Skip to main content

Give your agents a brand voice

Make AI Content Agent output sound like your brand using workspace-level and per-agent instructions, with a worked example you can copy.

Written by David Baum

Voice is the biggest quality lever you have over agent output. The same agent, with the same tools, can produce copy that reads like your brand or copy that reads like generic marketing. The difference is the guidance you give it. This article shows you where to set that guidance and how to write it so the output sounds right the first time.

Where voice lives

You set agent instructions in two places, and each has a job.

Workspace-level instructions apply to every agent working in that workspace. Every agent inherits this shared guidance, so it is the right home for your brand voice, tone rules, dos and don'ts, and formatting preferences. Set it once and all your content picks it up. To find it, open the workspace settings and go to the agent instructions.

Per-agent instructions apply to one agent and are for job-specific behavior. Use them when a single agent needs guidance that should not spread to the rest of your content, for example a different structure for a specific content type.

A simple rule: put what should be true across all your content at the workspace level, and put one agent's task-specific behavior on that agent.

Write a brand voice guide that works

Vague guidance produces vague output. Concrete guidance steers the agent. Cover four things.

  1. State the tone. Say plainly how your brand should sound, for example direct, warm, and plain-spoken.

  2. List banned words or phrases. Name the words and phrases you never want to see, so the agent avoids them.

  3. Give a short do and don't list. A few clear rules on each side do more than a paragraph of description.

  4. Include one or two example sentences. Show a sentence that sounds right. Examples anchor the agent faster than adjectives.

A worked example

Here is a compact brand voice block you can adapt. Keep it short and concrete.

Tone: direct, warm, plain-spoken. Write in second person.

Banned words: leverage, seamless, robust, game-changer, revolutionary.

Do: lead with the concrete benefit, use short sentences, name the reader's problem.

Don't: open with a generic claim, stack adjectives, use exclamation marks.

Sounds right: "You can set your brand voice once and every agent picks it up."

Test and refine

Guidance improves with one pass of feedback. Run a single project, read the output closely, and note where it drifts from your voice. Then tighten the guidance: add the word you saw and did not want, sharpen a rule that was too soft, or add one more example sentence. Run another project and compare. A short, specific guide that you refined against real output beats a long one you wrote from scratch.

Tips

  • Keep the workspace guide short. A page the agent applies well beats three pages it applies loosely.

  • Prefer examples over adjectives. One sentence that sounds right teaches more than five describing words.

  • Put brand voice at the workspace level so you maintain it in one place.

  • Revisit the guide after a batch of output, not after every sentence.

FAQ

Which wins, workspace or per-agent instructions?

Workspace instructions set the shared baseline every agent inherits. Per-agent instructions add job-specific guidance for one agent. Use the workspace level for what should hold across all your content, and per-agent instructions for one agent's task.

Will brand voice instructions change the tools an agent has?

No. These instructions steer how the agent writes. They do not change the tools it has.

Related articles

  • Build your first AI Content Agent

  • What are AI Content Agents

  • Content pillars, content types, and funnel stages

If you need help, message us in the Relato Slack community or email [email protected].

Did this answer your question?