The Brand Brain: 10 Marketing Assets for AI that Every Business Needs

58% of small businesses now use generative AI. Only 26% get real value from it.

That gap is not a tools problem. It is a fuel problem.

The AI produces generic work because you gave it generic inputs.

A marketing asset in the AI era is any structured, machine-readable, single-source-of-truth file that encodes one aspect of your business (who you serve, how you sound, what you sell, how you look, what you measure) so that humans AND AI agents can produce on-brand, on-strategy marketing from it.

This guide names the 10 assets every small business needs to future-proof its marketing.

It is the detailed companion to our earlier piece on AI marketing systems for small business, which introduced the ARMS framework. ARMS Layer 1 is the Intelligence Foundation. M10 is what exactly goes in that foundation.

Every asset in this guide comes with what it is, why it matters in the AI era, how to build it, which tools to use, and a 5-check “agent-ready” test.

At the end, we show how the 10 assets assemble into a single Brand Brain that you can load into Claude Projects, ChatGPT, Gemini Gems, OpenClaw, Hermes, Manus, or any agent workflow.

---

Why 2026 Changed What “Marketing Assets” Means

A marketing asset used to be a thing you handed to a designer, an agency, or a printer. A logo file. A brochure. A template.

Then it became a thing you handed to a marketer. A style guide. A persona doc. A content calendar.

Now it is a thing you hand to an agent. And the agent might be a person on your team. It might be Claude. It might be an automated workflow running at 3 a.m. The asset has to work for all of them.

That shift sounds small. It is not.

When the only consumer of your brand guide was a human designer, the guide could be a 40-slide PDF with aspirational mood-board images and a paragraph about “the feeling of the brand.” The designer would absorb the vibe and translate it into work.

An AI agent cannot absorb a vibe.

If your brand voice lives in your founder’s head, the AI writes in nobody’s voice.

If your ideal customer lives in a sales rep’s intuition, the AI writes to a generic “small business owner.” If your positioning lives in a 12-month-old slide deck that contradicts your current pricing page, the AI picks one and hopes.

That explains the paradox every small business is living right now:

The tools work. The inputs do not exist.

Marketing Assets as Code. Before Infrastructure as Code, every server was configured by hand and no two were quite the same. Every new environment was a fresh round of guessing, and the guesses silently diverged. Then the industry made server configuration machine-readable and auditable, and the guessing stopped.

The same shift is happening to marketing. Pandya calls this treating design decisions as infrastructure. We apply that frame to all 10 assets. When your ICP, voice, offers, design tokens, and editorial calendar live as structured files that every agent reads and every audit can enforce, AI stops guessing. Treat every asset in this guide as code: version it, lint it, and update it in one place.

Three more numbers tell the same story:

• 78% of marketing teams report data fragmentation across platforms (NinjaCat 2026). The brand lives in 30 different places.
• 50% of AI agents operate in silos with no shared data or coordinated output (ConvertMate 2026). Each tool invents its own version of you.
• Organizations with integrated AI systems see 3x better outcomes than silo’d ones (ConvertMate 2026). Integration begins with integrated assets.

The fix is not another tool. The fix is 10 files.

Build them, structure them, version them, load them into every AI agent you use, and the same model that produced generic slop last week produces a landing page that sounds like you, sells what you sell, and reads like your best rep wrote it.

The model did not get smarter. The inputs got real.

The rest of this guide is those 10 files.

---

---

The Marketing Foundation 10 (M10)

M10 is the detailed build specification for the Intelligence Foundation layer of ARMS. Four sub-layers, 10 assets.

Layer	Assets	What it answers
Identity	1. ICP + Voice-of-Customer Library 2. Brand Voice, Tone & Terminology 3. Visual Brand System 4. Design System (DESIGN.md + theme.json)	Who we serve, how we sound, how we look, what our components are made of
Market	5. Offer, Positioning & Competitive Intelligence 6. SEO & Keyword Strategy	What we sell, against whom, for what searches
Surface	7. Website & Core Pages Architecture 8. Content Library, Proof Corpus & Editorial Calendar	What the world sees, reads, and cites
Intelligence	9. Data, Analytics & Performance Library 10. Agent Operating System (“Brand Brain”)	What we measure, what we prompt, how it feeds back

Each asset has four required attributes:

1. Machine-readable. Markdown, JSON, or CSV. No PDFs. No proprietary formats. If an AI cannot parse it, it is not an asset.
2. Structured. Clear sections, labeled fields, predictable headings. Agents need to locate “the customer’s top objection” in one place, not hunt through prose.
3. Single source of truth. One file per asset. When something changes, you change it here, and every agent, tool, and workflow that reads it gets the update.
4. Versioned. Git, or at minimum a versioned document store. You should be able to see what the asset said three months ago and why it changed.

