How I'd build an AEO program: a machine per content type

I wrote a whole post already arguing that most early-stage startups shouldn't chase AEO yet. The gist of it, until you've found PMF and a channel that's actually growing the business, organic content is too slow to do anything for you. And too detached from the business to matter anyway.

So this post assumes you're past all that. You know who buys from you and why, you have some channel that's working, and you have source material to pull from - sales calls, customer conversations, a community, events you've run, that kind of thing.

Some background on me. I spent the last year building content programs for companies like Ramp, Lovable, and Webflow, and on the side I ran my own paid community where the content itself was the thing people were buying. I'll get to tactics further down. Tactics rotate every few months anyway; a year ago it was 2h workshops with experts from companies like Webflow and Lovable, and by now it's something else. The two principles I want to start with have held up across every program I've worked on.

New to GEO/AEO? This field guide is a good primer on what it is and how AI search works.

Principle 1: the content has to be as valuable as your product

The typical SEO program, it feels disconnected from the company that's paying for it. The articles are generic, the writing is bad, and none of it compounds with anything else the company is doing. A founder who cares about writing can tell a mile away. The models can tell too, and they're the ones doing most of the reading now.

The bar I use, would someone pay for this?

This past year I was building AI-Led Growth on top of my client work at GrowthX. A paid community, and the content was the product - more than 400 people paid to read it, and it sourced most of our pipeline too. If your content couldn't survive behind a paywall, it won't earn a free read either.

So how do you actually clear that bar? For me it always came back to the customer relationships we'd earned the hard way, the QBRs, the awkward sales calls, the questions that only come up live.

At GrowthX the channel was workshops and dinners, and the best content came straight out of them. Recordings got cut into clips, and the questions people asked us got turned into posts no competitor could write, because they weren't there for any of it. If you're not mining this stuff you're leaving your single biggest advantage on the table, and meanwhile you go write the same generic piece your competitor's agency happens to be writing this week.

Principle 2: keep a mental map of the buyer's journey

This one I'd tattoo on the wall. Before you write a single word, you need a map of the buyer's whole journey, from not knowing they have a problem all the way to picking you. Pretty much everything that matters hangs off of it somewhere.

I lean on it for two things. First, who each piece is for and why. A person who doesn't know my category yet needs a totally different post than a person comparing me against a named competitor, and without the map I'm writing into the void and hoping.

Second, whether the program is working at all. Every piece has an intention, move someone from this spot on the map over to that one, and the intention is what I measure against rather than some vanity number. Take a top-of-funnel post getting 50,000 views while moving nobody toward a decision. It's failing at its actual job, and without the map nobody catches it.

Most buyer journeys collapse down into three zones, and the content looks different in each one:

Aware of the problem. This was SEO's home turf back in the day. HubSpot got a crazy amount of awareness off ranking for "what is a CRM". AI ate most of this zone though, and the articles that still rank rarely give you an opinionated path to a product.

Aware of the problem and the products. This is where most of the money is. Searches like "best CRM for startups" or "alternatives to [competitor]". Whoever's typing those is already out shopping. First Page Sage puts the bottom-to-top-of-funnel conversion ratio somewhere around 25x, which is why comparison and alternatives pages usually end up being the highest ROI pages on a site.

Evaluating your specific product. Stuff like "does Webflow integrate with Stripe?". If you sell through sales teams or procurement, these answers decide whether a buyer closes or quietly disqualifies you.

The last thing the map gives me is progression. A single article converts about as often as a single BDR call does, so, not often. Each piece should know where its reader came from, and where I want them headed next.

HubSpot is still the best model I know here. Nearly every post has a CTA somewhere in it for a template or a calculator or a checklist, and almost none of them just say "sign up". The lead magnet gets your email, the nurture sequence walks you toward the product, and the post you landed on was the opening of that whole ladder.

Hold those two principles still and everything else in this post is just distribution nuance, i.e. how I'd go execute them this quarter. The tools will be different in a year. The principles won't.

Define what good looks like, then build the machine

Say a company handed me a working channel tomorrow and asked me to stand up AEO. Everything would get organized around one unit, the content type.

