What is a brand voice guide for Claude?

Quick Answer: A brand voice guide for Claude is a structured document you upload to Claude Projects that tells the AI your tone, audience, writing rules, and banned phrases. When built correctly, it eliminates the need to re-explain your brand every session and produces consistent, on-brand copy at scale.

B2B SaaS content teams waste hours re-briefing Claude on the same brand basics, session after session. The fix is a single, well-structured brand voice guide that lives inside Claude Projects and travels with every conversation. This guide shows you exactly how to build one: what to include, how to structure it, which examples to use, and how to test whether it's actually working.

What Is a Brand Voice Guide for Claude?

A brand voice guide for Claude is a reference document, typically written in Markdown, that you upload to a Claude Project as a knowledge file. It gives Claude the context it needs to write in your voice without you repeating yourself.

Unlike a general brand style guide built for human writers, this document is written specifically for how Claude processes instructions. That means precise, structured, example-led content rather than vague descriptors like "professional but friendly."

When the guide is well-built, Claude applies your tone, vocabulary, and formatting rules across every output in that project, whether that's a product page, a cold email sequence, or a LinkedIn post. If you need outside help refining that system, it can also be worth reviewing specialist B2B SaaS content marketing agencies that build repeatable editorial workflows.

Why Claude Projects Is the Right Place for This

Claude Projects lets you attach persistent files and instructions to a conversation workspace. Every new chat inside that project inherits the context from those files automatically.

This matters for B2B SaaS teams because:

• Content volume is high. You need consistency across dozens of assets, not just one-off prompts.
• Multiple contributors use the same tool. A shared project means every team member works from the same voice rules.
• Claude reads the file fresh each session. There is no "memory drift" the way there can be with long running chat threads.

The alternative, pasting brand context into every prompt, produces inconsistent results and burns time. A Claude Project with a proper voice guide is the production-ready setup.

How to Structure Your Brand Voice Guide for Claude

The structure below is the one that produces the most reliable outputs. Each section serves a specific purpose, and the order matters.

Section 1: Brand Overview (50-100 words)

Open with a short, factual summary of who the brand is and who it serves. Claude uses this as a grounding reference for every output.

Example:

SaaS Hackers is a content and growth consultancy for B2B SaaS founders and marketing teams. We help early-to-mid stage companies build content engines that drive pipeline, not just traffic. Our audience is commercially-minded: they want tactics, not theory.

Keep this tight. One paragraph. No mission statement fluff.

Section 2: Tone Descriptors with Examples

Tone descriptors only work when paired with examples. "Direct and confident" means nothing to Claude. Showing Claude what direct and confident looks like in a sentence does.

Use this format for each descriptor:

Tone quality: Direct What it means: We say what we mean without hedging. No qualifiers, no soft language. Do this: "This approach cuts CAC by 30%." Not this: "This approach could potentially help reduce CAC in some scenarios."

Aim for 4-6 tone descriptors. Common ones for B2B SaaS brands include:

• Direct vs. hedged
• Specific vs. vague
• Peer-to-peer vs. formal authority
• Dry wit vs. enthusiastic
• Opinionated vs. neutral

Section 3: Audience Profile

Tell Claude exactly who it is writing for. Not a persona template, a real description of the reader's context, goals, and frustrations.

Example:

Primary reader: B2B SaaS marketing managers and founders at companies with 10-150 employees. They are time-poor, sceptical of hype, and have been burned by agencies that delivered traffic without revenue. They respond to specificity, social proof, and clear next steps. They do not need to be educated on what SaaS is.

This section stops Claude from over-explaining basics or writing in a way that talks down to experienced readers.

Section 4: Banned Phrases and Words

This is the section most brand voice guides skip, and it is the most important one for Claude.

List every word and phrase you do not want Claude to use. Be specific. Group them by type.

Overused AI words to avoid: leverage, utilise, streamline, robust, seamless, holistic, game-changer, transformative, cutting-edge, empower, unlock, elevate, ecosystem, synergy, scalable, actionable, dynamic, pivotal, crucial

Filler phrases to avoid: "it is worth noting", "it goes without saying", "at the end of the day", "in today's world", "now more than ever", "in order to", "as we all know", "needless to say", "last but not least"

Structural patterns to avoid:

• Em dashes used as parenthetical asides
• Sentences that open with "Ultimately," or "Essentially,"
• Three-part lists ending with "and more"
• "Not only X but also Y" constructions

Claude follows these rules reliably when they are written as explicit prohibitions rather than style suggestions.

Section 5: Formatting Rules

Specify your default formatting preferences so Claude does not make decisions you will have to undo.

Cover:

• Paragraph length: "Maximum 3 sentences per paragraph."
• Headers: "Use H2s for major sections, H3s for sub-points. No H4s."
• Lists: "Use bullet points for features and steps. Use numbered lists for sequences only."
• Bold text: "Bold key terms and takeaways only. Not decorative."
• Sentence length: "Vary sentence length. Avoid three consecutive long sentences."

