I’ve seen an increase in output correlate directly with an increase in lead pipeline in recent months. That’s the story I’m going to explore in this post via one case study.
There’s a quality component to this story, I must note. Hyperscaling slop is not a viable lead generation plan.
Increasing content velocity without compromising quality is a systemic challenge. In this article, I’ll explain how we set up a content system that did exactly that.
TL;DR
- Speed matters less than quality and consistency.
- AI attribution is messy but workable with proxies.
- Strong context layers separate good AI content from bad.
Note: this article is adapted from a longer guide: “How to influence what AI is saying about your brand.”
Speed is the outcome but not the goal
This tends to get lost in the greater AI discourse, amid all of the hype and hyperscalers. Yes, the speed of AI is incredible. No, speed is not the goal here.
Efficiency and effectiveness in content production requires more than speed: it requires publishing high-quality content on strategic topics and doing so in a strategic way. This is categorically different from the goal of publishing content at scale.
I mean, you totally can publish every single variation of every keyword that you can possibly find – it's actually one of the easier things to do with AI – but I don’t recommend it. What you really want is velocity, quality, and consistency.
Example: an efficient (and speedy) content system
Below I have a content velocity report. It shows, month by month, how much a client is publishing. Different colors represent different types of content (i.e. traditional blogs, LLM-assisted blogs, updated pages, etc.)
This is data from a client engagement. This brand had been publishing traditional, human-made blog content. (That’s the red stuff in the bar chart.) Publishing everything by hand is hard, as we all know. So this company was publishing, typically, 4 to 6 posts per month.
Their content system went live in September 2025 with some AI-supported motions. Right away we started publishing programmatic, or batched-out, campaigns using LLM assistance and a context layer that pulls from the brand’s unique product positioning, documentation, and style guides.
The yellow parts of the bar charts represent programmatic content. In that first month their content output doubled.
We phased in other types of content, too: different LLM-assisted pieces as well as (manual) page updates. By January their output was at more than 300% of its original monthly volume.
Every one of those published pieces had strong fundamentals: addressing specific customer questions, integrating the most current product positioning, conveying it all in the brand’s idiosyncratic voice. The quality of the AI-assisted stuff is good. It’s not incredible but it’s doing strategic work, and we’re publishing it at a competitive rate. And the results speak for themselves…
Content velocity drives traffic and pipeline
One reason why I’m big on content velocity: it tracks pretty directly with inbound traffic. When velocity increases, inbound traffic does, too.
There's no paid ad traffic in the chart above. This is all organic traffic.
The traffic volume in this example includes labeled traffic as well as direct traffic. Traffic from AI platforms like Claude and ChatGPT typically shows up as “direct traffic.” Most of the time ChatGPT or Claude will not give you any way of knowing that it's coming from AI. So this is a good thing to test.
And this correlates also with pipeline. The chart below shows traffic from leads who are in our pipeline. 👇
People really are searching for these things. And, if you can get visible in LLMs, you really can drive additional pipeline opportunities for folks coming from there.
What I’m saying is: velocity drives traffic and this traffic drives pipeline.
Attribution is messy in AI but you can make it work
Marketers are constantly expected to provide granular data that proves their successes. We have to re-position this conversation at the right level internally to focus on the system on a holistic level first.
Because of the direct traffic issue, there’s a gap between AI visibility and pipeline. In other words: you might know that one of your converted sales clicked over from ChatGPT but you don’t know what ChatGPT told them before they clicked over to your site.
You can, however, infer some things by putting direct traffic data and conversion data on a matrix. Here are the four positions:
- No AI visibility and no revenue from AI traffic. This is the worst case scenario, of course. Something is fundamentally wrong with your content system.
- High visibility but no revenue from AI traffic. Then we’re probably talking about the wrong topics. You have a real depth on the wrong topic.
- High revenue but low AI visibility. Content influence on the pipeline is great, but you're not showing up. That's a huge opportunity to figure out how we can.
- High revenue and high visibility. This is the goal. Whatever you’re doing is working.
There are other proxy indicators, too, such as comments on a post that you published. If that kind of engagement is coming from people in your target audience then you can say, “Here’s evidence that we're reaching the right people. We know it's good stuff and we can see that pipeline is rising. So we know the system is working.”
What does it look like to build a content system?
This diagram shows the architecture that I use to build content systems for clients.
Analytics are the foundation. They tell us about how the operations are working (and if they’re working at all). Strategy sits on top of analytics. We have to have a conversation about what topics actually matter to us, and which ones don't, and what we’re going to prioritize.
On top of that strategy we have our brand assets: messaging, positioning, goals, priorities. All of that stuff kind of comes together to form the context that you really need for the system to work.
Sitting on all of those baseline layers are a variety of different content formats:
- Editorial content. Written by human writers (who might use AI now and then, as it suits them).
- Engineered content. AI-generated content that is published for strategic topics. (This is great for news-related topics and anything that needs to be addressed on the fly.)
- Batch content and plays. Using AI for specific maintenance tasks and experiments. Refreshing links throughout the blog, for example, or generating content based on sales call transcripts.
Even with the most complex content systems a good old fashioned content calendar is still very necessary. This is what it looks like when you’re managing traditional content alongside AI-assisted variations.
Note: AI does not make human bottlenecks go away! It just doesn't. It lets you produce more, but then we have to review more and think more. That’s why hyperscaling content doesn’t work.
The next frontier: run your content system out of GitHub
A lot of companies we're working with that are technical, we work with them to build that in GitHub. It’s a tremendous resource that requires its own management cadence.
Someone needs to be in charge of answering questions like: What does the workflow look like? Where are our brand assets? Are we keeping them up to date?
Once the build and management are set up, however, the potential is huge. You can run any variety of content generation plays. You can order AI agents to actually publish content in your CMS for you. (Claude is quite good at that now.)
And, most importantly, when you manage all of this in GitHub you’re able to really treat content as code. You’re able to continuously update your context layers so that anyone using these workflows will have the freshest possible results.
I think the difference between junk LLM content and LLM content that humans respond to and engage with is the quality of your context: how, how much unique data you have the ability to inject. Customer comments, quotes, case studies, material from sales calls, all of that is incredibly important. It's worth the pain of setup.
What to do next
I really think it's a moment for marketers to step back and say, “We can build this ourselves. Our context layer, our content systems, we can build them.”
Of course, it can be daunting at the start. So here are some small steps you can take toward AI enablement.
- Look up your pricing on Claude. See what Claude says about it. Where does it get information? That's a really illustrative exercise
- Start enabling Claude like you would a sales rep. Is there one thing we can do to help enable Claude? Even just one? Claude committed this error, how can we fix it?
- Build some basic functions with Claude Code. Work in GitHub, if you know how. And if you don’t know how then now is a great time to learn.
Try it out. Start by tinkering with Claude Code to whatever degree you’re comfortable. That might mean simply opening the interface and typing some commands at first. These tools give marketers more creative control and possibilities than ever before. All we have to do is seize it.
The system is the strategy
The bigger lesson here isn't about AI or content velocity in isolation. It's that marketers who build systems stop making individual bets and start compounding them. Quality context in, quality output out. Consistently, at scale, across every topic that matters to your buyers.
The easiest place to start is also the most instructive: look up your pricing on Claude right now. Whatever comes back tells you something about your content gaps. From there, the next step is figuring out where the gaps are and publishing content that fills them. Our guides to content velocity and content system basics are good places to start.
