I argued in the last post that most early-stage startups shouldn't chase AEO yet. Until you've found PMF and a channel already growing the business, organic content is too slow and too detached to matter.
Say you're past that. You know who buys and why, one channel works, and you have sales calls, customer conversations, a community, or events to pull from.
I spent the last year building content programs for companies like Ramp, Lovable, and Webflow. I also ran a paid community where the content itself was the product. Across those programs, two principles stayed the same.
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
Most SEO programs start from a keyword brief instead of the company paying for them. That cuts the articles off from the QBRs, sales calls, customer questions, workshops, and events that contain the material a competitor cannot copy.
The bar I use, would someone pay for this?
At GrowthX, I built AI-Led Growth alongside my client work. 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.
The source material came from customer relationships we'd earned the hard way: QBRs, awkward sales calls, and questions that only came up live.
At GrowthX, workshops and dinners produced our best content. We cut the recordings into clips and turned the questions people asked into posts no competitor could write because they hadn't heard those conversations.
Principle 2: keep a mental map of the buyer's journey
Before I write, I map the buyer's journey from not knowing they have a problem to choosing a product.
I use the map first to decide who each piece is for. A person who doesn't know the category needs a different article from someone comparing named products.
I also use it to decide whether the program is working. Every piece should move someone from one point on the map to another. A top-of-funnel post can get 50,000 views and still fail if nobody moves toward a decision.
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 from ranking for "what is a CRM." AI now answers a lot of these queries before the user reaches a website, and the articles that still rank rarely give the reader 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 shopping. That intent is why comparison and alternatives pages usually become 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.
A single article converts about as often as a single BDR call does, so each piece needs a next step.
HubSpot is still the best model I know. Nearly every post offers a template, calculator, or checklist instead of jumping straight to "sign up." The lead magnet gets the email, and the nurture sequence moves the reader toward the product.
Write the benchmark, then build one machine per content type
Say a company handed me a working channel tomorrow and asked me to stand up AEO. I would organize the program by content type.
Each content type needs source material that keeps the writing grounded in the company, plus a clear answer to what the company wants to own. The buyer map is usually too large to cover at once, so I choose one content type and build around it.
AI changed the execution. With the map in hand, I assign one agent workflow to each 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 I automate a content type, I write the best version I can. Sometimes I write every word by hand and sometimes I use AI. The process matters less than ending with a benchmark the agent can write toward.
Then I give the agent a research skill, a screenshot skill, and an internal-linking skill. The agent gets the outcome instead of a rigid script and figures out the steps. 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.
I keep iterating until the output matches the benchmark.
Pick the type
Out of what the company wants to own: a listicle, a product-vs-product page, a use-case guide. Each one is a type.
Define what good looks like
Write the single best version of that type by hand. That benchmark is the goal you point the agent at.
Build the machine
An agent with skills for research, screenshots, and internal linking, handed the outcome instead of a rigid script.
Iterate to the bar
Is it good yet? No? Make it better. Keep going until the output matches the version you made by hand.
Scale, then double down
Once it's a machine, one great listicle costs about the same as ten. Publish a lot, then put video and screenshots behind the winners.
Once a content type has a working machine, writing one good listicle costs about the same as writing ten. I publish more, then put manual effort into the pieces already producing results: a video, better screenshots, or a contractor improving the pages.
The benchmark should match the job of the content type. A recipe doesn't need a Jane Austen hook or half a page about the author's Christmas memories. The reader wants the answer quickly.
A note on tools
People always ask which tools I use. The stack changes faster than the two principles.
Today's stack starts with an AI writing workflow loaded with real sales calls and customer language. I use Ahrefs, or a similar SEO tool, for keyword and SERP research. I also use an AI-visibility tracker to watch what the models say: Scrunch has a solid guide to monitoring, or you could start with a free CheckThat.ai. Ahrefs has some useful data too if you're curious what AI actually cites.
Strong source material has to feed production, and the same brand language has to stay consistent across the web. Use whichever tools preserve those two things this quarter.
Two reasons to start now instead of later
The first reason is timing. The first six to twelve months of a new channel often produce disproportionate gains because fewer companies hold the citations. What you publish now gets indexed and cited. A year from now, you will be competing with companies already inside those answers.
The second reason is that the quality bar is rising. Generic writing is cheap now, so a writer and editor no longer signal effort on their own. First-hand data, original video, and events do. The tactics for influencing AI will keep changing, but you will still need to capture source material, publish strong content at scale, and keep the brand consistent across the web.
I sell this work, and I still think most companies should wait. The companies that are ready already know who buys, have a channel producing source material, and can name the first content type they want to automate.
If you want to talk this through for your company, reach out.
Get the next post in your inbox
Companies with unfair distribution win. I'm collecting field notes from the world's best operators to help you think through growth.
Join top operators from Zapier, Clay, Lovable.