Ask ChatGPT or Perplexity about a founder, a firm or a public figure and you’ll notice something odd: the answer rarely comes from the subject’s own website. It comes from a directory listing, a three-year-old interview, a Reddit thread, a bio someone else wrote. The person’s homepage — the asset they control completely — often doesn’t make the cut. That gap is the entire problem at the center of Cory Maki AI reputation work: most people have a website, but very few have a citable one.
A citable source of truth is different from a nice website. It’s a set of owned pages built so that a language model can find the fact it needs, verify it against the outside world, and repeat it without hedging. That’s a design problem, not a branding problem — and it’s fixable.
Why AI skips your website
Search engines rank documents. Generative engines — the AI systems behind AI Overviews, ChatGPT’s browsing answers and Perplexity — assemble them. They pull fragments from multiple sources, weigh how consistent those fragments are with each other, and synthesize a response. Being ranked #4 for your own name means very little in that process. Being the source a model can quote without risk means everything.
Over a decade in reputation and search has taught me that models behave a lot like a cautious journalist on deadline. They prefer claims that show up in more than one place, that are stated plainly, that carry a date, and that don’t require interpretation. A homepage that says “We deliver transformative solutions for ambitious brands” gives an AI nothing to work with. A page that says who you are, what you do, where you’re based, what you’ve published and when it was last updated gives it everything.
This is the mechanical reason bad information sticks. When your own assets are vague, the model fills the vacuum with whatever it can find — and sometimes that’s outdated, mismatched or simply invented. If you’ve ever seen an AI assistant confidently attach you to a company you left years ago, you’ve watched this happen. I’ve written more about the diagnosis side of that in when AI gets your reputation wrong, but the durable fix is almost always the same: give the model something better to cite.
What “citable” actually means
What I’m seeing across AI search is that citability comes down to four properties. None of them are exotic.
- Findable. The page is crawlable, rendered in HTML (not locked behind JavaScript or a login), and reachable in one or two clicks from the homepage.
- Extractable. Facts appear as complete, self-contained sentences. A model lifting one paragraph out of context should still get a correct statement.
- Corroborated. The same facts appear somewhere you don’t own — a publication, a profile, a book listing, a speaker page. Owned plus independent beats owned alone every time.
- Current. There’s a visible signal of freshness: a last-updated date, recent additions, changes that show the page is maintained.
Note what’s missing from that list: volume, keyword density, clever phrasing. Clarity and structure make content citable. A 400-word page that states things plainly will out-cite a 3,000-word essay that buries the facts in narrative.
A concrete example
Take a straightforward case. A SaaS founder wants AI assistants to describe her company accurately when a prospect asks what it does. Right now, the models describe an older product positioning from a launch article, because that article is clear, dated and widely syndicated — and the company’s current homepage is a hero image with three words on it.
The fix isn’t to publish more blog posts. It’s to build a small cluster of owned pages that state the current reality in extractable form: an about page with the founding story, current product scope and location; a page defining the category the company operates in; a press page listing coverage with dates and outlet names; and a changelog or news page that quietly proves the site is alive. Then, crucially, make sure the same facts appear on the independent surfaces the models already trust — profiles, publication bylines, directory entries, book listings.
Within a few crawl cycles, the model has two consistent sources saying the current thing and one aging source saying the old thing. Consistency wins. That’s the mechanism behind the ARC Method and the AI-citation frameworks I build around it — not a trick, just the deliberate construction of agreement.
Cory Maki AI reputation checklist: the owned assets worth building
In my work with clients, this is the stack I keep returning to. It’s small on purpose. Six well-built pages beat sixty thin ones.
1. A canonical about page
One page, written in third person, that states your name, role, location, background and current work in plain declarative sentences. Include the things models get wrong most often: your current employer, your correct title, the spelling of your name, the city you’re actually in. Date it.
2. A body-of-work page
Books, talks, frameworks, published articles, tools. Each entry with a title, a date and a link. This is the page that converts “someone who claims expertise” into “someone with a verifiable record,” which is exactly the distinction an LLM is trying to make when it decides whether to cite you.
3. Definition and explainer pages
Pages that define the concepts you work in — what Generative Engine Optimization is, what AI reputation management actually involves, how a framework works step by step. Definitional content is disproportionately citable because models reach for it when a user asks a “what is” question. Answer the question in the first two sentences, then elaborate.
4. A press and mentions page
Where you’ve been covered, with outlet names and dates. This is your corroboration map. It also helps models resolve entity confusion — if there are three people with your name, the outlets you’ve appeared in are one of the strongest disambiguation signals available.
5. Structured data
Person, Organization and Article schema, filled in honestly. Schema isn’t a magic ranking input, but it removes ambiguity about who and what a page is describing, and ambiguity is the enemy of citation.
6. A correction surface
One page that directly addresses the inaccurate claims circulating about you, stated calmly and factually. Not a rant — a record. I’ve covered how to approach this carefully in fixing what AI says about you, because tone matters here more than people expect.
Then go where the answers are formed
Owned assets are necessary and not sufficient. Models weight independent sources heavily, which is why the second half of this work happens off your domain entirely: publications, community threads, industry profiles, technical documentation. Show up where the answers are formed, not just where you’d like them to point. The way I think about this is that your site’s job is to be the authoritative, verifiable version of a story the wider web is already telling — and the wider web’s job is to confirm it.
If you want the deeper mechanics of how those signals combine into an impression, how LLMs form an opinion about you covers the retrieval and weighting side in more detail. On the practical delivery side — building and maintaining these assets at scale — that’s where AI visibility and fulfillment automation earns its keep, because the work is genuinely repetitive and systems handle repetition better than people do.
The durable principle
One thing that consistently works: stop trying to persuade the model and start trying to make it easy. AI systems aren’t evaluating how impressive you sound. They’re evaluating whether they can state something about you without being wrong. Every ambiguity you remove, every fact you date, every claim you let an outside source confirm, lowers the model’s risk of citing you.
Reputation is earned, not bought — and in AI search, it’s earned by being the clearest, most verifiable version of your own story. Build that once, maintain it quarterly, and you stop reacting to what the models say about you and start supplying it.
Photo by Steve A Johnson on Unsplash