The Brand AI Profile

How do I teach m18t's AI my brand voice? Learn the four Brand AI Profile fields, why Positioning grounds every prompt, and how blank fields are skipped.

The Brand AI Profile is how you teach m18t's AI what your brand is and how it sounds — once, in one place, so every AI feature speaks consistently. It's the steering wheel. It's also the reason you rarely need to touch a prompt template.

What you'll learn

  • The four fields of a Brand AI Profile and what each one does.
  • Why Positioning grounds every prompt while Voice, Rules, and Glossary only shape writing.
  • How blank fields behave (they're skipped, not broken) and how to fill them in.

The concept: steer the brand, not the prompt

m18t's AI features run on shared prompt templates that are tuned by the platform and Locked by default. If those templates were the only place to control output, every brand would sound the same — or you'd be forced to fork a template for each brand and maintain them by hand.

The Brand AI Profile solves that. It's a small set of brand-specific fields you fill in once. At run time, m18t composes a short brand preamble from those fields and prepends it to the System Prompt of the relevant feature — automatically, on every call. The shared template stays generic; the per-brand grounding comes from your profile. Switch brands and the next brand's profile is what gets injected instead.

This is why the prompt editor keeps pointing you back here. When you open a Locked platform prompt, m18t tells you outright: to change how the AI writes for this brand, edit its Brand AI Profile, not the template.

The four fields

A Brand AI Profile has four fields. Each one is a free-text box you fill in plain language.

Positioning & identity

Who the brand is, who it serves, what makes it distinct.

This is the grounding field. It is injected into every prompt — both writing prompts and scoring prompts — because the model always needs to know whose brand it's working for. A relevance scorer needs to know what your brand cares about as much as a copywriter needs to know who it's writing for. Keep this factual and specific: what you do, who you serve, what sets you apart.

Voice & tone

How the brand sounds — formality, warmth, energy.

This shapes how generated text reads. It is injected into writing (generative) prompts only. It is deliberately kept out of evaluative prompts like the news relevance scorer, because a yes/no decision shouldn't be coloured by how chatty or formal your brand is — that would skew the judgement, not improve it.

Rules — Dos & Don'ts

Hard guidelines: claims to avoid, words to never use, things to always include.

Also a writing-only field. Use it for non-negotiables: medical claims you can't make, a competitor you never name, a disclaimer you always include. These are constraints on generated content, so they belong with the other generative fields.

Glossary — preferred terms

Brand-specific terminology and the exact wording to use (and what not to call things).

Writing-only as well. If your brand calls a thing a "session" and never a "meeting," or capitalizes a product name a specific way, put it here. It nudges generated copy toward your house vocabulary.

Why the split matters

The division is intentional and worth internalizing:

  • Positioning → injected always (it's identity; everyone needs it).
  • Voice, Rules, Glossary → injected into generative prompts only (they shape style and content, which has no place in a scoring or evaluation task).

So when an AI step is writing for you — a news draft, a caption, a content rewrite — it gets all four fields. When an AI step is judging for you — scoring a news article's relevance — it gets only Positioning. Your brand's tone never leaks into a relevance decision.

Blank fields are skipped, not broken

You don't have to fill in all four to start. If a field is empty, m18t leaves it out of the preamble — there's no dangling "Voice:" label, no error, no degraded output. The brand preamble only ever contains the fields you've actually written. Fill in what you have; add the rest later.

This also flows through to the injection-token badges in the prompt editor: a brand-profile token shows green when the active brand has that field filled and amber when it's empty (meaning the token will be ignored for this brand until you fill it).

How to fill in a Brand AI Profile

  1. Open Configuration ▸ Brands (/studio/config/brands) and select the brand you want to tune.
  2. Find the brand's AI Profile section.
  3. Fill in Positioning & identity first — it grounds everything, including evaluation. Be concrete: what the brand does, who it's for, what's distinctive.
  4. Add Voice & tone, Rules — Dos & Don'ts, and Glossary — preferred terms as you have them. Each is optional; blanks are skipped.
  5. Save. The next AI run for that brand will pick up the changes — no template editing required.

To check your work, open any AI prompt for that brand at AI Prompts and look at the token badges: green means the field is set and will be injected; amber means it's still empty.

Variations

  • Per brand, never shared. Profile values are strictly per-brand. There's no cross-brand sharing of Voice or Rules — consistent with m18t's brand-scoping throughout. Each brand maintains its own.
  • Use it before you reach for a template. Most tuning belongs here. Only customize a prompt when a brand needs a genuinely different task structure, not just a different tone.
  • Extensible by design. The field set is defined at the platform level; today it's the four above. They show up everywhere the same way — settings form, prompt-editor badges, and the injected preamble.

FAQ

I edited my brand's voice but the AI output didn't change. Why? Voice, Rules, and Glossary are only injected into writing prompts. If you were looking at an evaluation step (like news relevance scoring), those fields are intentionally excluded — only Positioning is used there. For writing features, re-run after saving the profile.

Do I have to fill in all four fields? No. Blank fields are skipped cleanly. Start with Positioning (it grounds every prompt) and add the others over time.

Why is Positioning treated differently from the other three? Positioning is identity — every prompt needs to know whose brand it is, including scoring prompts. Voice, Rules, and Glossary are about how to write, which only matters when the AI is generating text, so they're held back from evaluation tasks.

Is the profile shared across my brands? No. Values are per-brand. Each brand's profile is injected only when that brand is active.

Where do I see whether a profile field is actually being used? In the AI Prompt editor, brand-profile tokens show green (filled, will inject) or amber (empty, ignored) for the selected brand. See Injection Tokens.

What's next

Was this article helpful?