By Hooman
The barrier to being the source AI engines cite is no longer money, distribution, or twenty years of domain authority. It's the willingness to put down, in your own actual voice, the things you already know that nobody else does. The reason it isn't already happening at every brand isn't strategic. It's that founders talk all day and almost none of it gets caught.
There's a moment, somewhere in the middle of a call with a founder, when you realize the article already exists. He's explaining a decision he made about the product. Why one version is different from another, what he had to give up to get it right, the reason most people in his industry wouldn't have made the same call. He's not pitching. He's not writing. He's talking the way you talk to someone who actually wants to know.
And the whole time he's talking, I'm thinking: this is the piece. This is the thing nobody else on the internet can write, because nobody else made this trade-off, at 11pm on a Tuesday, with these specific constraints. He didn't sit down to produce it. He's just doing what he does every day. The article isn't waiting to be written. It's being said, right now, in the most natural register a person has, which is the one they use when they're not trying to sound like anything.
That's the part most people miss. The substance is already being produced. Constantly. In calls, in voice notes, in hallway conversations, in the back-and-forth with a co-founder over text. The problem has never been that founders don't have anything to say. The problem is that the place where it gets said has historically been the wrong shape to reach anyone. It vanishes the moment it leaves your mouth. There's no copywriter standing next to you with a notebook, and there's no time later to sit down and try to remember.
That part is what's changed.
Open your analytics any morning right now and you'll see a pattern that didn't exist three years ago. A meaningful share of your traffic isn't from cities you'd expect, and it isn't going to buy anything from you, ever. It's from data centers. Sometimes more machine visits than human ones in a single day.
Those are the crawlers from the major AI systems, indexing the web in a way the old search engines never did. They're not measuring keyword density. They're building a model of what your brand is, what it offers, and whose question your page is the best answer to.
This is the new top of funnel, and it doesn't behave like the old one.
The old top of funnel was a person typing two or three words into a box. The new one is a person typing a paragraph, something like "I'm looking for a manufacturer in North America that can do small-batch CNC work with food-safe finishes and won't outsource to overseas suppliers," into a system that will give them one answer, maybe three, and a citation. They won't scroll through ten blue links. They'll read the answer. And here's the part that matters: people trust those curated answers more than they ever trusted Google's listings. If you're not in it, you don't exist for that customer. If you are in it, you've been pre-vetted in a way Google never pre-vetted anyone.
Two things follow from this that most teams haven't internalized yet.
The first: SEO as it was practiced for the last fifteen years (keyword stuffing, content farms, the same article reworded by twenty agencies) is actively being penalized. The models can tell. They've been trained on enough generic content to recognize it on sight, and they treat near-duplicate material as evidence that your page isn't a source, it's a copy.
The second: the thing they reward instead is specificity. Not specificity in the SEO sense. Specificity in the human sense. We chose this supplier because of how their quality control catches defects ours used to miss. We redesigned the housing three times because the first version flexed under load. We pay more for this component because the cheaper version fails in cold weather and we sell to people who work outside.
This is the kind of detail that used to be considered too in-the-weeds to put on a website. Now it's the single thing that makes the difference between being cited and being ignored.
Almost everything on the internet right now is a paraphrase of something else on the internet. News articles rephrasing the wire story. Brand blogs rewriting what a competitor wrote. Listicles citing listicles. Open five tabs on the same topic and you'll see the same six sentences in different orders.
The AI engines have read all of it. They know what derivative looks like. And they've gotten very good at finding the one source that wasn't.
If you're a founder, this is the most generous market opening in years, and almost nobody is treating it as one. The work is not to find more time to write. It's to change what counts as writing.
A founder we work with takes voice notes constantly. He sends them to his co-founder over text. Two or three a day. Sometimes a minute long, sometimes fifteen. Why he just switched a supplier. What happened with a production run. Why the new packaging matters. The thing the warehouse manager noticed about returns last week.
Every one of those voice notes is, in its raw form, more valuable than anything an SEO agency could produce for him. It's first-person, specific, ungeneric, full of named decisions and actual reasoning. It's exactly the kind of substance the citation engines are looking for.
The reason it doesn't end up as content is the gap between the voice note and the published page. Someone had to transcribe it. Someone had to write the article. Someone had to format it. Someone had to load it into the CMS. Someone had to push it live. By the time it got there, the founder had already had three more conversations worth more than the one being published.
The gap between the voice note and the live URL is what just collapsed.
Not because AI can write articles. It's been able to do that badly for years. The shift is quieter and underneath it. The new generation of AI agents can take an instruction in natural language and act on the actual systems your business runs on. Read your CMS. Write to it. Update existing pages. Cross-reference what you've already published. Push something live, formatted to your design system, without you opening the admin panel.
