How to Create a Voice & Tone Guide With AI in a Single Afternoon

[tldr]

TL;DR: How to get on-brand AI content fast

1. Start with 5 real examples of your best content—internal or external.
2. Give the model a format to follow (a simple voice & tone template works).
3. Build in hard constraints (length, banned phrases, structure).
4. Iterate in small steps, not one giant prompt.
5. Use the finished guide as the required input for every AI-generated asset.

[/tldr]

Creating content is easy with AI. Creating content that’s consistently high-quality and on brand? Still a tall order. 

In fact, 77% of companies deal with off-brand content—and it’s especially true in B2B SaaS. While they may be pumping out more articles than ever, consistency is down.

The more teams experiment with gen AI, the more they realize that brand voice can’t be an afterthought. It has to be operationalized. Voice and tone guides need to be considered part of the infrastructure.

So if you want to scale your content with AI, you need to teach it what “on-brand” actually means—in a way it can effectively apply across every piece of content. Remember: even one suspiciously off-brand piece can cause damage to your brand.

The good news: building that system can be done in a few hours, if you approach it the right way (and follow the upcoming guidelines).

[tldr]

Watch the full webinar

This piece is based on a session with Seema Kumar, GM, Marketing at Whispered, and Lauren Andrasco, VP, Marketing Operations & Performance Management at OutSystems, where they shared practical ways to operationalize AI in B2B marketing.

Watch the full webinar to get all the insights.

[/tldr]

Step 1: Reverse-engineer what already works

Most teams treat voice and tone as something they need to invent, if they haven’t done so formally. In reality, it’s usually embedded in the content you’ve already shipped.

Start by collecting five-ish pieces of content that reflect your desired voice. These don’t need to be polished marketing assets. Things like founder emails, product launch notes, or customer communications can represent your voice surprisingly well. 

Then, instead of asking an LLM to “write a voice and tone guide,” give it two inputs:

1. The five example assets
2. A simple template for what a voice & tone guide should look like (1–2 pages, structured, practical)

Then, ask the model to reverse-engineer the rules that govern the writing in your samples.

“Within two hours, start to finish, including time to get approval, I had a voice and tone guide done.” Seema Kumar, GM, Marketing, Whispered

Step 2: Stop chasing the “perfect prompt”

It may be tempting to build the perfect all-in-one prompt: outline the guide, define principles, add examples, suggest vocabulary, write do/don’t lists, and polish language.

But it doesn’t really work because, as it turns out, both humans and AI struggle with overly complex asks.

“The more that you try to condense into one step, the lower quality your result is going to be.” Seema Kumar, GM, Marketing, Whispered

Instead, treat the process like a recipe. For example:

• Step 1: Extract tone principles.
• Step 2: Refine each principle with examples.
• Step 3: Add structural guidance by asset type.
  Step 4: Insert vocabulary rules and banned phrases.
• Step 5: Tighten length and clarity.
  

These small steps allow you to incorporate clear feedback loops and human review between each stage. And by forcing you to articulate what actually makes your voice your voice, they improve output quality.

Step 3: Add hard constraints (this is where quality jumps)

Obviously, to get the results you want, you need to tell AI what you want it to do. But just as importantly, you need to explicitly tell it what not to do, and many teams miss that.

Examples of useful constraints in a voice and tone guide include:

• Maximum sentence length range
• Banned phrases (e.g., “seamless,” “game-changing,” “unlock”)
• Explicit rules about hedging language
• Clear stance on humor (allowed? sparingly? no dad jokes? only dad jokes?)
• Formatting guidelines (short paragraphs, bold subheads, etc.)

Constraints narrow the solution space, ultimately raising quality and consistency. So start telling AI “no.”

Step 4: Make it practical, not poetic

Many voice and tone guides fail because they’re too abstract. “Be bold. Be clear. Be human.” Those principles are fine, but they won’t make your brand voice special.

To make your guide more actionable and useful (for both humans and AI), include:

1. Voice principles with rewrites

Instead of:

• “Confident but not arrogant”

Add:

• A weak example
• A strong example
• A rewrite showing the difference

2. Vocabulary dos and don’ts

Clarify:

• Words you always use
• Words you never use
• Jargon you deliberately avoid
  

3. Asset-specific direction

Get more granular, and specify:

• How blog posts open
• How landing pages structure proof
• How emails balance brevity and persuasion (combine this with word count constraints mentioned in the previous section, for example)

4. A brand POV

If your company believes something specific about the market, state it clearly. AI can’t infer your point of view unless you explicitly provide it.

This is especially important because generative AI is optimized for plausible, average language. It defaults toward neutrality. Your job is to encode sharpness.

Step 5: Operationalize the guide across workflows

You’ve done most of the hard work. But the beautiful guide you’ve put together can only shine when it’s embedded in all your systems.

That means:

• Uploading it into your AI “projects” or custom GPT environments.
• Requiring it as context for blog drafts, landing pages, and email copy.
• Incorporating it into content grading rubrics.
• Making it a checkpoint before publishing.

The broader opportunity is significant: McKinsey has estimated that generative AI could drive productivity improvements equivalent to 5-15% of total marketing spend, depending on application. But, of course, those gains assume outputs are usable. If every draft requires heavy editing, the efficiency evaporates. That’s where operational excellence comes in.

“And once you have that infrastructure in place, you can go further. Campaign orchestration systems, for example, can draft initial email copy based on persona and campaign intent.” — Lauren Andrasco, VP, Marketing Operations & Performance Management, OutSystems

A one-week implementation plan

If you want to put this into practice starting next week:

Day 1: Collect five “gold standard” content examples.
Day 2: Generate v1 of the guide using an LLM and a simple template.
Day 3: Add constraints and rewrite examples.
Day 4: Review with 2–3 stakeholders (founder, brand lead, PMM).
Day 5: Upload the finalized guide into your AI tools and make it mandatory context.

This is assuming limited bandwidth. If you’ve got some free time, you can knock this out in an afternoon (hence the headline). 

Then, to measure success, start tracking:

• Reduction in edit cycles
• Faster approvals
• Fewer “this doesn’t sound like us” comments

[tldr]

Need help operationalizing AI?

AI can undoubtedly enable scale, but only when used the right way. Need help build the voice, workflows, and guardrails that let AI accelerate your marketing without creating brand drift? Contact Right Side Up today!

[/tldr]

‍