Why guidelines are the whole game now
A brand book used to be a PDF nobody opened. In the AI age it is the single most valuable brand asset you own, because it is the thing that steers every AI-generated word and image off the category average and onto your brand. In 2026, documented guidelines - especially verbal, with sample copy and a "never say" list - are a requirement, not a nicety. Tonight you make Cadence's, and turn it into a brand GPT that enforces itself.
Guidelines an AI can actually obey 6 min live
Human guidelines say "be confident and warm". An AI cannot use that. Machine-usable guidelines give rules, examples, and hard boundaries: sample copy for each context, explicit "never say" lists, and decision rules. Same brand, written so a model produces it on demand.
LiveThe four things AI needs from your guidelines3 min▶
Translate the human brand book into the parts a model can act on:
- Positioning + personality in a line each - what Cadence stands for and how it behaves, so the model has a target.
- Voice rules with sample copy - three or four principles, each with a do and a don't example. Examples teach far better than adjectives.
- The "never say" list - banned words, claims, and tones. This single list removes most off-brand output.
- Decision rules - how to handle length, formality by channel, and edge cases, so the model does not improvise your brand.
The "never say" list that saved a voice. One team's AI content kept drifting into hype - "revolutionary", "seamless", "unlock". Adding a six-word never-say list to their brand GPT did more for consistency than any amount of "be authentic" coaching. Boundaries, not adjectives, are what a model can actually follow.
Self-studyLiving guidelines, not a frozen PDF2 min read▶
Machine-usable guidelines are a living document: version them, add examples as edge cases appear, and keep the brand GPT in sync. A guideline the team cannot find or the GPT does not have is a guideline that does not exist. Store the source of truth somewhere both humans and your tools can reach.
Drive it yourself: watch consistency get built ★ 8 min · everyone plays
Here is the whole lesson in one control panel. The AI is generating Cadence brand assets. Every foundation you turn on - positioning, voice + never-say, personality, visual system, guidelines - steers it further off the average. Watch on-brand consistency climb and flip to the scorecard to see how many asset types come out on-brand. Notice what never moves it: a bigger model.
Assemble Cadence's machine-usable guidelines ★ 12 min · everyone builds
Pull everything from b2-b5 into the one-page system a model can obey. This is the artifact the brand GPT will run on.
Gather your Cadence outputs so far: positioning (b2), personality + archetype (b3), naming + voice (b4), visual direction (b5).
Ask the AI to compress them into a machine-usable brief: positioning line, personality line, 4 voice rules each with a do/don't example, a never-say list, and 3 decision rules.
Edit hard - especially the never-say list and the examples. These carry the most weight in steering the model.
This one page is your source of truth. It powers the brand GPT in Demo 2 and every content session after.
Build the Cadence brand GPT ★ 8 min · build your own
Load the system into a reusable assistant so anyone on the team generates on-brand without you in the room. This is how a small brand acts consistent at scale.
Open a custom GPT / project / custom-instructions space in your AI tool. Paste the one-page machine-usable brief from Demo 1 as its standing instructions.
Add the operating rule: "Generate strictly to this brand system. Refuse or flag anything that violates the never-say list. If information is missing, ask rather than invent."
Test it on three asset types - a tagline, a social post, and a sales-email opener. Check each against the system: on-message, on-voice, no banned words.
Compare to a plain chatbot with no system given the same ask. The gap you see is exactly the consistency meter in the simulator, made real.
Save the brand GPT and share it. You just turned a documented system into an enforcement engine - the b8 activation session runs on this.
How a five-person team sounds like one voice. A startup could not afford a copy chief on every asset, so they built a brand GPT from their system. New hires, contractors, and the founder all ran drafts through it first. The brand held together not because everyone had taste, but because the system did - and the model obeyed it.
Try it yourself - this week ◐ 30-45 min total
- Finish Cadence's one-page machine-usable brief and build the brand GPT if you did not live.
- Play the simulator again and predict the consistency score before each toggle - predicting it means you understand what drives it.
- Run your real brand (or one you admire) through the "compress to machine-usable" prompt. Note the biggest gap - usually a missing never-say list.
- Test your brand GPT on one asset type you actually ship. Keep what is on-brand; log what it got wrong to refine the guidelines.
- Optional: version your guidelines doc and decide where the single source of truth will live.
Official sources covered
Taught from public brand-AI guidance. Certificates and paid tooling stay official. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · In the simulator, what makes every asset type come out on-brand?
Consistency is the system filling in, not the model. Each foundation steers the AI further off the average and onto the brand.
2 · What do machine-usable guidelines add that a human brand book often lacks?
"Be confident and warm" is useless to a model. Examples and boundaries are what it can actually obey - the never-say list does the heaviest lifting.
3 · What is a brand GPT?
A brand GPT turns your documented system into an enforcement engine - the same system, applied a thousand times without you in the room. A human still audits.