How do you train AI on your brand?
You train AI on your brand by giving it the context, examples, assets and rules it needs before it produces anything. That means brand and strategy, plus whatever the piece depends on, such as audiences, objections, pricing and product features, along with examples of what good looks like and what it must never do. For one person in one tool, a custom GPT or project with your guidelines and ten strong examples gets close. For a team, each request (a deck, a post, a reply) needs to get the context and examples for its situation, plus the steps and template for that job, from one shared source.
The DIY way.
- Write the brand down. Turn the guidelines PDF into plain text or a brand .md file: positioning, audiences, voice, words to use and avoid, and visual rules. This alone makes output noticeably less generic. See machine-readable brand guidelines.
- Add the context the piece depends on. Write down what the brand file doesn't cover: the audience, the offer, pricing, product features, customer objections and proof points. Without it, output is on-voice but off-message. See production context.
- Add examples of good work. Collect ten or so pieces you'd happily publish, plus a few you wouldn't, and say why. Examples teach tone and structure faster than rules do. See brand examples and inspiration.
- Put it in a custom GPT, Claude Project or Gemini Gem. Upload the files and examples as knowledge, and write instructions for the job, such as LinkedIn posts. For one person in one tool, this gets close.
- Review and correct. Read every piece before it goes out, and feed fixes back into the instructions. This is where most of the time goes.
Where it breaks.
The DIY setup works for one person, one tool and one kind of output. A team multiplies it: every person builds their own project, every tool needs its own copy, and every situation (a sales deck, a LinkedIn post, a support reply) needs different context, examples and rules. A single set of instructions can't tell a deck from a post, so output drifts back to the average.
The setup also depends on whoever built it. When positioning changes, someone has to update every project, custom GPT and instruction file by hand, and nobody knows which copy each colleague is using. Review becomes the bottleneck, and AI saves less time than it promised.
What a system does differently.
A system starts from what the team already has, such as guidelines in Drive or Notion, PDFs, Figma files, the website and existing .md files, and structures it into production context: positioning, audiences, messaging, objections, proof points, pricing and product features. Alongside it sit examples of what good looks like, in general and for each format and channel.
When someone asks for a deck or a post, the system works out what the request is for and gives the agent the slice it needs: the context, the examples, the marketing playbook for that situation, the shell and the assets, in whichever tool they're working in.
Because the source is shared, an improvement made once reaches everyone, output is checked against the same standard whoever asked, and approved work becomes the next set of examples. The result is work that starts close to ready, rather than a draft someone has to rewrite. See what an agentic marketing brain is.
How Brail handles it.
Brail is an agentic marketing brain. You define what on-brand and on-quality looks like, and Brail understands what each request is for and equips the agent with everything it needs: context, examples and inspiration, playbooks, shells, tokens and assets. It keeps it all as versioned files synced across your tools, checks what goes out against your thresholds, and learns from approved work.
Checklist.
- Your brand is written down in a form AI tools can read, not only as a PDF.
- Agents get audiences, messaging, objections, pricing, product features and proof points, not just voice rules.
- Each common output (deck, post, email, reply) has examples of what good looks like, and a note of what must never appear.
- Each request gets the context and examples for its situation, not one set of instructions for everything.
- Recurring jobs have a playbook, so the steps, shell and checks aren't rewritten each time.
- Every AI tool your team uses works from the same, current version.
- Logos, colours, fonts and approved images are on call in every tool, not re-uploaded by hand.
- Approved work becomes the next set of examples.
Terms in this challenge.
- Agentic brand brain: An agentic brand brain is the brand side of an agentic marketing brain: the identity, voice, brand kit, assets, design system and examples that every AI agent uses, kept current and applied to each request.
- Brand system: A brand system is the method a brand operates by: the strategy, rules, components and examples designers build so a brand can be applied consistently across every channel and situation.
- Agentic brand system: An agentic brand system is a brand system written so AI agents can follow it as well as people.
- Machine-readable brand guidelines: Machine-readable brand guidelines are brand rules written in formats AI tools can read directly, such as .md, .json and .css.
- Production context: Production context is everything an AI agent needs to know to produce a specific piece of work: brand and strategy, audiences and messaging, GTM and marketing strategy, plus whatever the piece depends on, such as customer objections, pricing strategy, product features, proof points, competitors and launches.
- Marketing context: Marketing context is one part of AI production context: the market-facing part, covering audiences and their pains, positioning, channel strategy, campaigns and the messaging each audience should hear.
- GTM context: GTM context is one part of AI production context: what an agent needs to sell, not just describe.
- Company context: Company context is one part of AI production context: what an agent needs to talk about a company correctly to the outside.
- Marketing context layer: A marketing context layer is the governed source of customer, brand and business context that every marketing application, agent and workflow reuses, instead of each tool holding its own copy.
- Context layer for marketing AI: A context layer for marketing AI is the governed source between a company's knowledge and the AI agents that use it, giving every agent current, approved context instead of whatever each tool happens to hold.
- Customer context: Customer context is what an AI agent needs to know about the people it's producing for: their segment, account history, recent behaviour and where they are in the buying cycle.
- Marketing ontology: A marketing ontology is a shared model of the things marketing talks about (products, audiences, segments, messages, proof points, channels) and how they relate, so every agent reads them the same way.
- Approved claims library: An approved claims library is the set of statistics, customer quotes, results and statements an agent may use, each with its source, its owner and any rules about when and how it applies.
- Marketing playbook: A marketing playbook is a set of instructions for one situation, such as a sales deck, a LinkedIn post or a customer report.
