Brand Atlas AI as a concept

Brand Atlas AI is the umbrella term for the AI features inside the portal. It refers to Oswald, Henry, and any future AI assistant Brand Atlas builds. The naming is deliberate. The features are presented as properties of Brand Atlas, not as access to a third-party language model.

​

The naming choice

The product copy never says “ChatGPT,” “Claude,” “Gemini,” “GPT-4,” or any other provider name. The features are Oswald and Henry. The category is Brand Atlas AI. The underlying models are named only in the AI Usage & Disclosure Policy and the BYOK Policy, where transparency requires it. Three reasons:

1. The features are products in their own right. Oswald is not “Claude inside Brand Atlas.” It is an editorial assistant that uses Claude as part of how it works. The system prompt, the brand-record grounding, the tool-use loop, the user-experience design, and the operational decisions about what Oswald can and cannot do are Brand Atlas’s, not the provider’s.
2. Provider names date the product. A docs site that says “powered by GPT-4” today is awkward when the underlying model is GPT-5 next year. Oswald keeps working.
3. The product relationship is with Brand Atlas. A brand owner using Guardian has a relationship with Brand Atlas. The provider is plumbing.

​

What Brand Atlas AI is, in design terms

Two products, designed for two distinct roles:

• Oswald. Editorial. Used by the brand owner. Has read and draft access to the brand record. Drafts changes for the brand owner to approve. Available on Guardian.
• Henry. Read-only. Used by the team. Has read-only access to the brand record limited to what the user can already see. Answers questions; cannot edit. Available on Keeper and Guardian, paid for by the customer through BYOK.

The two are designed not to overlap. A Guardian customer typically has both: Oswald for the brand owner’s editorial work, Henry for the team’s reading work. The two are visible in different parts of the portal; they are not conflated in the UI.

​

What the AI does at all

A working definition: Brand Atlas AI is a collection of language-model-backed assistants that read the brand record, understand questions about it, and produce answers and drafts that respect the brand’s vocabulary, voice, and decisions. Three properties matter:

1. Grounded in the brand record. Both assistants read the actual atlas (the eight standard sections, Horizons, history) as context. Answers are not hallucinations from general training data; they are drawn from the brand’s own material. When the brand record is silent on a question, the assistants say so.
2. Voice-aware. Both assistants follow the brand’s voice rules. Banned vocabulary is excluded; owned vocabulary is preferred; tone aligns with the Voice section.
3. Bounded. Both assistants have explicit boundaries. Oswald drafts, does not unilaterally apply. Henry reads, cannot edit. Neither assistant invents brand decisions.

​

What Brand Atlas AI does not do

Three things worth stating explicitly:

• It does not make brand decisions. Brand decisions are authorship work. The assistants draft and explain; the brand owner decides. A Guardian customer using Oswald daily is still the author of every change.
• It does not replace the studio. MadeBy_ services exist for work the AI is not the right tool for: original identity work, evolution projects, strategy. Oswald assists; it does not strategise.
• It does not have memory beyond the session. Henry does not remember last week’s conversations. Oswald does not remember last month’s drafts beyond what is in the brand record’s history. Each session starts from the current state of the atlas.

​

How the AI is presented in the UI

In the portal, the AI features appear as distinct elements:

• Oswald has its own surface in the context panel of sections and Horizons, and in the Update Requests panel.
• Henry appears as a chat icon in the corner, opening to a conversation panel.

Neither assistant interrupts. Neither makes suggestions unprompted. Both wait to be asked. This is by design: the team’s relationship with the atlas should be primary; the AI is auxiliary.

​

When the AI is the wrong tool

Three cases:

1. Original strategy. The brand’s positioning, audience, mission. The AI can summarise what exists; it should not be asked to invent.
2. Sensitive decisions. Crisis responses, legally-fraught statements, political moments. Decide off-platform; record the decision afterwards.
3. Replacing the brand owner’s read of the atlas. A brand owner who never reads the atlas because Oswald summarises it is missing the point. The summaries are an aid to attention, not a substitute for it.

​

How this docs site uses the assistants

Brand Atlas’s own docs are written by humans without AI drafting in the loop. This is a deliberate choice. The docs are evaluated by AI engines for citations (see AEO and GEO strategy), but the writing is human.

​

Related pages