BARELY SHIPPING Season 1 · Episode 04

Your Best Content Is Trapped in Your Sales Calls

Your competitors can run the same model and the same prompt. What they can't copy is what you feed it. Build one queryable source of truth (a Brand Brain) and the content takes care of itself.

Published June 22, 2026 Runtime 14 min Host Nathan Thompson
▶ Watch on YouTube
Barely Shipping · E04
Listen: YouTube

In this episode

  1. 00:00Back to basics: the centralized source of truth
  2. 00:40The nuggets trapped in your silos (calls, docs, CRM)
  3. 02:00Why no competitor can replicate your conversations
  4. 03:30One client, five systems, none of them talking
  5. 05:00Context is the real constraint, not tools
  6. 06:00What a Brand Brain actually is
  7. 07:00Garbage in, garbage out (and the reverse)
  8. 08:00Keep a human on the ends (the club sandwich)
  9. 09:00Why "AI slop" is the editor's fault, not the model's
  10. 10:00How to build it: Supabase, embeddings, chunking
  11. 11:30What it unlocks across every GTM team
  12. 13:00Don't fix a tool problem with another tool
  13. 13:30Pipes before chocolate, and what's next
Context is the real constraint right now. Nathan Thompson · Barely Shipping E04

Most brands are sitting on a moat they can’t see, because it’s scattered across five tools that don’t talk to each other.

I worked with a client whose sales team used one call recorder and whose CS team used a different one. Sales talked to prospects over email. CS lived in Slack and Intercom. Someone had spun up a Claude project but never loaded the right connectors, so half the team was generating off one knowledge base and half off another, in two different voices. The closed-won and closed-lost deals sat in Salesforce, untouched by any of it. Five systems, and not one of them could read across the others.

That’s the wasted opportunity almost everyone has right now. Every one of those systems is full of information no competitor can ever get their hands on.

Your conversations are the part that can’t be copied

Every sales call has a prospect asking about a feature gap, pushing back on pricing, asking how you stack up against a competitor. Your competitor will never hear that exact conversation, in that exact tone, about your exact product. They might field the same question, but they will never give the same answer, because the answer only lives inside your team’s heads and your scattered tools.

That’s the thing people miss when they panic about everyone having the same AI. Same model, same prompt, same shape of output. What changes everything is what’s actually loaded in: five hundred sales calls, a thousand Slack messages, the internal docs a human actually wrote instead of the AI-generated version that says nothing.

“Context is the real constraint right now.”

Tools are not the constraint. Everyone has the tools. Context is the thing in short supply, and you already own it. It’s just trapped.

What a Brand Brain is

I’ve been calling this a Brand Brain for two years now: every call, every doc, every proof point, your ICPs, your voice rules, your product truths, every win and loss pattern, sitting in one queryable store.

A folder full of files is storage. A Brand Brain is something every person on every team can query and build from. That’s the difference, and it’s the difference between five silos and one system.

Garbage in, garbage out runs both ways

You know the old line, garbage in, garbage out. The useful part is that it runs in reverse.

“AI is essentially a scaling machine. As long as you’re loading it with the right content and context, you can make really strong outputs on the back end.”

Load a system with real, specific, contextual input and the output comes out specific too. AI scales whatever you give it, including mediocrity. So the leverage isn’t in a cleverer prompt. It’s in what you feed the thing.

Keep a human on the ends

None of this is “let the AI run the company.” I always want a human in the loop on the ends, like the bread on a sandwich. Sometimes a club sandwich: strategy at the start where a human decides what’s worth making, a check in the middle to make sure the system stayed on the rails, and an edit at the end before anything publishes.

Which is also why I can’t stand the phrase “AI slop.” It implies the model published itself.

“If you see AI slop, it is not AI’s fault. A lazy editor said ‘good enough’ and hit send.”

The model didn’t choose to publish. A person did. The slop is a process problem, not a technology problem.

How to actually build one

It sounds more tedious going in than it is. Having built it a few times, the order I’d run it in:

  1. Find where your context already lives: call recorders, Slack, Intercom, the CRM, Drive.
  2. Pick one queryable store. Not another tool to “consolidate” your tools. One place everything lands.
  3. Load the easy structured stuff first. Scrape your sitemap into rows and columns. Add case studies and internal docs.
  4. Then load the conversations: sales transcripts, webinar transcripts, YouTube transcripts. This is the part nobody can copy, so it matters most.
  5. Add embeddings and chunk it down so it can search across hundreds of calls instead of choking on them.
  6. Connect a generation layer on top that draws only from this store.