The implication is concrete. A founder can talk into their phone for fifteen minutes about a decision they made (the kind of thing they'd normally voice-note to a friend) and the words that come out the other end aren't a transcript. They're a piece of writing, in the founder's voice, structured for how AI engines and humans read now, on a live URL by the time the call is over.
One important thing this isn't: a system that publishes everything you say. Half of what comes out of a founder's mouth in a day is private, half-formed, or not for public consumption. The piece that matters is the filter sitting in the middle. A taste layer. Something that knows what your brand sounds like, what's worth keeping, what's an article and what's just a thought you had on the way to lunch. We build that filter as part of the system. It's a whole other piece in itself, and we'll write it next.
And the part that genuinely matters: when something changes six months from now (a supplier, a product spec, a position you held that you've since updated), the same system can update every article that referenced the old version, in your voice, with the new context, in five minutes. The thing that historically made content publishing a debt, every article you publish is a future maintenance burden, stops being a debt.
This isn't a product. It's a way of working. The reason it has to be built per-company is that the value sits in the specificity. Your CMS. Your design system. Your taxonomy. Your voice. The decisions you actually make. A generic version of this would produce generic output, and we'd be back where we started.
There's a smaller observation underneath the bigger one, and it's worth saying out loud because it explains why this works for the people it works for.
The founders who are best positioned to win in AI citation are, almost without exception, the ones who hate the existing version of "content marketing." They find it embarrassing. They can tell that a blog written by a hired SEO writer sounds nothing like them, and they refuse to put their name on it. They've spent years feeling like the price of visibility is sounding like a fraud.
What's actually happened is the price has changed. The thing being rewarded now is the exact thing those founders have been refusing to give up: their own voice, their own perspective, their own taste. The agencies who sold them on generic content for the last decade were selling them the wrong product. The new product is the one they were already producing for free, in voice notes to their team, every single day.
The work we do is building the system that catches those notes and turns them into something the world can find. The pieces of that system have proper names (design system, taxonomy, agentic layer, maintenance pass) and each of them deserves its own piece. We'll write those too. For now, the shape is this: catch the conversation, run it through a filter that knows your taste, publish it as something that looks like your brand and not like a template, and keep it current as the company evolves.
We do this slowly, per company, because it has to be specific to be valuable. The same way the content does.
The bots in your analytics this morning are not a problem to be solved. They're the audience writing the next chapter of how your brand gets found, and they're hungry for something almost nobody is giving them: the thing you would have said anyway, on a Tuesday, to someone you respected, if they'd asked.
Volume has never been the point. The old SEO playbook rewarded it because the algorithm couldn't tell the difference between five good articles and fifty mediocre ones. It just counted. The new systems can tell. They've read everything, and they've developed something closer to judgment. What they reward is originality. Not originality in the artistic sense, not novelty for its own sake, but the thing you could only say because you actually lived it. If you had five ideas today that are worth saying, publish five things today. If you have nothing worth saying for three weeks, publish nothing. The rhythm follows the thinking, not a content calendar. The practical implication is that the volume question is actually the wrong frame. The right question is: do I have something real to say? If yes, say it. If no, wait until you do. A founder who publishes one thing every two weeks that contains actual reasoning will outperform a brand that publishes four times a week out of discipline. The models are not impressed by discipline. They're looking for the thing that can't be replicated by a content farm.
This depends entirely on what you put in. There's a principle in machine learning called garbage in, garbage out. The same principle applies here, just in the other direction. If your input is a two-sentence brief, you get a generic article. If your input is a fifteen-minute voice note where you're working through an actual decision, talking the way you talk when you're not trying to sound like anything, what comes out the other end reflects that. The model has something real to work with. The reason long-form voice input specifically matters here is that it's hard to be generic for fifteen minutes. You run out of generalities. You start naming things. The supplier. The version that failed. The Tuesday it happened. That specificity is exactly what the system needs to preserve your voice, and it's also exactly what the AI citation engines are trained to look for. The sanding-down risk is real, but it's a risk of the brief, not of the tool. Give the system the full conversation and the full context, and you get something that sounds like you. Give it a prompt, and you get a prompt response.
The difference is the same as the difference between a general-purpose hospital and a doctor who knows your whole history. When you open ChatGPT and type a prompt, the model has no idea who you are. It doesn't know your design system, your taxonomy, your past articles, what positions you've already taken publicly, what your voice sounds like at its best. It's working from zero every time. The output reflects that. It sounds like a well-informed person who just met you, which is fine for some things and completely wrong for this. What we build is different at the architecture level. The system is connected to your actual CMS. It knows your article history. It has your voice embedded as a set of operating instructions, not just a style guide sitting in a prompt, but a persistent layer that shapes every output. When you talk for fifteen minutes about a decision you made, it isn't just transcribing and formatting. It's cross-referencing what you've already published, checking for consistency, placing the new piece in the context of everything your brand has said before, and writing to a structure that your design system already knows how to render. The closest analogy in our world is the difference between hiring a freelancer from a job board and working with a studio that's been inside your business for a year. Both can write. Only one of them already knows what you'd never say.
