AnswersDeciding what to spend

What does AI visibility tracking cost?

The cheapest version is free and you should do it this afternoon: open 5 assistants and ask them what a buyer in your category would ask. What money buys beyond that is repetition and coverage, and those are worth buying only once you know why one look is not enough. Here is what changes at each size, and what to refuse to pay for at any of them.

What is the cheapest way to do this?

Ask the assistants yourself, in a fresh chat with no history, and write down what they say. It costs nothing, it takes an afternoon, and for a single brand checking a handful of questions occasionally it is genuinely enough.

Write down the five or ten questions a buyer actually types before choosing something like yours. Put each one to each assistant in a new conversation, because a chat that already knows who you are is not the conversation a stranger has. Record which brands come back and in what order.

Nobody should buy a tool before doing this once, and a vendor who discourages it is telling you something. What it gives you is a real answer to whether you appear at all, which is the first thing worth knowing and the cheapest thing to find out.

So what is the money actually for?

Repetition and coverage. An assistant does not give the same answer twice, so one look is one sample. Doing it by hand across five assistants, several samples each, every month, is the point at which a person's week is more expensive than software.

The figure beside this section is why. If an assistant returns a different set of names most times it is asked, then a single reading tells you what one assistant said once, and a change you notice between two readings might be the measurement moving rather than your position. Knowing which is the difference between acting and guessing.

Doing that properly means every question, to every assistant, several times, on a schedule, with the answers kept so a figure can be traced back to the sentence that produced it. That is arithmetic on somebody’s time before it is a software decision: ten questions across five assistants at three samples each is a hundred and fifty conversations a month, transcribed.

What does a small marketing team need?

One brand, a question set they wrote themselves, and a monthly reading with the answers kept. Anything beyond that is capacity they will not use, and the thing to refuse is a score with no denominator behind it.

At this size the useful output is a short list of the questions where somebody else is being recommended, with the actual answers underneath. What is not useful is a dashboard of composite figures: a team of two does not have a week to spend working out what a number means.

The one thing worth insisting on at any price is that a figure can be opened. If a score cannot be traced to the answers it came from, it cannot be argued with, and a measurement nobody can check is a number nobody should act on.

What changes as a team grows?

Several brands or several markets, which turns a reading into a comparison, and somebody who needs the work sequenced rather than listed. That is where a plan starts being worth more than the measurement.

Two things arrive together. More than one brand means the readings have to be comparable with each other, which is a real constraint rather than a feature: the same instrument, the same cadence, the same denominators. And more than one person acting on it means a list of findings is no longer enough, because a list does not say what to do first.

This is where the price steps in most of this category, ours included, and it is a defensible step: the recommendation layer is the expensive half to build and the half that decides whether anything changes.

What does an enterprise need that a small team does not?

More brands and more markets, a bigger question set because the category is wider, and evidence that survives being forwarded. The last one is the real difference, because a figure reaching a board has to be defensible by somebody who was not in the room.

Scale changes the shape rather than only the volume. A wider category needs a wider question set to be representative, and a wider set costs more to sample because every question is asked of every assistant several times.

What matters more than any of that is what happens to a figure after it leaves the team. It gets forwarded, quoted in a deck and repeated in a meeting nobody who understands it attends. So the thing to buy at this size is not a bigger dashboard, it is a measurement whose margin of error travels with it and whose evidence can be produced on demand.

What should you refuse to pay for, at any size?

A score with no denominator, a movement reported as a result without a margin of error, and a claim about assistants nobody can show you the answers for. None of the three costs a vendor anything to fix, so an absence is a choice.

Ask how many prompts a score rests on and how many samples per assistant. Ask what the same measurement returned last time and how far it would move if it were taken again with nothing changed. Ask to see the answers behind one figure.

Those three are cheap for an honest vendor to answer and impossible for one whose number is a single sample. They are worth more than any feature comparison, and they are the same three questions whether you are spending a hundred a month or thousands.

Which of the five is holding your brand back?

Start with the free version, and buy repetition when checking by hand costs more than it tells you.

Start measuring Read the method