If an asset in your company fails any of those four tests, it is not yet an asset. It is notes.

Each section below includes a 5-check agent-readiness test so you can grade your own files as we go.

---

Asset 1: ICP + Voice-of-Customer Library

What it contains. One file per primary persona with firmographics (company size, revenue, industry, geography), demographics, psychographics, jobs-to-be-done, trigger events, pain points, objections and rebuttals, buying-committee roles, budget range, decision criteria, content-consumption habits, disqualifiers, and a Voice-of-Customer section: direct quotes from sales calls, reviews, support tickets, Reddit, Quora, and G2.

Why it matters in the AI era. Every agent decision about tone, angle, proof, and CTA starts here. If the ICP is one line (“small business owners”), every AI output is one-size-fits-nobody. If the ICP is 1,500 words of structured detail plus 50 customer quotes, agents write in language customers actually use.

How to build it:

1. Pull 10 recent closed-won deals and 10 recent closed-lost deals. Look for patterns in industry, size, trigger, and objection.
2. Interview or review 5-10 recordings with your best customers. Ask them to describe the problem, the alternatives they considered, and why they chose you. Transcribe the exact language.
3. Scrape public sources for the language your market actually uses: Reddit threads in the customer’s industry, G2/Capterra reviews of you and competitors, Quora answers to related questions, YouTube comments under relevant videos.
4. Draft one persona file per primary buyer. Include Jobs to Be Done, triggers, objections, rebuttals, and a VoC quote block.
5. Version the file. Update it every quarter or after every 10 closed deals, whichever comes first.

Agent-ready checklist:

• Machine-readable format (Markdown or YAML), not slides or PDF.
• Fields are labeled, not buried in prose.
• Voice-of-Customer quotes are tagged by source and date.
• Single file per persona, linked from a top-level PERSONAS.md.
• Last updated in the last 90 days.

Example: Our starter template at 01-icp/ICP.md uses “Brenda, Operations Director at a 50-person home services company” as a worked example. You can fork it and replace Brenda with your own.

---

Asset 2: Brand Voice, Tone and Terminology Guide

What it contains. Point-of-view statement (what we believe about our market), brand archetype, voice attributes (4-6 adjectives with do/don’t examples), tone modulation by channel (LinkedIn vs. email vs. sales deck), reading-level target, sentence-length rules, punctuation rules, formatting conventions, an annotated sample library with 5-10 do/don’t pairs, a preferred terminology dictionary (the words we use and how we define them), a banned-words list (with reasons), and claims and compliance guardrails (what the agent can and cannot assert).

Why it matters in the AI era. An AI model can mimic any voice once. It cannot invent yours. The terminology dictionary is the single biggest determinant of whether AI output sounds like you or like a generic AI. “Leads” vs. “customers” vs. “buyers.” “Team” vs. “staff” vs. “crew.” Every word choice compounds across thousands of outputs.

How to build it:

1. Audit your last 20 published pieces (emails, blogs, ads). Pull 20-30 sentences you are proud of. Pull 10 sentences that missed. Label what makes each work or fail.
2. Write a 150-word point-of-view statement. What do you believe about the market that most competitors do not? What will you never say? What do you always say?
3. List 4-6 voice attributes. For each, write a do/don’t sentence pair.
4. Build the terminology dictionary. Start with every term that appears in your homepage, top 10 blog posts, and sales deck. Mark each as “preferred”, “use with caution”, or “banned”, with a reason.
5. Write 5-10 do/don’t pairs as annotated samples (same sentence written two ways, with commentary on why one works and one fails).
6. Add a compliance section: industry-specific claims you cannot make, required disclosures, review triggers.

Agent-ready checklist:

• Voice attributes are specific and paired with do/don’t examples.
• Terminology dictionary is a list, not a paragraph.
• Banned words have reasons, not vibes.
• Sample library includes 5+ annotated sentence pairs.
• Compliance section exists even for unregulated industries.

---

Asset 3: Visual Brand System

What it contains. Logo family (primary, secondary, monochrome variants, clear-space rules, minimum sizes) with the actual SVG and PNG binaries, color palette (primary, secondary, accent with hex, RGB, HSL, and OKLCH values), typography (display, body, mono with web and print sizing), spacing scale, radius and shadow tokens, accessibility rules (WCAG contrast minimums per color pair), imagery and photography direction, and baseline AI image prompts for Midjourney, Adobe Firefly, Nano Banana, Canva AI, and ChatGPT image generation.

