What a brand can influence in an AI answer, and what it cannot

Once a team has read the AI answers given about its category, the next reaction is nearly always the same: how do we change them? It is the right question, but it hides a harder one underneath. Before deciding what to do, it is worth being clear about what is actually within reach. A great deal of the effort spent on AI visibility goes into things no brand can move, while the things it can move are left half done.

This piece sets out the line between the two, and argues that drawing it honestly is the most useful thing a marketing team can do before it spends anything on the work.

Two lists, not one

It helps to sort everything that shapes an answer into two lists. The first holds what sits outside your reach: the systems, the choices and the variation that belong to the providers and to the buyer. The second holds what sits inside it: the evidence you publish, the claims you make and the way you measure the result. Between them sits a smaller middle ground of evidence that other people publish, which you can earn but never write yourself.

Most disappointment with this kind of work comes from confusing the lists. A team that believes it can steer the model spends on tactics that promise to do so, and is puzzled when the answers do not move. A team that believes nothing can be done leaves its own evidence thin, and is puzzled when a rival with a weaker product is recommended ahead of it.

What sits outside your reach

The model itself is the obvious place to start. What a model absorbed in training, how it weighs one source against another and how it chooses to phrase an answer are all decided by the provider. They change when the provider changes them, on the provider's schedule, and usually without notice.

The buyer's choices sit here too. You do not choose which assistant a buyer opens, which model it runs or how the question is worded. Small differences in wording produce different answers, and a buyer in a hurry will not phrase a question the way your positioning document would.

Variation belongs on this list as well. The same question, asked of the same assistant on different days, can return different answers. That is not a malfunction to be repaired but the way these systems behave, and no amount of effort removes it.

Finally, the evidence your rivals publish is outside your reach. If another vendor has published clear, specific material about a capability you share, the answer may well draw on theirs rather than yours. You cannot remove it. You can only make sure your own is at least as clear.

Anyone offering to change the items on this list directly should be asked to show how. The honest answer is usually that they cannot, and the other answer tends to be a short-lived trick that depends on a provider not noticing.

What sits inside it

The second list is shorter, but everything on it is yours.

The first and largest item is your own published evidence. What you say about your product, where you say it and how specifically you say it are entirely within your control. An integration that exists but is described nowhere cannot be cited. A security standard you meet but never write down cannot support a recommendation. A sector you serve well but never name in a buyer's words will not connect you to that buyer's question. Each of these is a gap between what is true and what is findable, and closing that gap is work a brand can do.

The second item is consistency. When your website, your documentation and your partner listings describe the same capability in different terms, or contradict one another, the evidence for any one claim is weaker. Saying the same true thing, in the same words, wherever it appears is unglamorous and entirely within reach.

The third item is the set of questions you hold yourself to. You decide which buying questions matter commercially, and those are the ones your position is judged on. Choosing them in the words a buyer would use, rather than the words you would prefer, is what makes everything after it meaningful.

The fourth item is how honestly you measure. You decide whether to compare like with like over time, whether to keep a mention, a citation and a qualified recommendation apart, and whether to report the results that stayed flat beside the ones that moved. None of that changes an answer, but it decides whether you know what your work achieved.

The middle ground

Some of the most influential evidence is published by other people: comparison articles, analyst coverage, partner directories, community discussion and reviews. You cannot write it, and attempts to fake it tend to be noticed and to do lasting harm.

What you can do is give other people a reason, and the material, to write accurately about you. A partner listing that is complete and current, a technical detail documented clearly enough for a reviewer to quote, a claim backed by something a third party can check: each makes it more likely that independent evidence exists and that it says the right thing. The influence is real but indirect, and it is earned over time rather than bought.

Why the line matters

Drawing the line honestly changes where a team puts its effort. Work on the second list compounds: evidence published this quarter is still there next quarter, available to every assistant that goes looking for it. Work aimed at the first list does not compound, because the thing it targets belongs to someone else and changes when they decide.

It also changes what a team can promise. Nobody can promise that a model will recommend a brand, because nobody outside the provider controls that. What a team can promise is that the evidence for the brand's real strengths will be clear, specific, consistent and findable, and that the effect will be measured on the same questions over time. That is a smaller promise, and a far more useful one.

And it changes how results are read. When an answer moves, a team that has kept the lists apart can ask whether the move followed its own work or something it does not control, such as a change on the provider's side. When an answer stays flat, the same team can say so plainly, rather than reaching for a different number that happens to look better.

Evidence, not tricks

The argument comes down to this. The parts of an AI answer a brand can influence are the parts built from evidence the brand controls or can earn. They are worth the effort precisely because they last and because they are honest: a brand that is easy to find, easy to verify and easy to recommend is better placed with every assistant, including the ones that do not exist yet.

The parts it cannot influence are worth knowing too, if only so that nobody spends money pretending otherwise. Accepting that the model is not yours to steer is not a defeat. It is what lets a team put its effort where effort works.

The working steps that follow from this argument are set out in the AI visibility checklist. This piece is the reasoning; the checklist is the method.