tl;dr: as DESIGN.md is a spec for your brand guidelines and guardrails, MESSAGE.md is a spec for your company's positioning and messaging. Together, let the LLMs go.
Check it out: https://github.com/fortyfivan/message.md
Here we go yo. What's the AI scenario.
Let's assume that from now on, most marketing content will be both written by AI and consumed by AI. Pack it up, it's over.
Well, maybe not. Look at what has happened to the software engineering discipline over the past few years. Coding has largely been "solved", but the software craft remains. We can say the same for the content marketing discipline - writing has largely been "solved", but the narrative craft remains.
But "solved" isn't automatic, especially when it comes to quality and taste. AI is trained on what good looks like, and we all know the result. The leap from good to great is what every craft must be about from now on. This only comes from deep subject matter expertise and system design.
Engineering teams have invested heavily in designing systems that enable a safe, scalable software development lifecycle. GTM teams must do the same. To-date, the GTM Engineering discipline has placed a lot of emphasis on the workflows, but less so on the content.
To serve as GTM Engineering's foundational positioning and messaging context, I'm introducing MESSAGE.md as a spec for AI content factories (so they don't become AI slop factories).
Context Rules Everything Around Me
MESSAGE.md is a layered specification designed for optimal messaging and optimal token use.
It was born out of my work as a Product Marketer in cybersecurity – where every market is a complex, multi-dimensional environment that typically involves multiple categories, products, personas, segments, and competitors. Every day I feel like I'm in a multiverse, but I digress.
The thesis is as follows: your markets are dynamic, your message must be too. I'm not referring to just what your homepage says, rather how your company narrative spreads across the employee base, the field, the market, the customer base, the partner ecosystem, industry analysts, and the community at large. At today's pace, maintaining relevant and personalized messaging across channels is paramount, but a few compounding issues arise with traditional messaging frameworks:
- Messaging frameworks go stale the moment they get written (and they get buried in docs nobody reads)
- Every campaign, launch, and play is a bespoke messaging snowflake (and your marketers and sellers will say so)
- AI with little to no context and/or direction gets it wrong every time (and produce nonsense worse than slop)
I've been working on a Claude-native messaging system to counter these challenges for a while now, and first released an open source edition about 6 months ago. Through that work I landed on a structured design pattern that I believe is transferrable to any GTM AI system. If your positioning and messaging are flat, you can fit everything into one document - this is overkill. For those who operate in multi-dimensional environments like me, this is for you.
I said who's messaging house? Your messaging house!
Messaging frameworks and messaging houses are used interchangeably - I say house. A MESSAGE.md conformant messaging house is the sum of the following elements, which agents load progressively:
- MESSAGE.md: sets the altitude. Your company facts, glossary of terms, brand guardrails, and common scenarios.
- Pillars: tell the story. Foundational messaging documents for the company, narrative, market, audience, portfolio, and proof.
- Collections: hold the details. Granular profiles loaded on demand for specific personas, products, competitors, and more.
- Assets: define the shape. Structured documents for what every asset type looks like for you, with the conventions for automated design, production, and delivery.
The spec is opinionated in its structure and flexible in its contents. While the design is purpose-built as the context layer for agents, the structure of a messaging house is only as good as the contents within. Consider how agents interpret context and use it for research, writing, and production tasks. There's obvious criteria like your ICP and product portfolio, but the real value starts to rise up when you document your unique market perspective and your unique value propositions. The repository includes inline instructions, format guidance, and tips for how to populate.
What makes this system dynamic is the concept of a messaging room – runtime context assembly based on the scenario of a task at hand. Running a play, launching a product, building a campaign, planning a conference, or responding to an analyst are better served with just the right slice of the messaging house and awareness of whether you're talking about a new feature, a competitive takeout, or a vertical-specific outreach. Once the messaging house is populated, set progressive loading rules in your CLAUDE.md or AGENTS.md files so the agents know what context to load and what elements to focus on.
This spec is focused purely on the design of the context layer – you can wrap it however you please to make a system. To keep the positioning and messaging current, I recommend adding periodic research agents to run market and competitive analysis, read live customer and prospect feedback, and analyze calls. And for content factories, it's easy to build any type of content production pipeline based on this context as the source of truth.
It all takes a lot of work, there's no denying that part. Populating the messaging, tuning the assets, building the workflows. But once you do this work, you can place a lot more emphasis on the positioning and messaging craft, while the system carries the heavy lifting across many day-to-day GTM tasks.
Take the spec as-is into your own environment, or you can fork my open source project that includes a lot of the surrounding research and content tasks, as well as guided workflows to bootstrap and tune the system to make it uniquely yours. See: https://github.com/fortyfivan/claude-message to get started.
You down with MESSAGE.md? Yeah, you know me