Why it matters in the AI era. Every AI image tool needs a style anchor. Without baseline prompts and banned styles, Midjourney gives you generic “business imagery” that looks like a stock site. With baseline prompts that include your palette hexes, composition rules, and banned styles, Midjourney produces work that matches the rest of your brand.

How to build it:

1. Export every logo variant as SVG and PNG at sizes 64px, 256px, 1024px. Commit to /assets/logos/.
2. Define your color palette with exact values. Include 1 primary, 1 accent, 2-3 neutrals, plus any functional colors (success, warning, error).
3. Pick your fonts and declare fallbacks. Note exact weights and sizes for headings, body, and small text.
4. Write the imagery direction: subject matter (abstract shapes, product photography, illustrations, no people, etc.), composition (centered, asymmetric, full-bleed), color treatment, and lighting.
5. Write 3-5 baseline AI image prompts for your most-used tools. Each prompt should include your palette hexes, composition rules, and “always exclude” modifiers (no stock-photo aesthetics, no generic business imagery, no AI-generated humans if your brand does not use people).

Agent-ready checklist:

• Logo binaries are checked into the repo or asset library, not “in Figma somewhere.”
• Color values are given in multiple formats (hex + OKLCH minimum).
• Imagery direction includes banned styles, not just preferred ones.
• AI image prompts are tested and checked into a /prompts/ folder.
• WCAG contrast is verified for every text-on-background pair.

---

Asset 4: Design System (DESIGN.md + theme.json)

What it contains. A DESIGN.md file that translates the visual brand system into machine-readable tokens (colors, typography, spacing, radius, shadows, breakpoints, motion) plus a theme.json file with the same tokens in parseable JSON, plus component patterns (buttons, cards, forms, layouts) and rules for how components are built.

Why it matters in the AI era. AI coding tools (Cursor, GitHub Copilot, v0, Lovable, Bolt) are now building production UI. They need a design system they can reference. Handing an AI coding tool “a vibe” produces Tailwind slop. Handing it theme.json produces on-brand components the first time.

The spec is now open. In April 2026, Google Labs open-sourced the draft DESIGN.md specification through its Stitch design tool. The format encodes the reasoning behind a design system (what each color is for, WCAG rules, component intent) so AI agents can generate interfaces that match your brand and validate their own choices against accessibility rules. This is the direction the industry is moving: a portable, cross-platform standard for machine-readable design context. We recommend authoring your DESIGN.md against the Stitch draft spec so your file works today in Cursor, v0, Lovable, and Bolt, and works tomorrow in Stitch and whatever tools follow.

The four things a design system needs to be AI-ready. Pandya’s public teardown of this for Atlassian (hvpandya.com/llm-design-systems) is the clearest breakdown available. In one AI coding session, an LLM makes 200 to 300 silent visual decisions: what padding, what shade of blue, what border radius, what font weight. Each decision looks fine in isolation. Two hundred decisions later your prototype is off and you cannot say why. The next session the LLM starts over with zero memory of yesterday’s choices and picks 200 new values. By session ten it looks like three products built by three teams who never spoke.

A design system is AI-ready when it constrains every one of those decisions:

1. Spec files the AI reads every session. Structured Markdown documents for foundations (color, spacing, typography, motion) and for each component. Solves the memory problem. If the spec does not exist, the AI guesses. If it exists, the AI looks it up.
2. A closed token layer the AI picks from. Instead of hundreds of hardcoded hex values and pixel measurements scattered across files, one tokens.css with named variables. The AI cannot fabricate a new blue because the only blue it can reference is var(--color-link).
3. An audit script that catches what the AI gets wrong. A lint task that scans your codebase for hardcoded values and suggests the correct token for each violation. Runs in CI. Returns exit code 1 on any error. Drift cannot ship.
4. Drift detection for upstream design systems. If you depend on a component library (Atlaskit, Carbon, MUI, Radix, shadcn), a sync routine flags your spec files when the upstream ships changes so your specs stay current instead of quietly going stale.

Three-layer token architecture. The shape that makes all of this work is a token file with three layers of indirection:

/* Layer 1: upstream design system tokens */
--ds-text: #292A2E;
--ds-space-100: 8px;

/* Layer 2: your project aliases, with the raw value as fallback */
--color-text: var(--ds-text, #292A2E);
--space-100: var(--ds-space-100, 8px);

/* Layer 3: components only reference aliases, never raw values */
color: var(--color-text);
padding: var(--space-100);

The alias layer is what protects you. If the upstream library renames a token, you update one alias. Dark mode, high contrast, future themes: the chain resolves automatically. No component file ever touches a raw hex or pixel value.

