How to Win AI Citations When Your Category Does Not Have a Name Yet

Every AEO playbook assumes the category already exists. Pick the query, study who gets cited, write the better answer, earn the slot. That works when a buyer types "best contract management software" and an engine returns a shortlist. It falls apart when your product does something no one has a word for yet. There is no query, no shortlist, no incumbent to displace, and no training data teaching the model what bucket you belong in. You are not fighting for a slot in an existing answer. You are trying to make the answer exist. This is how to earn AI citations before your category has a name.

Why a nameless category breaks standard AEO

AI engines do not recommend products in a vacuum. They recommend within a category, because a category is how the model narrows a universe of entities down to a handful worth naming. Before an engine can say "you should look at X," it has to have already filed X under the thing the buyer is asking about. When your category has no established name, three things are missing at once.

There is no head query. Nobody searches "reverse supply-chain attestation tool" because they have never heard the phrase, so there is no volume to optimize against and no SERP to reverse-engineer. There is no training signal. The model saw few or no documents during pretraining that place your kind of product in a coherent group, so it has no prior about what the category contains or who leads it. And there is no consensus corpus. The review sites, roundups, and community threads that engines lean on for category answers do not exist for a category the market has not agreed on.

The instinct is to invent a clever name and repeat it everywhere. That is half right and half a trap. A name nobody is searching for is a label the engine cannot connect to demand. The winning move is to bridge two things: the new name you want to own and the old problem buyers already ask about in words they already use. The full mechanics of how engines assign a brand to a category are worth reading alongside this, in how AI engines decide which category your brand belongs to.

Start from the problem query, not the category name

Buyers whose problem has no category name still describe the problem. They just describe it as a symptom, a workaround, or a job to be done. A team that will eventually buy "reverse supply-chain attestation" is today asking AI things like "how do I prove to an auditor that a supplier three tiers down actually did what they claim." That question has volume. It has intent. And almost nobody is answering it well, because the vendors who could are all busy naming their category instead of answering the question that leads to it.

This is the anchor. Your first citations will not come for the category term. They will come for the problem question, where the engine is actively looking for the single most extractable answer and finding thin content. Map the five to fifteen problem questions that sit upstream of your category, phrased the way a buyer with the pain would type them, and write the definitive answer to each. Inside those answers, and only after you have earned the reader's trust by solving the problem, you introduce the category name as the shorthand for the class of solution. You are teaching the engine the association in the exact context where a buyer needs it.

Name the category so it survives being repeated

If you are going to coin a term, coin one that holds up under machine repetition. The concept of deliberately defining and owning a market frame is what practitioners call category design, and a name built for AI has specific properties.

• Descriptive over cute. "Continuous supplier attestation" tells an engine what the thing does; "AttestFlow" tells it nothing until it has already learned the brand. Descriptive names carry meaning even on first exposure, which is exactly the situation a model is in.
• Grammatically stable. Pick one canonical form and one expansion, then never drift. If it is sometimes "supplier attestation," sometimes "vendor attestation platform," and sometimes "attestation-as-a-service," the model sees three weak signals instead of one strong one.
• Anchored to a known parent. Position the new term as a species of a genus the model already understands: "a category of third-party risk tooling," "an AEO discipline," "a type of supply-chain audit software." The parent gives the engine a shelf to put the new label on.
• Paired with the problem. Every time you state the name, keep it near the problem it solves. Co-occurrence is what turns a coined phrase into a recognized entity, not the phrase alone.

Seed the entity associations engines actually read

Models file brands through co-occurrence: the entities, problems, and competitors your name repeatedly appears next to become the model's definition of what you are. That is measurable. Brand mention density correlates with AI citations far more strongly than backlinks do. So the job is not to say the category name more times on your own site. It is to make your name, the category, and the problem appear together across many independent sources the engine already trusts. A practitioner breakdown of how this works is documented in Victoria Olsina's analysis of entity co-occurrence for AI brand visibility.

Concretely, that means the same tight association, brand plus category plus problem, showing up in your own answers, in analyst or trade coverage, in review-site listings, in community threads where the problem gets discussed, and in structured data. A clean way to declare the term itself to machines is schema.org's DefinedTerm, which lets you mark the category name as a formal defined concept rather than leaving the engine to guess. The through-line for all of this, making a brand legible as a distinct, trusted entity, is covered in depth in how to build a brand entity AI engines recognize and trust.

What to build, in what order

The order matters. Skipping to the category page before you own the problem gives the engine a definition with no demand attached. Skipping the earned mentions leaves the whole category resting on your own domain, which engines discount because roughly nine in ten citations in most answers come from sources a brand does not own.

Prove it with evidence, not adjectives

A nameless category has a credibility problem: the model has no prior that the category is real, so it treats claims about it cautiously. The counter is specificity. Engines cite the source that states a checkable number, names a real example, and shows a clean before-and-after. When a brand that started effectively invisible becomes the consistent answer, it is because the content carried extractable, verifiable substance rather than category evangelism. The FastTrackr AI case study walks through what earning citations from a standing start actually looked like.

Feed the engines a machine-readable map of your best answers so the category page and the problem answers are easy to fetch and hard to miss. A fast way to publish that map is a free llms.txt generator, and the standing distribution that keeps the answers in front of every engine as they recrawl is the AI Feed Engine. Category creation is a repetition game across sources and time, and the surfaces that carry your definition have to stay current for the association to hold.

Measure the association, not just the mention

You cannot manage what you are not sampling. For a nameless category, the metric is not "did we get cited," it is "did the engine connect the three things." Run three prompt panels repeatedly across ChatGPT, Claude, Gemini, and Perplexity: the problem questions, the emerging category term, and the head competitors' names to see whether you surface as an alternative. Log whether the engine names you, whether it uses your category term, and whether it uses your term for competitors too, which is the strongest sign the category is taking. Because AI answers vary run to run, one check is noise; you need an inclusion rate across many samples. Watching which source wins each answer and closing the gap deliberately is the loop OnlyAEO runs end to end.