Teach AI Your Brand Voice with Examples You Can Test

“Sound friendly and authoritative” is too vague to guide an AI writing workflow. A usable voice brief shows what good writing looks like, explains why it works and identifies language that would mislead your readers.

Build an example set with context

Choose a few approved passages serving different purposes: an explanation, an error message, a product comparison and a call to action. Include the audience and the action each passage supports.

Do not select examples solely because a founder likes the wording. A joke that works in a newsletter may be inappropriate in a payment failure message. Voice needs to adapt to the situation while keeping consistent principles.

Convert taste into observable rules

Add a short explanation to each example: “This paragraph names the user's problem before introducing the feature.” That helps the model reproduce a decision rather than copy a surface mannerism.

Worked example: removing a false promise

Suppose a hypothetical draft says, “Unlock effortless SEO success with our revolutionary dashboard.” A stronger version is, “See which pages lost clicks, then open the affected queries to investigate.”

The second sentence is useful only if the dashboard actually supports those actions. Brand voice review cannot substitute for feature verification. Keep the factual acceptance check separate from tone.

Test the brief before adopting it

Prepare five unseen tasks: a feature explanation, a limitation, an onboarding step, a comparison and a correction. Generate drafts using the same factual inputs. Have reviewers assess clarity, accuracy, audience fit and unnecessary verbosity without knowing which prompt produced each draft.

Record disagreements. If reviewers cannot agree whether a passage follows the brief, the rule may be too subjective. Add an example or clarify the audience rather than expanding the brief into pages of adjectives.

Maintain a small living guide

Save the brief version with each evaluation. Add examples when a recurring failure appears; remove rules that duplicate each other or conflict. Do not rewrite the guide after one unusual output.

A good voice system makes accurate content easier to recognize and use. It should not make every article sound identical, or hide uncertainty behind confident phrasing. The reader's task remains more important than a distinctive verbal flourish.

Separate voice from facts in the prompt itself

The most common failure in brand-voice work is not tone. It is that the voice brief and the factual inputs arrive in the same block of text, so the model treats the brief's confident examples as facts and the facts as things to be styled. Keep them physically separate:

SECTION A — Facts you may use (treat as the only source of product claims):
- Feature: exports the current filtered view as CSV
- Limit: exports are capped; the cap is not documented, say so
- Plan: available on all plans

SECTION B — Voice rules (apply to wording only; never add product claims):
- Name the reader's problem before the feature
- State supported conclusions directly; qualify anything from Section A marked uncertain
- No superlatives, no promises about results

Task: write the feature paragraph using only Section A facts and Section B rules.
Flag any sentence where a rule in B would require a claim not present in A.

The final instruction is the important one. When a voice rule such as "be confident" collides with an uncertain fact, the model should surface the collision instead of resolving it by inventing certainty. Reviewers then see exactly where the brief and the evidence pull in different directions.

What a voice evaluation set looks like

Five unseen tasks are enough to start, but the tasks need to be chosen so that a generic draft fails at least some of them. A set that has worked in practice covers:

Score each output on three separate scales: does it follow the voice rules, is every product claim traceable to the supplied facts, and would the intended reader be able to act on it. Keep the scales apart. A draft can be perfectly on-voice and factually wrong, and mixing the scores hides that.

Signs the brief is doing harm

A voice guide can make content worse while every draft technically passes. Watch for these patterns across a month of output:

• Every article opens the same way. A rule such as "start with the reader's problem" has become a template sentence rather than a decision.
• Limitations and caveats have become shorter or disappeared. "Confident" has quietly overridden "accurate".
• Reviewers are correcting the same tone issue repeatedly, which means the brief describes it but the example does not demonstrate it.
• Drafts sound identical across audiences that need different registers, such as developer documentation and marketing pages.

When any of these appear, fix the brief by adding or replacing an example, not by adding another adjective. Models reproduce examples far more reliably than they interpret abstractions, and a good example also shows human writers what is meant.

Put this into practice

Copy the worksheet columns below into a spreadsheet and keep one row per item you check. The filled row is an illustrative example, not a reported customer result; replace it with your own verified records.

Use the following prompt only after supplying the records it requests:

Rewrite using the approved examples and voice rules. Keep every factual limitation. Return the revision plus a short explanation of which rules changed the wording. Flag any unsupported feature claim instead of polishing it.

Research context

Reusable instructions need examples and review criteria, not just a list of tone adjectives. The related Ahrefs starting points are Claude Skills for SEO and Marketing: What They Are and How to Use Them and Every Marketer Says You Need “Taste”. Here’s What That Actually Means. This guide’s checklist, examples and proposed workflow are independently written; they are not results of a SEOVision experiment.

Continue with the next task

• How to Check an AI Draft Before Publishing
• A Human–AI Content Workflow That Protects Quality and Trust
• AI Content Refresh: Update Evidence, Not Just the Date

Sources

• Claude Skills for SEO and Marketing: What They Are and How to Use Them — Research starting point; not an endorsement of this original workflow
• Every Marketer Says You Need “Taste”. Here’s What That Actually Means — Research starting point; not an endorsement of this original workflow