The short answer is: they've been trained to find the source that can't be reproduced anywhere else. At the technical level, the large language models that power AI search (the systems behind ChatGPT's browsing mode, Perplexity, Google AI Overviews, and others) don't rank pages the way traditional search engines do. They're not counting backlinks or measuring domain authority in the traditional sense. They're assessing whether a piece of content represents a genuine primary source, something that contains knowledge which only exists in this form, on this page, because someone who actually did the thing wrote it down. The signal they're trained to weight heavily is specificity that can't be derived from other sources. Generic claims get treated as low-value noise because every page on the internet says the same thing. Named decisions, specific trade-offs, first-person reasoning about why one option was chosen over another, these read as evidence of actual knowledge. The model treats them as source material worth surfacing. The implication for a founder is counterintuitive. The things you'd normally consider too niche or too in-the-weeds for a public article, the reason you switched a supplier, the spec that kept failing, the constraint that shaped a product decision, are precisely the things the engines are looking for. The professional-sounding brand language you spent years refining is, from the model's perspective, a signal that you might just be paraphrasing someone else.
The test is simple, even if the judgment isn't: would you bring this up with a friend you respect, unprompted, because you think it would interest them? Not to impress them. Not to sell them anything. Just because the thought is interesting enough that you'd want to share it with someone who's thinking about similar things. That's the bar. Anything that clears it is a candidate for publishing. Anything that doesn't, leave it in the voice note. The deeper principle is that every founder is already producing this material constantly. The calls. The team conversations. The reasoning behind a decision that you explained to your co-founder and then never wrote down anywhere. Most of what a thoughtful person thinks about in a given week would be genuinely interesting to someone else in the world who's working on the same category of problem. The gap isn't in having things to say. It's in recognizing which things are worth capturing. One practical heuristic: if you found yourself explaining something twice in the same week, that's the article. If multiple people asked you the same question and you gave them the same answer, that's the article. The questions your customers are asking that you keep answering one-on-one are exactly what the AI citation engines are also being asked. If your answer already lives on a page somewhere, you get cited. If it doesn't, someone else does.
Not at all, and the entry point looks different depending on where you're starting from. If you're beginning from scratch, the system gets to know your voice without any prior material to reconcile. Every article that goes live is yours from the first one, built on what you actually think, not on what a content brief said your brand should sound like. If you've already been publishing, the starting point is what you have. The system can read your existing articles, understand the positions you've taken, and work from there. If your views on something have shifted since you wrote it, you can talk through where you are now and the piece gets updated to reflect that. Next year your thinking shifts again, you update it again. Every time you do, you're signalling something true: that this isn't a static document sitting on a server somewhere, it's a living record of how you actually think about a topic, maintained by the person who's still thinking about it. This matters to the AI engines for the same reason it matters to a reader. A page that's been revisited and updated carries different weight than one that was published and abandoned. The signal isn't just that the content exists. It's that someone is still paying attention to it.
This is one of the underappreciated advantages of building the system properly the first time. Because the agentic layer is connected to your CMS and knows your full article history, updating the record isn't a manual process. If a supplier changes, a product spec gets revised, or a position you held publicly has evolved, the system can go through every article that referenced the old version and surface what needs updating. You give it an instruction in plain language: "We changed our packaging supplier in March and the new one uses recycled materials. Update anything that references the old one." It reads everything, finds every instance, drafts the updated language in your voice, and brings it back to you for review. The whole pass happens in minutes. The reason this matters more than it sounds is that published content has historically been a compounding liability. Every article you put live is a future maintenance burden. Details change and the page doesn't, so over time your public record drifts from reality. The system inverts that. The same infrastructure that makes publishing faster makes keeping it current easier. The debt disappears. What doesn't disappear is the human judgment at the end of the process. The system surfaces what changed and drafts the revision. You decide whether the draft is right. That step stays with you, because taste can't be delegated.
Because this isn't content marketing. Content marketing, in the form most people encountered it, was about saying things you calculated might sell. Finding the keyword. Writing the article the algorithm wanted. Putting your name on copy that sounds nothing like you because the goal was reach, not truth. The founders who found this unbearable weren't wrong to find it unbearable. They were responding accurately to something that asked them to perform a version of themselves that didn't exist. What's changed is what the engines actually want now. The thing being rewarded is authenticity, not as a buzzword, but in the literal sense. Content that could only have been produced by someone who actually did the thing. Voice that is consistent across years of output because it belongs to a real person, not a brand brief. Specificity that comes from lived experience rather than research. The frame isn't marketing. It's finding your people. Say the things you actually think, in the way you actually think them, and the people who think the same way will find you. That's a better starting point for a working relationship than any funnel.