Lions Law: train the AI to understand the brand

You may not agree with AI on your brand — have you trained it to understand your brand?

That is Lions Law, on the record. Not a mood. A test.

If a strategist and a model cannot recognise the same brand truth, you do not have a partner. You have a talented stranger with access to the account. Most “brand voice” fields in tools fail this test in a sitting. They store adjectives. They do not store standing.

Recognition is the job

Humans and machines must recognise the same brand truth. Same promise. Same bans. Same audience. Same claims that are allowed to leave the building, and the ones that are not.

Recognition is not mimicry. A model can mimic “warm, expert, slightly irreverent” for fifty brands before lunch. That is fluency. Fluency is not authority. Lions Law asks a harder question: would this brand’s own people accept this as a description of themselves — and would the machine, asked again on Thursday, give the same answer?

If the Thursday answer drifts, you did not train it. You chatted with it.

What training actually is

Training, here, is not fine-tuning a foundation model in a lab. It is not a PDF in a shared drive. It is a durable brand record the work is generated from, not a prompt the work is generated near.

At minimum the record has to answer:

• Who we are for, and who we are not for
• What we promise, in language a buyer could repeat
• What we will never claim (the ACCC does not care that the caption was “AI”)
• Proof we are allowed to use — and proof we are not
• Visual and verbal non-negotiables
• The organising idea, if you have one, as protected state — not a slogan to paste under a stock scene

If a junior cannot run the brand from that record, neither can an agent. Lions Law is a kindness to both.

The NIST AI RMF talks about mapping context before you manage risk. Agencies skip the map and go straight to captions. Then they are surprised when the OECD still wants a human accountable for the output. Of course they do. The machine never took the brief.

ACCC: claims should be true, accurate, based on reasonable grounds. Training is how you stop the model inventing grounds.

Prompting is not Lions Law

A prompt is a request. Training is a constitution.

You can prompt “don’t sound corporate” and still ship corporate, because the model’s prior is the internet. You can prompt “we’re the challenger” while the brand’s actual proof is a 40-year incumbency. The machine will pick the more common story. Lions Law forbids that substitution.

In an online agentic closed loop, brand truth is step one and step seven. It goes in. It comes back. Learning is not allowed to hijack it because a reel performed. That is the closed. Open loops let Thursday’s engagement rewrite Monday’s meaning.

For agencies, per-brand isolation is part of the law. Your franchise client must not leak into your boutique client. Qy Partners is the operational form of that — not because a logo is white-label, but because each brand’s truth is sovereign.

If you cannot yet write the record, do not generate. Write the record. That is the work. Then generate inside it. So What Law still kills anything the buyer cannot use.

Kitchen table, same law: your own name has to mean one thing to you and to the machine. Restart. Soft door: app.theagencyiq.ai.

Frequently asked questions

What is Lions Law?

You may not agree with AI on your brand — have you trained it to understand your brand? Humans and machines must recognise the same brand truth.

Is a brand voice document enough?

Not if it is adjectives. Training is a durable record of promise, bans, audience, allowed claims, and protected idea — the work is generated from it, not near it.

How is this different from prompting ChatGPT with a brand kit?

A kit in a prompt is a request the model may ignore. Lions Law requires recognition that holds on Thursday, and a loop that cannot hijack the truth because content performed.