I built mine in Supabase, mostly because the embeddings and chunking let me search across hundreds of calls in a way no single chat tool handles natively. Nobody’s paying me to say that. I just like the tool.

Once that store exists, everything cascades. Sales gets battle cards built from real objections. Marketing gets blog posts, ABM pages, and SEO and AEO content built from real customer language. Every department draws from the same source instead of guessing at “brand voice” off a PDF an agency wrote eighteen months ago. You capture it once, and you reuse it everywhere.

Don’t fix a tool problem with another tool

The instinct under pressure is to go buy something. Resist it.

“If you’re getting more than one tool to consolidate your many tools, you’re just creating a bigger problem.”

That’s not consolidation. That’s one more silo with a login.

Build the pipe, and the chocolate follows. The source of truth comes first. The good content is the easy part once the pipe exists.

This is Episode 4 of Barely Shipping. The full episode is on YouTube, and the free book it’s based on, Pipes Before Chocolate, walks through the whole framework. If you’d rather have the system built with you than build it alone, that’s what I do.

Frequently asked

// These power the FAQ schema, questions and answers stay in sync
What is a Brand Brain?
A Brand Brain is one queryable store that holds everything unique to your company: every sales and CS call, your internal docs, proof points, ICPs, voice rules, product truths, and win/loss patterns. Instead of that context sitting in five disconnected tools, it lives in one place every team can search and build from. It's the difference between storage and a system.
Why do competitors with the same AI tools get different results?
Because the tool isn't the differentiator, the context is. Same model and same prompt produce the same shape of output for everyone. What can't be replicated is what you load in: your actual customer conversations, in your tone, about your exact product. Competitors can buy the same AI. They can't buy your five hundred sales calls.
Is "AI slop" the model's fault?
No. The model never publishes anything on its own. A human decides good enough and hits send. Slop is a process failure, specifically a missing editor, not a technology failure. Keep a human on strategy at the start and on the edit at the end and the slop problem mostly disappears.
How do I start building a source of truth?
Pick one queryable store rather than buying another tool to consolidate your tools. Load the easy structured content first (sitemap, case studies, internal docs), then add your transcripts (sales, webinar, YouTube). Add embeddings so it can search across everything, then connect a generation layer on top that draws only from that store. It takes a day or two to stand up a solid version.
Full transcript click to expand

All right, today we are going back to the basics. This is the fourth episode of Barely Shipping, where we are talking through the importance of a centralized source of truth. Everybody knows this in theory. However, there are very few people doing it in practice. So today we're going to touch on something you probably already know but might need a reminder on, and then some guidance on how you can set these up.

Right now, most brands have all of this scattered information and all of these scattered tools. But what's interesting are all of the nuggets living in those little silos. Sales calls, customer conversations, internal docs. These are all things absolutely unique to you that no competitor can replicate.

Every day, your sales team is taking prospect calls, and every day those conversations are loaded with questions about features that are lacking or confirming that they're there, questions about pricing and how often it comes up, questions about how you compare to competitors and your value on the market. These are all things happening in a conversation your competitors are never going to have. They might get the same questions, but everyone provides a different, nuanced answer depending on the exact product they're selling.

So just in your sales calls, you have loads of transcripts that are inputs that can inform the product team ("this feature request keeps coming up over and over") or the marketing team ("we need more content around these pain points, how our product specifically handles them, we need to educate the market on X, Y, and Z"). It's the same with your customer conversations in Slack or with your CS team, and with your internal docs, assuming they weren't just written by AI. If somebody took the time to write down your mission, vision, values, the benefits and solutions of your product, that strong documentation specific to your brand, these are all inputs that can inform the entire go-to-market motion that no competitor will be able to replicate with AI, even if they're using the same AI tools.

But the only way this works is if it's all stored together. The problem right now is you have calls living in one place. I was working with a customer, and their sales team used one type of call recorder, and their CS team used a different one. The sales team was mostly chatting via email with prospects, and the CS team was talking via Slack and Intercom. You had people going off and creating prompts, maybe in a team space in Claude, but they didn't load up the right connectors, or they're drawing from MCPs or knowledge bases other people aren't, so they're getting a different voice and tone. And then you have closed-lost and closed-won deals happening in a CRM like Salesforce. So you have all of these different places, and no system can reliably read across them, especially at the volume they're coming in. Which is why it's important to find a solution that lets you take all of that and put it in a structured table where you can identify your ICP, how it relates to your product, the feature requests that came up, and have that be queryable across everyone on your team.