Spec file hierarchy. Organize specs in three tiers so the AI can find what it needs in a predictable place:

specs/
├── foundations/        # color, typography, spacing, radius, elevation, motion
├── tokens/             # master map of every CSS variable and when to use it
├── atoms/              # button, input, icon, avatar, badge
├── molecules/          # tabs, dropdown, modal, form, tooltip
├── organisms/          # navigation, table, page header, content panel
└── patterns/           # layout rules, form layout, three-column, error handling

Each component spec follows the same 8-section template: metadata, overview (when to use, when not), anatomy, tokens used, props, states (default, hover, active, focus, disabled, error), code example, cross-references to related components. The consistency is the point. When the AI builds a form it reads patterns/form-layout.md for spacing, molecules/form.md for structure, atoms/input.md for fields, and tokens/spacing-tokens.md for exact values. Every choice is a lookup, not a guess.

How to build it.

1. Copy your visual brand tokens (from Asset 3) into a theme.json file using OKLCH for colors and rem for spacing. Add a tokens.css with the three-layer indirection above.
2. Write a DESIGN.md that explains how the tokens relate. Which spacing scale is canonical. Which radius for cards vs. buttons. What motion duration for hover vs. enter. Follow the Google Labs DESIGN.md open spec so any compliant tool can parse it.
3. Create a specs/ directory. Write foundation specs first (color, spacing, typography). Then write one component spec per component that actually exists in your UI, using the 8-section template. A small, accurate spec layer beats a comprehensive but stale one.
4. Add scripts/token-audit.js (or a shell equivalent) that scans every CSS file for hardcoded values and prints file, line, violation, and suggested token. Wire it into CI. Zero errors required to merge.
5. Add a section to your AI instruction file (CLAUDE.md, .cursorrules, or equivalent) that says: “Before writing or modifying any UI code, read the relevant spec file in specs/. Use only tokens from tokens.css. Run the token audit script before committing.”
6. Pin the version of any upstream design-system package you depend on. Add a weekly or monthly sync routine that flags spec files when the upstream ships changes.

The bootstrap prompt. If you are starting from scratch, Pandya’s six-step setup prompt (audit hardcoded values, build the token layer, write spec files, create the audit script, replace hardcoded values, wire up project instructions) gets an AI coding agent through the whole build in one session. Paste it at the root of your project in Cursor or Claude Code. Typical turnaround is a few hours, not a few days. Full prompt and walkthrough: hvpandya.com/llm-design-systems.

AI-ready design systems worth studying. Four public examples, all with spec files, closed tokens, and structured documentation:

• Atlassian Design System (the system Pandya ran his teardown against).
• IBM Carbon Design System (mature three-tier token architecture, heavy enterprise adoption).
• CMS Design System (Centers for Medicare & Medicaid Services, rare public-sector example, strong accessibility documentation).
• Nordhealth Design System (compact, well-structured, easy to adopt as a reference).

Fork the structure, not the visuals. Your tokens and voice stay yours. The skeleton they use is what makes those systems AI-legible.

Agent-ready checklist:

• theme.json is valid JSON, parseable without errors.
• tokens.css uses three-layer indirection (upstream tokens, project aliases, components reference aliases only).
• DESIGN.md follows the Google Labs open spec (or a documented equivalent structure).
• specs/ directory contains foundation specs plus one file per component, each following the 8-section template.
• A token audit script runs in CI and returns exit code 1 on any hardcoded value.
• Upstream design system dependencies are version-pinned, with a documented sync routine.
• Files live in the code repo, not in Figma.
• Production code references the same tokens (no drift between spec and shipped).

Your tenth AI session should produce the same visual quality as your first. If it doesn’t, the system is leaking somewhere in one of those eight checks.

---

Asset 5: Offer, Positioning and Competitive Intelligence

What it contains. Productized offers (name, scope, deliverables, timeline, outcomes, pricing), packaging tiers, outcome promise statements, differentiation story, proof points, a competitor map (direct, indirect, substitute), per-competitor breakdowns (positioning, pricing, strengths, weaknesses, public messaging themes), battle cards, and win-loss notes from closed deals.

Why it matters in the AI era. Every piece of copy positions you against alternatives, whether you wrote it to or not. If the AI does not know your competitors or your differentiation, it writes generic “we are a great partner” language that sells nobody. If the AI knows you compete against X and win on Y, it writes copy that frames Y on every page.