- Brand skills: Brand skills are instructions packaged for AI agents, usually as SKILL.md files, that teach an agent how to apply a brand.
- Situational direction: Situational direction is the guidance an AI agent needs on top of brand context to do one specific job well.
- Brand routing: Brand routing is matching each request to the right context, examples, playbook and assets based on what the prompt is asking for, so an agent gets what this task needs and nothing it doesn't.
- Output brief: An output brief is the package an AI agent receives before it produces one piece of work.
- Brand context engineering: Context engineering for brand and marketing is designing what an AI agent receives before it produces anything: context, examples, playbooks, assets and rules.
- Brand toolbox: A brand toolbox is everything a person or agent needs to produce on-brand work: production context, examples and inspiration, playbooks, shells and blocks, the brand kit and assets, the design system and rules.
- Brand kit: A brand kit is a brand's core visual set in usable form: logos, colours, fonts and design tokens, plus a few key visuals, ready for an agent to apply.
- Design tokens: Design tokens are a brand's visual decisions (colours, type sizes, spacing, radius, shadows) stored as named values, usually in .json or .css, so tools and agents apply exact values instead of guessing.
- Brand shell: A brand shell is a ready-made, on-brand HTML starting point for one type of output, such as a sales deck, proposal or one-pager, with slots an agent fills with content.
- Brand blocks: Brand blocks are reusable on-brand sections, such as a hero, stat row, quote, case card or call to action, written as plain HTML and CSS so any agent or tool can assemble them.
- HTML-first marketing: HTML-first marketing is producing decks, proposals, reports and pages as HTML instead of .pptx or .png, because AI agents build and edit HTML well, and the result can be branded, versioned, published and tracked.
- Off-brand AI output: Off-brand AI output is content that sounds generic or wrong for the company, made when the AI has no brand context, only part of it, or an outdated copy.
- Launch-ready output: Launch-ready output is AI-generated work that can be published as it comes out, without rework: on brand, on strategy and on quality, often from a single prompt.
- Brand HTML: Brand HTML is a brand expressed as working HTML and CSS that AI agents fill with content instead of designing from scratch.
- Brand context: Brand context is one part of AI production context: who the company is, how it sounds and looks, and which rules and assets apply to what an agent makes.
- On-brand: On-brand describes output that matches a brand on eight checks: look right, sound right, right message, right audience, right channel, right asset, on strategy and on playbook.
- On-brand AI: On-brand AI is AI output, from text and design to code and images, that is on brand, on strategy and on quality from the first draft.
- Brand voice: Brand voice is how a brand sounds in writing and speech: its personality, word choice, rhythm and level of formality, and how those shift by audience and channel.
- Brand framework: A brand framework is the structured core of a brand's strategy: its purpose, positioning, audiences, value propositions, personality and key messages.
- Brand inspiration: Brand inspiration is the set of examples that show what good looks like for a brand, such as strong designs, decks and past campaigns, which people and AI agents follow to match its feel.
Comparisons.
- Prompting from scratch vs. an agentic marketing brain: Prompting from scratch works for one person and a few pieces.
- Claude Projects and custom GPTs vs. an agentic marketing brain: Claude Projects, custom GPTs and Gemini Gems are a good first step for one person in one tool.
- Marketing context layer and an agentic marketing brain: how they fit together: A marketing context layer makes customer, brand and business context reusable across tools and agents.
- Customer data platform vs. a marketing context layer: A customer data platform unifies data about customers so tools can target and personalise.
- Knowledge base vs. a marketing context layer: A knowledge base stores information for people to search and read.
Questions people ask.
- Why does AI content sound generic?
- Because the AI has no brand context, only part of it, or an outdated copy, so it falls back on the most common patterns in its training data. Giving it your voice, audiences, messaging and examples for the specific job fixes most of it. See why AI content sounds generic.
- How do I train ChatGPT on our brand voice?
- Create a custom GPT (or a Claude Project or Gemini Gem) with your voice rules, words to use and avoid, and around ten examples you'd publish, then write instructions for one job at a time. It works for one person; for a team, each situation needs its own context and examples, kept current for everyone in every tool. See the ChatGPT on-brand copy workflow.
- What context does AI need to write good marketing?
- More than the brand: the audience and their pains, positioning, the message for that audience, the offer, and what the piece should achieve, plus examples of good work for that format. See what context AI needs to produce marketing.
- How do I give AI examples of good work?
- Pick pieces you'd publish again for each format and channel, say what makes each one good, and keep a few that missed with the reason. Examples for the specific situation teach structure and tone faster than rules, and adding approved work keeps the set current. See brand examples and inspiration.
- How do I give AI our pricing, product features and customer objections?
- Write them down as structured context, such as a pricing summary, a feature list with what each does for the customer, and common objections with your answers, and include them in the instructions for the jobs that need them. Keep one current copy, or answers will contradict each other across tools. See GTM context and company context.
- How do I get AI to use our colours, fonts and logo in slides?
- Give it the exact values and files, not a description: colours and type sizes as design tokens, font files, logos and a slide shell to fill in. Agents build on-brand slides far more reliably from a ready-made shell than from a list of rules.
- Can I just upload our brand guidelines PDF?
- It helps a little, but a PDF is written for people to interpret. Agents do better with the brand in plain text or structured files, plus examples and the context each job needs. See machine-readable brand guidelines.
- What is a marketing context layer?
- A governed source of customer, brand and business context that every tool, agent and workflow reuses, instead of each tool holding its own copy. See marketing context layer and how it fits with an agentic marketing brain.