STK   Stack

Social media forced every kind of business to become content producers. Coffee shops, local restaurants and nonprofits all got pushed into a space where a steady stream of good content is simply the cost of competing.

A quiet digital presence reads like a closed shop, and that silence costs a business trust with the people deciding whether to buy from it. Before you get to any of the upside a strong content strategy can bring, just keeping a credible presence takes steady work.

The question for most of these businesses is whether AI can help them compete in a content game they never meant to play.

Doing content well means producing at scale. For a business that isn't a media company, that means publishing every week while also running the actual business. Gary Vaynerchuk argues that feeds now run on interest instead of on who follows you, so reach depends on how good each post is rather than how big your audience is. That lowers the barrier to being seen, since a small account can break through with one strong post, and it puts more pressure on the quality and value of every piece. He still argues for making a lot of them.

However you look at it, doing content well takes a lot of work: finding stories, sharpening ideas, capturing material, editing, polishing the creative and getting it distributed. Every one of those has to happen for content to move the needle, and that's where AI gives a small team a real chance.

That chance comes through a system, and we'll walk you through ours. It was partly inspired by Alex Lieberman's content machine, which he walked through on How I AI, and it now helps small shops, including our own, share their voice and compete on content.

The dividing line is simple to state. The human brings taste, judgment, and instinct. AI takes the work that is tedious or voluminous. Each side does what it is good at.

You've probably seen the version of AI content sold all over YouTube, where you build a system once and an army of agents makes content and money while you sleep. That version isn't real. At best it runs a slop factory, turning out the kind of content that does a business more harm than good. What AI does is save us a large amount of time and make specific parts of the work better: research that used to take hours, first drafts that arrive in minutes, and a publishing rhythm we can actually keep.

Before any AI touches anything, there has to be an editorial strategy, written down. It covers four things:

AI does good work when you know exactly what you're looking for and forgettable work when you hand it a line or two about your business. The strategy is what you're looking for, written down and encoded so AI can work from it. If your answers are still loose, talk them through with AI and let it put scaffolding around them, the same move we described for building a company knowledge base.

To start: answer those four questions on a single page.

With the strategy set, AI can go looking. Give it your source list (the publications, writers, companies, creators, and accounts where your topics get discussed) along with plain instructions on what to filter for and what counts as trustworthy. It scans those sources and comes back with a digest of what's being said and what you might write about.

Your own notes belong on that list too. Jot ideas down as they come up during the week and keep them somewhere the AI can read, so your half-formed thoughts get weighed alongside everything else it finds.

Different tools suit different jobs. We use Grok for sourcing because it works natively with X and runs well as an agent, surfacing ideas and conversations worth joining. Claude handles our strategy work and certain writing tasks, and plugs in on creative.

We used to do this with social listening software. The tools pulled in everything across hashtags and accounts, and then a person had to sift through all of it, which took hours. Humans were the curation layer. Now the AI does the first pass of curation and synthesis, and the sifting takes minutes. Because it holds our editorial strategy as context, it also flags when someone is talking about an idea we've parked, and suggests whether to reply or fold it into a piece we're writing.

The human job at this stage is choosing. The AI ranks the ideas, and we pick the one worth our time.

To start: list a dozen sources you already trust, add a running notes file of your own, and ask AI for a weekly digest written against your strategy.

The blank page is the hardest part of writing. You have a head full of ideas and no order for them. Writers are taught to open paragraphs with a topic sentence, one line that holds the purpose of what the rest of the paragraph will explain. Advice that has stuck with us: don't try to write that sentence first. Start writing, get the ideas out, and the topic sentence usually shows up a few sentences in.

AI takes that problem off the table. You can hand it scattered, half-formed thoughts and count on it to help shape them into what needs to be said.

That's what our interview stage does. Before every issue, we answer five to seven questions out loud, loosely, without worrying about order. We answer with a voice dictation tool instead of typing, which gets the thinking out faster and gives the AI a truer sense of how we actually sound. The AI listens for the signal: the points we made, the ones we spent the most time on, the idea underneath the rambling. Then it helps distill them, and we go back and forth until the kernel we started with becomes a point worth making.

This is the most time-consuming step in the whole system, and it's the one we'd never hand off. Every piece starts from what a person said here.

To start: before your next piece, have AI ask you five questions about it and answer them by voice.

From the interview, the AI writes a first draft in minutes. Without it, that draft would take hours or days, longer if the week gets away from you. It writes from our words and follows written rules about voice, style and the mistakes we've already corrected, so the first version is strong. It still needs an editor.

Read it the way you'd read copy from a staff writer. Check whether it says what you actually mean, mark what it got wrong, and send it back with notes. A recent issue of ours started as an observation about how modern professionals present their skills publicly. The first draft followed our early notes faithfully, and reading it showed us that the point we cared about most, that marketers are becoming the people who build, was sitting near the end. The next version led with it.

Every correction we make gets logged as a lesson, so the next draft gets closer to a polished product.

To start: keep a running file of every correction you make, and hand it to the AI with each new draft.

A finished long piece becomes the source for everything else. By the time the draft is done, the key ideas are already clear: the thesis, plus the handful of points strong enough to stand on their own. Short-form platforms need those pulled out and reshaped. X needs the sharpest single lines. Instagram and TikTok need them as images and video.

Design runs on the same principle as writing. AI writes well against a documented voice, and it designs well against a documented design system with structured data it can read. Without one, you get generic AI visuals or something that looks like another company's brand. With one, a code agent connected to a design tool like Paper or Figma, or an image generator, can turn your key ideas into graphics, simple motion and infographics that look like yours.

The other half is capture. Document the work in every format as it happens, including writing, photos, screen recordings and video, and pour it in raw so the AI has material to cut from.

To start: skip this stage until the earlier ones have run a few cycles. When you get here, write down your design system before you ask AI to design anything.

Content still has to get out the door, and AI helps there as well. It drafts platform-native copy for each post, loads it into a scheduling tool, and reads how posts perform so it can propose changes. It also watches the conversations around your topics and flags where to join in, because a feed that only broadcasts its own links gives nothing back to the people you want to reach.

Distribution opens a bigger conversation than fits here, and we'll come back to it. In our system it's the last stage, and a human still approves what goes out.

To start: have AI draft the social posts for your next piece, then schedule them yourself.

None of this was built in a day, and none of it is finished. The system is a living thing, the same way a knowledge base is. You don't rebuild the architecture every week, but you are always refining it, partly because the AI tools inside it change almost daily. Two years from now this exact setup may be obsolete, with newer models taking on parts that still need us today.

The principles will outlast the tooling. Stories for people still have to be chosen with care and told with craft, especially now that everyone holds a microphone. A good system lets a small team do that at the volume the internet now expects.

It's also the work we do with the brands we work with: building systems like this one, so a small team can publish at a pace that used to take a whole department.