How to build it:

1. Write one offer sheet per productized service. Name, scope, what is included, what is not, timeline, pricing, outcomes the customer can expect.
2. Map your competitor landscape. Direct competitors sell the same thing to the same buyer. Indirect competitors solve the same problem differently. Substitute options are “do nothing” or “do it in-house.”
3. For each competitor, document: their positioning line, pricing if public, 3 strengths, 3 weaknesses, the messaging themes they push, and where they typically beat you or lose to you.
4. Write 5-10 battle cards: “when a prospect brings up X, we say Y, and here is the proof.”
5. Build a win-loss log. After every closed deal (won or lost), log why in 100 words. Read the log quarterly for pattern shifts.

Agent-ready checklist:

• Offers are fully productized, not “custom every time.”
• Competitors are categorized, not a random list.
• Each competitor entry has 3 strengths and 3 weaknesses.
• Battle cards are sentence-ready for copy use.
• Win-loss log has at least 10 recent entries.

---

Asset 6: SEO & Keyword Strategy

What it contains. Primary target keywords, secondary targets, long-tail cluster map, branded vs. non-branded split, search-intent mapping (informational, navigational, commercial, transactional), SERP-feature targets (featured snippets, People Also Ask, AI Overviews), GEO/AEO citation targets, content-cadence plan, internal-link architecture, and a research log.

Why it matters in the AI era. Traditional SEO and GEO (Generative Engine Optimization, the practice of making your content citable by ChatGPT search, Perplexity, Gemini, and AI Overviews) both depend on structured intent mapping. Without a keyword strategy file, agents write “content” without a target. With one, every piece maps to a cluster, an intent, and a reason to exist.

How to build it:

1. Run a keyword audit. Start with your top 3 competitors’ ranking pages (via SEMrush, Ahrefs, or free tools like Ubersuggest).
2. Group keywords into topic clusters. Pick 5-10 pillar topics. Under each pillar, list 10-30 cluster pages.
3. Map each keyword to intent (info, nav, commercial, transactional).
4. For each cluster, name your target SERP features. Featured snippet for the top-of-funnel term. PAA for common sub-questions. AI Overview citation for “what is” queries.
5. Write the cadence plan: one pillar refresh per quarter, N cluster pages per month, N updates per month.
6. Maintain a live research log: what we tried, what ranked, what didn’t, what to try next.

Agent-ready checklist:

• Keywords are grouped into pillar-and-cluster structure.
• Each keyword is tagged with intent.
• SERP-feature targets are named per cluster.
• Internal-link plan is documented, not ad-hoc.
• Research log is updated monthly.

---

Asset 7: Website & Core Pages Architecture

What it contains. Inventory of core pages (home, about, services, pricing, proof/case studies, FAQ, contact, legal), information architecture tree, canonical URL rules, navigation logic, schema markup plan per page type, and conversion-path maps.

Why it matters in the AI era. Your website is the primary surface AI search engines cite. Page structure, schema markup, and FAQ clarity directly determine whether Perplexity and ChatGPT pick your content when a prospect asks a relevant question. It is also the surface every AI-generated ad, email, and social post points back to, so conversion-path clarity compounds across every channel.

How to build it:

1. Audit current pages. For every URL, note: purpose, target keyword cluster, target intent, conversion action, schema markup status, last updated.
2. Rewrite your IA tree. Start from the home page, 2-3 clicks to any destination, grouped by job-to-be-done.
3. Standardize URL structure (lowercase, hyphens, no dates in URLs for evergreen content, plural or singular consistent).
4. Write a schema plan. Organization for the homepage, Service for each service page, Article / BlogPosting for blog, FAQPage judiciously (Google limits rich results), LocalBusiness if local, Product for SaaS pages, BreadcrumbList everywhere.
5. For every high-intent page (pricing, contact, top-of-funnel blog), draw the conversion path: what the visitor reads, the one thing they click, where they land, the confirmation.

Agent-ready checklist:

• Page inventory is a table with purpose, keyword, intent, conversion, and schema status.
• URL conventions are documented.
• Schema markup is live on every core page (verified in Rich Results Test).
• FAQ pages are on-page but not over-marked-up (one FAQPage block per site is often enough).
• Conversion paths are mapped for every high-intent page.

---

Asset 8: Content Library, Proof Corpus & Editorial Calendar

What it contains. An inventory of every content asset (blog posts, guides, videos, podcasts, webinars, lead magnets, case studies) with metadata (title, URL, status, target persona, funnel stage, topic cluster, last updated, performance), a structured proof corpus (testimonials, case studies, reviews, screenshots, UGC) tagged by use case, and a 90-day rolling editorial calendar with week, date, channel, content type, topic, persona, owner, and status.

