How Do I Become Cited by AI?

Being mentioned and being cited are different problems. Almost everyone is working on the first one.

There is a version of AI visibility that is genuinely easy. Get indexed, keep your pages crawlable, publish regularly, and eventually a model will say your name when someone asks a question narrow enough that you are the only plausible answer.

Then there is the thing people actually want, which is for a system to reach for you when the question was not about you. To name you as the source, in an answer where three or four companies could have gone in that slot.

That second one is not more of the first. It runs on a different mechanism, and most of the work aimed at it is aimed at the wrong layer.

Mentioned is about presence. Cited is about risk.

A generated answer is a series of small bets. Every claim in it is something the system is prepared to state to a stranger who may act on it.

Presence gets you considered. What gets you through is whether stating something about you is low risk, and low risk means one thing: the claim holds up across sources that did not come from you.

This is the distinction I call the corroboration gap. The distance between having been covered and having been confirmed. Coverage is volume. Corroboration is agreement between independent parties. Systems weigh the second and mostly ignore the first.

The audit almost nobody runs on themselves

List every place on the internet that describes what your company does. Now cross out every one you control or filled in yourself. Your site. Your LinkedIn. Your Substack. Your Medium. Your GitHub. The database entry you submitted. The directory listing you paid for. The guest post you wrote.

Whatever is left is your actual corroboration surface.

Ten copies of your own claim are still one witness.

For most companies, including well funded ones with real customers, what is left is nothing. And that is the entire explanation for why a company can do everything right and still not get cited. It is not that the system does not know about them. It is that everything it knows traces back to a single origin, and a single origin is a claim rather than a fact.

What becomes evidence, and what never does

Not all third-party mentions are the same weight, and the difference is not prestige.

What becomes evidence. Someone who is not you describing what you do, in their own words, on a surface you do not own, in a way that agrees with how you describe yourself. A journalist's paraphrase. A researcher citing your method. A customer explaining what they used and why. A conference listing you as a speaker with your own one-line description. Another practitioner disagreeing with you by name, which is corroboration too, because disagreement confirms the entity even while contesting the claim.

What never does. Syndicated copies of your own text. Aggregators that scraped your site. Directory entries generated from your metadata. Press releases you wrote, distributed on a wire. Sponsored placements that reprint your language verbatim. All of these increase volume and none of them add a second witness, because the words are still yours.

That last category is where most budget goes, which is why most budget produces nothing here.

The article that changed nothing

Suppose you do get real coverage. A publication with a real editor writes about you. Nothing happens.

This is common enough to have a shape. The article describes you as an AI-driven platform. Your homepage calls you a marketplace. Your founder's LinkedIn says consultancy. The system now has a genuine independent source that disagrees with your own material, and disagreement between sources does not produce confidence. It produces hedging, or it produces the competitor whose story does not have a seam in it.

The work of converting a placement into something a model can actually use is separate from getting the placement, and it is the part that gets skipped. One identical self-description everywhere you control, so the article has something to agree with. Reliable signals tying the piece to the entity. Claims scoped tightly enough to be checked. All of it reflected in the machine-readable layer, not only in the visible copy.

I call that placement activation. Without it, you paid for a contradiction.

Give them something worth citing

There is a supply-side half to this, and it is more under your control than the rest.

Systems cite text that resolves a question. Not text that describes a company. If everything you publish is about how good you are, there is no sentence in it that answers anything, and nothing to lift.

What gets lifted, in my experience of watching which of my own pages come back:

Definitions. A term with a stated meaning, a date, and an author is a citable object. It has edges. It can be quoted without the system taking on your marketing.

Methods someone else could run. A protocol with steps beats a claim about outcomes. The protocol can be attributed. The claim cannot be verified.

Numbers with their conditions attached. "Conversion rose 30 percent" is not citable, because the reader cannot tell in what market, over what period, from what base. The same number with those conditions stated is a fact, and facts travel.

Reasoning you published before you knew the outcome. What you tested, what broke, what you concluded. A public decision record is the rare artifact that gets cited when someone asks who understands a market, because it is evidence of judgment rather than an assertion of it.

What failed. Almost nobody publishes this, which is exactly why it gets picked up. A negative result is the one kind of claim nobody has an incentive to fabricate, so it reads as observation rather than positioning. It also fills a real gap: ask a model what does not work in a category and watch how thin the answer is. Whoever writes that down first owns the answer.

What this actually takes, in time

The honest version, because the dishonest version is everywhere.

Your own consistency is days of work and it is a prerequisite, not a strategy. Do it first, and do not confuse having done it with progress on citation.

Corroboration moves in months, and it moves through other people. There is no version you can execute alone. You can make yourself easier to write about, easier to quote accurately, easier to describe correctly on the first try. You cannot make anyone do it.

Anything promising you citations inside a quarter is selling you syndication and calling it corroboration. The tell is whether the deliverable is a number of placements. Placements are an input. Nobody sells you the output, because nobody controls it.

Four questions

You do not become citable by being described more often. You become citable when the description stops depending on you.

AI does not cite confidence. It cites corroboration.