How to choose an AI recommendation tracking tool
Tools for tracking AI recommendations are young and their price lists are hard to compare, because each counts something different. The questions below are the ones that decide after purchase whether a measurement is useful. Put them to every vendor, including us.
Seven questions before you sign
- Which platforms are included in the base price and which are an add-on? The home page list often names more than the plan you are about to buy contains.
- What is the unit being counted: a prompt, a domain or a seat? Twenty prompts on one domain and twenty domains with one prompt can be the same number in a price list and entirely different services.
- How many repetitions per question? A model does not always answer the same question identically. One repetition means the model variation gets recorded as a change in your visibility.
- In what language are the questions asked and who decides that language? Outside the English-speaking market this is decisive, because an answer in English often names different companies than an answer in the local language.
- Does the tool also measure the technical accessibility of the site or only mentions? A mention tells you where you stand; a technical check tells you what you can change yourself.
- What does the advertised frequency actually mean? A daily run is useful only if the measured substance changes daily too. Ask what a daily run actually executes and whether it covers the full set of questions or a sample.
- What remains yours when the subscription ends? Ask about exporting the measurement history and about the period during which the data stays reachable after cancellation.
Why price per prompt is a poor yardstick
A price list with several hundred prompts sounds cheaper than one with a hundred, until you ask what happens to a prompt. A prompt asked once on one platform is one measurement. The same prompt asked three times on three models is nine measurements and nine calls that somebody pays for.
In our case that ratio is stated openly: the paid report asks one hundred and twenty questions with three repetitions on three models, which is one thousand and eighty calls per report. That is also why we do not offer daily tracking; every run is a full paid analysis, not a refresh of a view.
What a trial measurement should show
Before deciding, run a measurement at two vendors on the same domain and compare three things: whether the competitor lists are at least similar, whether the report states the questions that were actually asked, and whether the report shows when the measurement took place.
A report that does not say which questions were asked and when is a score without a trail. A score without a trail cannot be reproduced and therefore cannot serve as evidence of work done.
Frequently asked questions
Is daily tracking better than weekly?
It depends on what you measure. A daily measurement catches model variation that is caused by nothing on your side and presents it as movement. Our offer is weekly or every fourteen days; anyone who needs a daily figure will not get it from us, and that is written in our configuration too, not only here.
How many platforms are enough?
More names on a list is not the same as more usable data. What matters more is whether the ones your buyers use are measured, and whether they are in the base price. We measure three providers and do not plan a fourth; other vendors list more, and charge separately for some of them.
Do I need a tool if I already have a search agency?
A tool and an agency do not replace each other. The tool produces a number and a list; the agency makes the fixes and decides what is worth fixing. Without a measurement, though, an agency reports what it did rather than what changed.
Start with one measurement
Eight questions on three models, a technical check and a score from 0 to 100. No card required.
Check AI visibility