Why it matters in the AI era. Agents are constantly asked to “write a new post” or “repurpose this case study.” Without an inventory, they duplicate, contradict, or miss the best reuse opportunities. Without an editorial calendar, agents publish content that competes with itself. The proof corpus is especially important: when an AI needs to cite proof, it should pull from a curated, permissioned bank, not hallucinate.

How to build it:

1. Inventory everything you have published in the last 24 months. One row per asset with the metadata fields above.
2. Rank each asset by performance (traffic, conversions, citations). Tag the top 20% as “evergreen flagship” and reference them in the Brand Brain.
3. Build the proof corpus. For every testimonial, log name, role, company, quote, date, written permission, and use cases (email, landing page, case study, social). For every case study, log customer, problem, approach, results with numbers, pull quotes, and proof artifacts.
4. Build the editorial calendar as a table, 90 days rolling. One row per planned piece. Refresh weekly.
5. Make the calendar the single source of truth. Every agent, freelancer, and internal writer starts here.

Agent-ready checklist:

• Inventory is a structured database, not a list of links.
• Every top-performing asset is tagged with performance data.
• Proof corpus has permission status per testimonial.
• Editorial calendar covers at least 30 days forward at all times.
• Calendar is accessible to every agent, human or automated.

---

Asset 9: Data, Analytics & Performance Library

What it contains. Tracking stack inventory, event taxonomy (event name, trigger, properties, owner), conversion-event definitions, UTM conventions, attribution model, dashboards with KPI definitions, reporting cadence, CRM hygiene rules, and a top-performers library: a curated set of actually-converting assets (best-performing emails, posts, ads, pages) with their performance data attached so agents can pattern-match on what already works.

Why it matters in the AI era. Agents write best when they can see what has already worked. Without a top-performers library, agents produce work that is on-brand but untested. With one, agents can reference “the email that converted at 47% last quarter” and write variations that are more likely to repeat that performance. This is where AI stops being a content machine and starts being a compounding growth system.

How to build it:

1. Inventory your tracking stack. GA4 or Plausible or Fathom. Ad platforms. CRM. Email platform. Call tracking. Review the last 90 days for gaps and broken events.
2. Write an event taxonomy. Every tracked event gets a name, trigger, properties, and owner. Standardize naming conventions (snake_case or camelCase, not both).
3. Define conversion events. Primary conversion (signup, demo booked, purchase). Secondary conversions (email signup, content download). Assisting conversions (video watch, deep scroll).
4. Standardize UTM conventions. Source, medium, campaign, content, term: what each means, who can create them, who cannot.
5. Build 3-5 dashboards: acquisition, engagement, conversion, revenue, retention. One KPI per dashboard that is the single number you watch.
6. Curate the top-performers library. Every quarter, pull the top 10 pieces (by conversion or engagement) and log them with performance data for agents to reference.

Agent-ready checklist:

• Event taxonomy is a document, not tribal knowledge.
• UTM conventions are enforced by a link-builder, not by memory.
• Dashboards have owners and review cadences.
• Top-performers library is updated quarterly.
• CRM is hygienic enough that an agent can trust it (dedupes, required fields, tagged properly).

---

Audit Every Asset, Not Just Design

The design-system crowd learned this first: if an asset is worth having, it is worth enforcing. A tokens.css file with 230 named colors does nothing if a developer can still write color: #2563EB in a component and ship it. The fix is an audit script in CI that fails the build on any hardcoded value.

The pattern generalizes across M10. Every asset deserves an audit:

• Voice & Terminology (Asset 2): a lint script that flags banned words, unapproved jargon, and terminology drift in any draft before it publishes. Writer.com and Grammarly Business both do this. A shell script over your own TERMINOLOGY.md does it for free.
• Offer & Competitive (Asset 5): a quarterly diff check on your competitor entries. If a competitor’s pricing page changed and your battle card did not, the audit flags it.
• SEO & Keywords (Asset 6): a crawl that finds pages in your keyword clusters with no internal links pointing at them. Orphans are invisible to both Google and GPT.
• Website & Schema (Asset 7): Google’s Rich Results Test on every canonical URL, scheduled monthly. Schema breaks silently when CMS templates change.
• Content Library (Asset 8): a freshness audit that flags any evergreen URL not updated in 12 months and any duplicate topic cluster.
• Design System (Asset 4): the token audit script from Pandya’s bootstrap prompt.