Every engagement, the work is basically the same on repeat. I need context that captures the important stuff, so the content doesn't drift, stays on brand, sounds like you. And I need a vision. What is this company trying to own, what do they want to be known for? Once I can answer that I have my map. Though the map is usually huge, there's no way to do all of it at once.

Execution is the part AI actually changed. With the map in hand, I start mapping agents and workflows to a content type.

What counts as a "type" depends on the company. For a tool like Ordinal, a listicle is one type, a "product A vs product B" comparison is another one, a use-case guide is another. Every workflow gets exactly one job, which is getting really good at writing its one type. The workflow that's great at listicles is usually bad at an integration guide or a how-to, so they get built separately and prioritized.

Before automating anything though, I define what good looks like. At the start of every engagement I sit down and write the single best version of that content type, sometimes by hand every word, sometimes with AI, doesn't really matter how. What matters is ending up with the absolute best example of a great listicle I can possibly make. That becomes the benchmark, and the agent finally has something to write toward.

Then I go build the machine that can hit it. My setup these days is agentic. I give the agent a handful of skills, and every skill is good at just the one thing, so there's a research skill, a skill for grabbing screenshots, a skill that handles the internal links. The agent itself gets handed the outcome I want rather than some rigid script, and it figures out the rest. One thing I picked up along the way, a listicle covering ten products comes out better when the agent writes one product at a time instead of trying to do all ten in one go.

From there it's iteration, over and over. Is it good yet? No. Okay, make it better. I keep going until the output matches the version I wrote by hand, the one that took forever.

Once a type turns into a machine, volume is close to free, since writing one really good listicle costs about the same as writing ten of them. So I publish a lot. Whatever ends up working gets the extra attention, maybe a video, maybe a contractor who goes in and drops screenshots into everything. The manual effort goes to the stuff that's already paying off.

Same place is where the quality bar lives. Most of these content types don't need some Jane Austen hook, and nobody who's looking up a recipe wants half a page on the author's Christmas memories before they get to the ingredients. They want the answer. For most of these types "good" just means the reader gets what they came for, and fast.

A note on tools

People always ask me which tools. Honestly the tools matter way less than the two principles, and whatever stack I name will have rotated by the time you're reading this anyway.

Still, today's answer looks like this. An AI writing workflow with my real source material loaded in as context, so a draft starts from actual sales calls and customer language and never from a blank prompt. Ahrefs, or a similar SEO tool, for the keyword and SERP research. And some kind of AI-visibility tracker to watch what the models say about you - Scrunch has a solid guide to monitoring, or you could just start with a free CheckThat.ai. Ahrefs has some useful data too if you're curious what AI actually cites.

The pattern that lasts is pretty simple though. Strong source material feeding into production, and a brand that stays consistent across the web. Whatever tool gets you that this quarter is the right tool.

Two reasons to start now instead of later

Reason one is staking your claim while the surface is still open. The first six to twelve months of any new channel is where the disproportionate gains are, and it goes double for a channel with momentum. What you publish today gets indexed and cited, and the models lock it in as their reference for your category. Wait a year and you're muscling against companies that already hold those citations.

Reason two is building the muscle, because the bar keeps rising. A lot of marketing boils down to signaling to a buyer that your product is worth their time. Content used to carry that signal through cost, you needed a writer and an editor and weeks of everyone's time. Now generic writing is cheap and noisy, so the signal moved on to curated editorial with first-hand data, original video, events. The tactics for influencing AI specifically will keep changing. The machinery behind them won't change much at all. You'll still be producing strong content at scale, you'll still be keeping your brand consistent across the web, and you'll still be capturing source material from sales calls and customers. Build all that and the tactic shifts sort themselves out.

I should flag the irony here. I just spent a whole other post arguing against AEO, and I sell this work. Now here I am telling the people who are ready that they should invest more, and invest earlier. The word "ready" is doing all the work in that sentence, because most companies still aren't. If you are though, the two principles are the whole game, and a machine per content type is just how I'd go execute them.

If you want to talk this through for your company, reach out.