So context is the real constraint right now. If your competitors all have the same AI tools, in theory they could have all the same AI outputs. But if you have a central knowledge base or central source of truth, what I call a Brand Brain, if you have that context, then you can make really unique content.

You've heard the expression garbage in, garbage out. The cool thing is it also works in reverse. Great in, great out. Unique in, unique out. AI is essentially a scaling machine, and as long as you're loading it with the right content and context, you can make really strong outputs on the back end.

How much you want a human in the loop to edit all of that is your company's call. I always recommend having a human in the loop, at least as the bread ends. Sometimes a club sandwich: the start of the process on strategy, the middle as a stopgap to make sure everything's going well, and the end to edit before anything gets published.

I've said it before and I'll say it again, AI is not publishing itself. So if you see AI slop, it is not AI's fault. In fact, I hate the term AI slop, because it implies AI was publishing itself, and it's not. A lazy editor said "good enough," hit click, sent it off. But if you have a central knowledge base, then every single person on any department can draw from the same sources and create on-brand content, whether that's a sales follow-up email or a blog post, top, middle, or bottom of the funnel. They can do anything as long as they're loading the right information in, and it is a company's responsibility to provide them with that shared knowledge base.

So we get every call, every doc, every proof point, your ICPs, your voice rules, your product truths, all the win and loss patterns into a queryable store. I personally have been using Supabase. I'm a huge fan because I can make it add embeddings and chunk everything down, so it can search across five hundred sales calls and a thousand Slack messages in a much more granular way than any chat tool can handle on its own, once it's connected with an MCP or a custom proprietary app. That's just my preference. I know some people have other solutions, but I've been a big fan of Supabase, and they're not even paying me to say that, which is crazy.

Once you have that central source of truth, everything cascades. You can create on-brand content with your unique, nuanced perspective for the sales team with battle cards and follow-up cards. You can create blog posts and social pages. You can create AEO and SEO and discoverable answers. You can create ABM pages. You can create all of your success documentation. But everything's drawing from the same inputs, that same queryable central source of truth.

It sounds tedious, but it's a lot easier than you might think. You can scrape your sitemaps and load all of those in as separate columns and rows. The same with your case studies and internal docs. But what you also need to do is contextualize that store with all of your sales transcripts, webinar transcripts, and YouTube video transcripts. Having built it myself a few times, it takes maybe a day or two to get something solid in place. After that, you can connect a chat experience or use a separate tool. I was using Copy.ai for years and still do, where you can draw from that central knowledge base and create strong outputs that are on brand and loaded with your unique perspectives that competitors can never replicate.

So you capture it once, then you reuse it everywhere. All your posts, all your email drafts, all your slide deck pulls. You're never writing anything totally from scratch, and every starting point is something on brand and unique to you that competitors can't replicate.

In the next episode, we're going to talk about how you take that central source of truth and apply it specifically to content marketing: strong top-of-funnel content for SEO and AEO, strong middle-of-funnel content for thought leadership and brand awareness, and bottom-of-funnel content that drives trust and generates pipeline, all drawing from that same central source of truth.

It might sound basic, but it would surprise you how many clients I work with who simply don't know how to get started, or who think the answer is to add more tools to simplify the data, which almost never works. If you're getting more than one tool to consolidate your many tools, you're just creating a bigger problem.

So the expression I like to say is you have to build the pipe, and the chocolate follows. If you don't understand what that metaphor means, head over to the website. There's a book, Pipes Before Chocolate, that runs through this and the entire framework of how the system works together. And if you ever have questions, reach out to me on LinkedIn. I'm building these companies with some surprisingly great success in terms of making unique outputs that drive traffic, but also pipeline, and I'd be excited to chat with you about that.

Wait for the next one. We're going to talk about top of funnel, middle of funnel, bottom of funnel. Visit systemsledgrowth.ai and reach out on LinkedIn if you have any questions.

Barely Shipping

New episode every week. No fluff, by policy.

Short, opinionated episodes on building lean GTM systems with AI. Get each one plus the written version in your inbox.

Subscribe on Substack See how I build this for teams

More from the season

S1 · E03
There Are No Low-Level Tasks
S1 · E02
Nobody Should Be Charging You Extra for AEO
S1 · E01
Your ChatGPT License Is Not a Strategy