The common shape: a machine-readable asset, a set of rules the asset implies, and a scheduled check that fails loudly when reality and asset disagree. Without audits, your assets rot the quiet way. With audits, the system tells you exactly which file stopped being true.

---

Asset 10: Agent Operating System (The “Brand Brain”)

What it contains. The meta-asset. A master index file that references all 9 other assets and aggregates them into one loadable context. Plus a template library (email, landing page, social post, one-pager, deck, proposal), channel playbooks (LinkedIn, X, email, blog, Quora, Medium) with format rules and voice modulation per channel, campaign recipes (product launch, nurture, re-engagement, SEO sprint), a prompt library with system prompts tuned to assets 1-9, and approval workflow rules (when a human must review).

Why it matters in the AI era. Assets 1-9 are the raw material. Asset 10 is the operating system that makes the raw material usable. The Brand Brain is the single file you paste into every new agent’s context. It references the paths to everything else. It contains the system prompt that tells the agent who to be, what to read first, what to check before producing, and when to stop and ask.

Without a Brand Brain, you re-explain your company to every new tool, every new prompt, every new agent run. With one, you load it once and the agent behaves.

How to build it:

1. Create a BRAND-BRAIN.md file at the root of your docs repo.
2. Include a folder tree showing where every asset (1-9) lives.
3. Write a “load order” section. Which assets load first. Which are read-only. Which are updated by which team.
4. Write a “core system prompt” that aggregates the essential context from every asset into one copy-paste block. This is what goes into Claude Projects, ChatGPT custom contexts, Gemini Gems, and agent workflows.
5. Add the template library. Each template has clearly labeled variables ({{ICP_NAME}}, {{OFFER_NAME}}, {{PROOF_POINT}}) and tells the agent where to fetch them.
6. Add channel playbooks. Each playbook has format rules, cadence, length limits, link-placement rules, voice modulation, and 3-5 example post structures.
7. Add campaign recipes. Multi-touch sequences with timing, assets involved, channels, and decision points.
8. Add the prompt library. Production-ready prompts per content type. Each references the relevant assets via path.
9. Add approval rules. What an agent can publish autonomously, what needs human review, what needs legal review.

Agent-ready checklist:

• BRAND-BRAIN.md exists at the root and links to every other asset.
• Core system prompt is tested in at least 2 AI tools and produces on-brand output.
• Template library has at least 5 templates with labeled variables.
• Prompt library has at least 3 production-ready prompts.
• Approval workflow rules are written, not implicit.

We built a complete starter version of this at github.com/imforza/marketing-assets (companion repo for this post). Fork it, replace our examples with yours, and load it into your agent of choice.

---

Asset Stacking and Priority Order

You do not build all 10 at once. You build them in the right order, matched to your current marketing maturity.

M3 Tier	Where most are stuck	Build these first (in order)
Ad-Hoc (~40% of SMBs)	“I know I should be marketing more”	1. ICP + VoC 2. Brand Voice & Terminology 5. Offer & Positioning
Activated (~35% of SMBs)	“Marketing is happening but scattered”	3. Visual Brand System 4. Design System 6. SEO Strategy 7. Website Architecture
Systematized (~20% of SMBs)	“Marketing works but depends on me”	8. Content Library + Editorial Calendar 9. Data & Performance Library
Architected (~5% of SMBs)	“Marketing runs as an asset”	10. Brand Brain (formalized)

If you only do three assets, do 1 (ICP + VoC), 2 (Voice & Terminology), and 10 (a minimum-viable Brand Brain). Those three alone will triple the quality of your AI output overnight.

One warning. Skipping Layer 1 (Identity assets 1-4) and jumping to Layer 4 (Intelligence asset 10) produces the most common failure mode we see: a Brand Brain built on vibes. The meta-asset only works when the base assets are real. Build in order.

---

Feeding the Agents: How the 10 Assets Become One Brand Brain

Here is the exact folder structure we recommend for the Brand Brain. This is the structure our starter repo uses and what most modern AI coding tools can load directly.

