Brand
“A blank model knows everything, and nothing about you.”
The answer in 30 seconds
- →You keep AI on brand by giving every agent a brand contract — surfaced as brand memory: your tone of voice, product catalog and channel rules, in a form the model reads on every task.
- →The brand memory is enforced before review, not after — agents draft against your brand, not a blank model, so the human at the gate corrects taste, not basics.
- →Prompting one chat window per person doesn't scale: the knowledge lives in one head, and nothing learns. A shared brand memory is read by the whole team.
- →Every edit and rejection flows back in, so month-three drafts are sharper than week-one drafts — because your people corrected them.
Why does AI sound generic by default?
A blank model knows the whole internet and nothing about you. Ask it for a launch email and you get competent, forgettable copy that could belong to any company in your category. Generation has gone to zero; what's scarce is work that sounds like you, fits your catalog, and survives a compliance question six months later. The gap isn't model quality — it's that the model has never been told who you are.
What is a brand contract (and why “brand memory”)?
A brand contract is a single, machine-readable set of rules that defines what may — and may not — go out under your brand, for humans and AI agents alike. In practice that means your tone of voice, product catalog, claims and channel rules — the things a good agency would internalise before writing a line. We surface it publicly as your brand memory, because that's what it does: every agent reads it before drafting, so it doesn't start from zero every time. One agent drafts the campaign email, another the sales follow-up, a third the support reply — all from the same brand memory, so the bank never sounds like the brewery.
[ figure — one brand memory, several agents drafting from it ]
Why enforce brand rules before review, not after?
Most teams run AI as an open loop: prompt, paste, edit, ship. The edit — the most valuable judgment in the chain — evaporates. If the brand rules only live in the reviewer's head, every draft arrives wrong and the human spends their time fixing tone instead of making judgment calls. Enforced before review, the brand memory catches the rule-level issues up front. The named human at the publish gate approves taste, not typos.
How does the brand memory get sharper over time?
The approval isn't overhead — it's how the system learns your taste. Edits and rejections flow back into the brand memory. The drafts you see in month three are sharper than the drafts in week one, because your people corrected them. Honest constraint: in week one, the drafts will need real editing. That's not a flaw in the loop — it is the loop.
Questions your team will ask
Is this just a better prompt?
No. A prompt lives in one chat, in one person's head, and disappears when the tab closes. A brand memory is read by every agent on every task, and it improves as your team corrects it.
What goes into the brand memory?
Whatever you already have: brand-book PDFs, old campaigns, tone-of-voice docs, the product catalog, that one Notion page nobody can find. You connect sources as they are — you don't clean up first, you connect first.
Can each client or brand have its own?
Yes. A separate brand memory per client, so the bank never sounds like the brewery. Every deliverable still passes your approval gate.
Does it replace my marketers?
No. It removes their production grunt work and leaves them the judgment calls.
Brand