guests conversationally, with occasional humour. Both can be welcoming while sounding quite different.
Look at writing your business has already approved. Consider:
- Vocabulary: Do you favour everyday words, specialist terminology or evocative descriptions?
- Rhythm: Do sentences allow room for explanation, move briskly or vary according to the subject?
- Relationship with readers: Do you offer suggestions, ask questions or make direct invitations?
- Expression of confidence: Do you explain your strengths through detail or make more prominent assertions?
- Personality and emotional expression: Does your writing use humour, convey enthusiasm or take a more restrained approach? Where are those qualities appropriate?
These choices help turn an impression of your voice into something you can discuss and apply.
Voice also allows for changes in tone. A playful brand may use straightforward language when explaining a cancellation policy. Recognisability doesn’t require every piece of content to sound equally cheerful or informal.
Give AI Examples and Explain What Works
When prompting AI, include a few representative pieces of approved writing. Choose examples relevant to the task: a social media caption alone may offer limited guidance for a detailed service page.
Explain what you want the AI to take from them. For example:
Use these examples as a reference for our voice. Notice how they address the reader directly, explain product features precisely and use humour sparingly. Follow those qualities without repeating the examples’ phrases.
That last distinction matters. A memorable expression can become laboured if it appears in every new draft.
Examples can also show boundaries more clearly than adjectives. You might explain why a sentence feels too familiar, or why an enthusiastic rewrite loses the restraint you value.
Here, giving AI examples means supplying reference material with your instructions. It doesn’t mean technically training a new model.
Turn Your Voice Into a Practical Writing Brief
An AI brand voice brief should connect the qualities you want with writing decisions.
| Broad Preference | More Useful Direction |
| Welcoming | Address readers directly and explain what they need to know without assuming how they feel. |
| Confident | State confirmed strengths clearly and support them with relevant details. |
| Lightly humorous | Use an occasional understated observation grounded in the subject. Avoid forcing a joke into every section. |
| Understated | Let specific details convey the appeal; use praise selectively. |
Include the audience, purpose and publication context too.
For a hotel, “write a website introduction” leaves the placement open. A more precise instruction would be:
Write a 70–90-word introductory paragraph for the hotel’s homepage, to appear beneath the main headline. Address prospective guests directly and help them decide whether to explore the rooms.
The voice instructions then guide how that particular task should be expressed.
Example: Two Voice Briefs for the Same Hotel
To explore how written voice instructions affect a draft, I used two prompts for a fictional independent hotel. Both supplied the same facts and requested a 70–90-word website introduction in British English.
The fictional hotel was described as having 18 rooms in a coastal town, five minutes’ walk from the harbour. Some rooms overlooked the harbour; others faced the courtyard. Breakfast was served until 10.30 am, with bread from a local bakery. A lounge contained books and board games.
The prompts allowed selection and reordering of those facts but prohibited invented facilities, arrangements or claims about guests’ feelings.
The comparison used ChatGPT with GPT-6 on 5 November 2026, in separate new chats outside a Project. Account memory may still have influenced the responses. These are two illustrative outputs, rather than evidence of how consistently the model would respond across repeated attempts.
Voice One: Calm and Understated
The first brief requested measured sentences, concrete details and a welcoming manner without exaggerated promises. It also asked the model not to assume guests’ feelings or tell them how to spend their stay.
Unedited output:
Our independent hotel sits in a coastal town, just five minutes’ walk from the harbour. With 18 rooms, it offers a choice of outlooks: some rooms overlook the harbour, while others face the courtyard.
You can enjoy breakfast until 10.30 am, with bread supplied by a local bakery. There’s also a lounge with books and board games.
Whether you’re interested in a harbour view or would prefer a room facing the courtyard, take a look at our rooms to see what’s available.
The restraint is apparent. The description relies on supplied details and gives readers a clear invitation to explore the rooms.
However, much of it reads as a factual inventory. “It offers a choice of outlooks” is more formal than the surrounding wording, and the final paragraph repeats the harbour-versus-courtyard distinction.
An editor could preserve the measured approach while reconsidering that repetition and finding a more natural connection between the details. The instruction to be understated has produced restraint; whether it has produced a distinctive voice remains a separate question.
Voice Two: Conversational With Light Humour
The second brief requested contractions, direct address and one understated, good-natured observation grounded in the facts.
Unedited output:
Our independent hotel has 18 rooms in a coastal town, just five minutes’ walk from the harbour. Some rooms overlook the harbour itself, while others face the courtyard, so you can explore which suits you best.
Breakfast is served until 10.30 am, with bread from a local bakery. There’s also a lounge with books and board games, should you fancy spending some time there. And with breakfast running until half past ten, there’s no need to be quite so enthusiastic about the alarm clock.
“Should you fancy” gives the description a more conversational register. The alarm-clock observation introduces the requested humour through an existing detail.
That is a meaningful response to the voice brief. However, it still leaves decisions to make.
The closing sentence repeats the breakfast time, and the humour assumes that guests can leave their alarm later. That may suit leisure visitors, but it could feel less relevant to people travelling for work or leaving early.
There’s no universal verdict on the humorous statement. The business needs to judge whether it reflects how it wants to address its guests.
What the Comparison Shows
Changing the voice brief produced an observable difference in these outputs. It didn’t settle every question of audience fit or expression.
Both versions provide usable material. Both also repeat information. The second follows the request for humour, but deciding whether that humour belongs on the homepage requires knowledge of the brand and its audience.
The briefs therefore give the work direction. Editorial review assesses how well the choices serve it.
Review What a Revision Changes
When AI rewrites existing copy, compare the versions before accepting the smoother one.
Ask:
- Has a precise observation become a general promise?
- Has the relationship with the reader become more familiar than intended?
- Does the humour still feel natural in this context?
- Have useful explanations been replaced with promotional language?
- Which original phrases already express the voice well?
Requests such as make this more engaging can leave the desired change unclear. More focused feedback gives the revision a firmer purpose:
Keep the restrained tone and the reference to the local bakery. Remove the repeated room-view information and make the closing invitation more direct.
That instruction identifies what to preserve as well as what needs attention.
For the broader checks before publication, including product information and structure, see my article on editing AI-generated content.
Build Approved Decisions Into Future Work
Keep examples of finished copy alongside brief explanations of the choices behind them. Record useful feedback when a draft misses the voice.
“Too playful” identifies a concern. Explaining that a joke interrupts an otherwise calm description gives the next writer or editor more to work with.
Over time, this can make your brand voice guidelines more practical. They become a record of decisions people can understand and apply, with room for different formats and situations.
An experienced editor can help identify those patterns, assess AI-generated alternatives and refine the wording without making every piece sound alike. The same understanding also supports decisions when your voice needs to carry into another language, as explored in What Makes a Good Marketing Translation?
If you’re using generative AI for English marketing content and want help keeping the writing recognisably yours, get in touch with a draft and examples of your preferred voice.