marketing-assets/
├── README.md
├── AUDITS.md                          (audit patterns for each of the 10 assets)
├── 01-icp/
│   ├── ICP.md
│   ├── VOICE-OF-CUSTOMER.md
│   └── BUYER-COMMITTEE.md
├── 02-voice/
│   ├── VOICE.md
│   ├── TERMINOLOGY.md
│   └── COMPLIANCE.md
├── 03-visual/
│   ├── VISUAL-BRAND.md
│   ├── IMAGERY.md
│   ├── AI-IMAGE-PROMPTS.md
│   └── assets/                        (logos, palette swatches, type samples)
├── 04-design-system/
│   ├── DESIGN.md
│   ├── theme.json
│   ├── tokens.css                     (three-layer token indirection)
│   └── specs/                         (foundations, atoms, molecules, organisms, patterns)
├── 05-offer/
│   ├── OFFER.md
│   ├── COMPETITORS.md
│   └── WIN-LOSS.md
├── 06-seo/
│   ├── KEYWORDS.md
│   ├── TOPIC-CLUSTERS.md
│   └── GEO-AEO.md
├── 07-website/
│   ├── SITE-MAP.md
│   └── SCHEMA.md
├── 08-content/
│   ├── CONTENT-LIBRARY.md
│   ├── EDITORIAL-CALENDAR.md
│   └── PROOF-CORPUS.md
├── 09-data/
│   ├── ANALYTICS.md
│   ├── TRACKING-PLAN.md
│   └── TOP-PERFORMERS.md
├── 10-brand-brain/
│   ├── BRAND-BRAIN.md                 (master index + system prompt)
│   ├── templates/                     (email, landing, social, deck)
│   ├── playbooks/                     (linkedin, x, email, blog, quora)
│   ├── campaigns/                     (launch, nurture, re-engagement, seo)
│   └── prompts/                       (blog-writer, email-writer, social-writer)
└── scripts/                           (token audit, voice lint, schema check)

To load this into an AI agent:

• Claude Projects: upload the folder, set the system prompt to reference 10-brand-brain/BRAND-BRAIN.md, and Claude will pull context as needed.
• ChatGPT Projects: create a new Project, upload the key files (compressed or as separate docs), paste BRAND-BRAIN.md into the instructions field.
• Gemini Gems: paste BRAND-BRAIN.md into the Gem instructions, attach the folder as knowledge.
• OpenClaw, Hermes, Manus, or custom agents: mount the folder as the agent’s knowledge base; reference BRAND-BRAIN.md in the system prompt.
• Zapier, Make, or n8n: load the relevant asset files at workflow start via a file-read step, pass into the LLM node’s context.
• Cursor, v0, Lovable, Bolt (for design and dev work): commit to the same repo as your product code. The tools will read theme.json and DESIGN.md automatically.

The goal is that every agent you load, today and in the future, starts with the same foundation. Your Brand Brain is portable across models. The asset structure is the permanent layer. The model you use this year is the replaceable layer.

---

What’s NOT in the Core 10 (and Why)

We get this question every time we walk through M10: “where is the email list? Where is the Google Business Profile? Where are my social accounts?”

We deliberately excluded them. Here is why.

Category	Examples	Why it is not in the Core 10
Distribution channels	Email list, social accounts, ad accounts	These are pipes, not inputs. They distribute the outputs of assets 1-10.
Operational infrastructure	Domain, DNS, hosting, CRM records, payment systems	Necessary but not an input to marketing content.
Discovery assets	Google Business Profile, directory citations, backlink profile	Mostly SEO execution; lives downstream of SEO Strategy (asset 6) and Website Architecture (asset 7).
Legal and compliance artifacts	Terms of service, privacy policy, contracts	Referenced by Compliance (inside asset 2) but not standalone marketing assets.
Lead magnets	Checklists, ebooks, templates, calculators	Products of assets 1-10, not inputs. Produced by agents once the 10 are in place.

The test is simple: if an asset feeds into content production, it is Core 10. If it distributes, measures, or operates the content, it lives downstream of Core 10.

This is the most important distinction in the framework.

---

Frequently Asked Questions

For the quick-answer version of this guide, see our knowledge base article What Marketing Assets Does a Small Business Need for AI?

---

Build the Foundation, or We Will Build It For You

If you have made it this far, you already know the honest answer: you have fragments of these 10 assets but not all 10, and the gaps are what is producing generic AI output across your marketing.

There are three paths from here.

Path 1: Build it yourself. Fork our starter repo at github.com/imforza/marketing-assets. Every file you need is there with prompts, fill-in-the-blank structure, and worked examples. Plan on 60-120 days of focused work.

Path 2: Audit your current foundation. Schedule a free Marketing Asset Audit. In 30 minutes, we score your current assets against M10 and identify the top three gaps holding back your AI output. You walk away with a prioritized build order whether you work with us or not.

Path 3: Have us build it. We build the full M10 stack and the AI marketing system on top, so your foundation is done right and your agents are running from day one. This is what we mean by “we build AI marketing systems for small businesses.” M10 is always the first deliverable. Start the conversation here.

Fix that once, and every tool you buy from this point forward gets a head start.

---