Section 6: Voice Tests

A voice test is a short passage written in your brand voice that Claude can use as a calibration reference. Include 2-3 examples.

These work best when they cover different content types: one editorial paragraph, one product description, one short social post.

Example editorial paragraph:

Most B2B SaaS blogs are written for Google, not for buyers. They rank, they get traffic, and they convert nobody. The fix is not better SEO. It is writing that treats the reader as someone who already understands the problem and wants to know what to do about it.

Example social post:

Your content strategy does not need more volume. It needs better targeting. One article that speaks directly to your ICP outperforms ten that speak to everyone.

When Claude has these examples, it can pattern-match your voice rather than interpret abstract descriptors. Teams pairing AI writing systems with search performance work often also document adjacent rules in the same workflow, similar to how B2B SaaS SEO agencies standardise briefs and editorial constraints.

Section 7: What This Brand Does Not Do

A short list of positioning guardrails stops Claude from generating content that is off-brand or off-strategy.

Example:

• We do not write thought leadership that avoids taking a position
• We do not use case studies as proof of concept without specific metrics
• We do not write content that targets keywords without a clear buyer intent
• We do not use corporate language or write as though addressing a committee

How to Upload Your Guide to Claude Projects

1. Go to Claude.ai and open or create a Project for your brand or content work
2. Navigate to the Project Knowledge section
3. Upload your brand voice guide as a Markdown (.md) file or paste the content directly
4. Add a brief instruction in the Project Instructions field: "Always refer to the brand voice guide before writing any content. Follow the tone, formatting, and banned phrase rules exactly."
5. Test with a sample prompt before using the project for live work

The Project Instructions field is processed before every conversation, so this is where you anchor Claude's behaviour at the session level.

How to Test Whether Your Brand Voice Guide Is Working

Do not assume the guide works because you uploaded it. Run these three tests before using the project for real content.

Test 1: The banned phrase check Ask Claude to write a 200-word product description. Scan the output for any banned words or phrases. If they appear, the guide needs stronger prohibition language ("Never use the word X under any circumstances").

Test 2: The tone comparison Ask Claude to write two versions of the same paragraph: one in your brand voice and one in a generic B2B SaaS tone. If you cannot tell the difference, your tone descriptors need more examples.

Test 3: The audience calibration check Ask Claude to write an intro paragraph for a technical topic in your space. Check whether it explains basics your audience already knows. If it does, your audience profile needs to be more explicit about expertise level.

Run these tests each time you update the guide. Treat the voice guide as a living document, not a one-time setup. If your team is managing this across channels and operators, support from B2B SaaS digital strategy agencies or B2B SaaS fractional CMOs can help tighten governance.

Common Mistakes That Break Brand Voice Guides for Claude

Using adjectives without examples. "Confident and approachable" is not an instruction. A sample sentence is.

Writing the guide for human readers. Brand guidelines built for designers or agencies use language Claude does not weight the same way. Write directly for how Claude processes instructions: specific, structured, example-led.

Skipping the banned phrases section. This is the fastest way to get Claude outputs that sound like every other AI-generated article. Explicit prohibitions are more effective than positive style guidance alone.

Treating it as a one-time document. Your brand evolves. A voice guide that was accurate six months ago may not reflect your current positioning. Review it quarterly.

Making it too long. A guide over 1,500 words risks diluting the most important instructions. Keep it tight. Every section should earn its place.

FAQs

What should a brand voice guide for Claude include? A brand voice guide for Claude should include a brand overview, tone descriptors with examples, an audience profile, a banned phrases list, formatting rules, voice test examples, and positioning guardrails. The most important section is the banned phrases list, as explicit prohibitions produce more consistent outputs than style suggestions alone.

How do I stop Claude from sounding generic? The fastest fix is adding a banned phrases section to your brand voice guide. List every overused AI word and filler phrase you want Claude to avoid. Pair this with 2-3 voice test examples written in your actual brand tone so Claude has a pattern to match rather than abstract descriptors to interpret.

Where should I store my brand voice guide in Claude? Store your brand voice guide in Claude Projects as a knowledge file. This makes it available automatically in every conversation within that project, so you do not need to paste it into each new prompt. Add a brief instruction in the Project Instructions field telling Claude to apply the guide to all content it produces.

How long should a brand voice guide for Claude be? Aim for 800-1,200 words. Long enough to cover tone, audience, banned phrases, formatting rules, and voice examples. Short enough that the most important instructions are not buried. A guide that is too long risks Claude deprioritising key rules.

Can one Claude Project serve multiple content types? Yes, but only if your brand voice is consistent across formats. If your blog tone and your ad copy tone are significantly different, create separate projects with separate voice guides. Trying to handle both in one guide usually produces outputs that are too averaged to be useful for either format.