Your brand guide used to have exactly one reader: a designer who could ask you a question. It now has a second reader that cannot.
Through 2026 the tools your team already pays for stopped starting from a blank canvas. They start from a stored brand kit — your colours, your fonts, a written description of your voice — and produce finished work from it. That changes what a brand guide is for. A document written to be interpreted by a person is now being read by software that does not interpret. It executes what it can parse and guesses at the rest.
What changed while you weren't looking?
Three moves, all this year, all inside tools a small business already uses. LinkedIn put brand rules into Campaign Manager: since June 2026 an advertiser can set a colour palette, a heading face and a body face, and a written description of its voice, and the platform references those when it drafts an ad. Where no voice is supplied, LinkedIn assembles one from your past posts and your Company Page — which is the whole problem in one sentence.
Figma opened its canvas to agents. On 15 October 2026 it added 20 new Model Context Protocol partners, and a connected agent can read design context — variables, components and layout data — then create code layers, edit specs and generate images on the canvas. Your design system is no longer a reference for humans. It is an input to whatever drafts next week's campaign.
And the brand platforms started shipping the same pipe. Frontify's own guidance, published in August 2026, names the format as the problem rather than the tooling: a PDF brand book was never built to hand an AI tool structured, retrievable rules, and what it recommends instead is a short file of explicit rules a model can execute. Its connector exposes 52 tools to an agent, the guidelines themselves included, so tone and usage rules arrive inside the tool where the work is happening, without anyone opening a portal.
Why does a 40-page PDF fail here?
Because a brand book is written in advice and a generator needs values. “Use the blue confidently” is a sentence a designer can act on. A model cannot. Three things go wrong, in the order we see them. Ambiguity: the file holds two blues and no rule for which one is the button. Absence: there is no rule for the situation the tool is actually in — a dark ground, a square crop, a 30-word caption, a second language. Unreachability: the rule exists, correct and complete, in a file nobody connected to the tool doing the work.
The cost is quiet, which is why nobody bills it. The output looks plausible, lands roughly on palette, and no one files a ticket. IAB's survey of 125 US advertising executives, published in August 2025, found 70% had already hit an AI incident, with off-brand and offensive material named among the failure modes — while fewer than 35% planned to increase investment in AI governance or brand-integrity oversight over the following year. The distance between those two numbers is the whole opportunity.
Your brand guide is now an interface
Every tool that generates on your behalf works from some set of brand rules. If you do not supply them, it infers them — from your old posts, your website, or its own defaults. Silence is not neutral. It is a decision to let the model guess.
What goes into a brand file a machine can use?
Keep it short, explicit and in one place. One test decides every line: could a tool with no taste and no context follow it without asking you a question? If the answer is no, that line is a preference, not a rule.
- Name every colour with a role and an exact value, never a mood.
- State which text colour sits on which ground, with the contrast ratio.
- Fix the type scale: heading face, body face, weights, and what is never used.
- Give the mark its clear space, its minimum size and the grounds it may sit on.
- Write voice as do-and-do-not pairs, each with one example line.
- List banned words, banned claims, and the names you refuse to translate.
- Set the per-language rules: script, digits, direction, and the face each language uses.
- Keep one versioned file with a date and a named owner.
Where do you put it?
In two places, and they are not interchangeable. Fill the brand-kit fields in every platform that has them, because that copy is what the generator reads at the moment it drafts. Then keep the original somewhere you own — a file in your own repository or drive — so each vendor's version is a copy you can re-issue rather than the only record. The order matters when the rules change: you edit the original, push it out, and can tell who is still running an old version. One more rule, and it is not about files: nothing publishes straight out of a generator. A person approves, every time.
A brand guide nobody can parse is not a standard. It is a preference, and the model will overrule it.
How do you know it worked?
Run one brief twice. Give a tool your brand file and ask for a single-image ad for your current offer; run the same brief again with the brand file removed. Put the two results side by side. If you cannot tell them apart, the file is decoration — either nothing is reading it, or it says nothing a machine can act on. That is a check worth having, because it can fail. Then run the human version: hand one line from the file to someone outside your team and ask them to apply it without asking a question. Every line they have to ask about is a line the model will quietly guess at.
Write the rules once
Hamed Azimi
Founder, NexusWave
Hamed Azimi is the founder of NexusWave Technologies, a Vancouver-based AI-native digital agency and product studio.
We write these guides from real client work — GEO, SEO, and builds we've shipped. If a piece raised a question about your own site, we're happy to look at it.
- Straight answers, no retainer required
- We reply in your language — EN · فارسی
- Based in Canada, working worldwide
