If you manage AI search visibility, you need a repeatable record of how answers describe your business for relevant buyer questions. A tracked prompt is a question sent to an AI assistant on a schedule you set. The report shows whether your brand was named, which sources were cited, and how that compares with the competitors named in your campaign.
Where prompts go
A tracked prompt can go to three providers: ChatGPT, Gemini and Google AI Mode, and you choose which of them each prompt is sent to. It is sent word for word, with no campaign text or persona added to it.
Claude and Perplexity are not available and cannot be added on request. If either must appear in your reporting, this is the wrong product for that requirement and other products cover them.
Choosing the days
You tick which days of the week each item needs fresh data on, using seven checkboxes. You never choose a time.
Collection runs the night before each ticked day, so a Monday tick is a Sunday night run. Reports are read in the morning. A collection starting at breakfast would not be finished when the report is opened.
How usage is counted
One credit is one call: one tracked item, on one provider, on one day. The weekly cost of an item is a multiplication:
providers chosen × days ticked = credits per week
Two hundred prompts on all three providers, twice a week, is 1,200 credits. The same two hundred prompts checked five days a week is 3,000.
Because frequency multiplies, it usually decides whether an allowance is enough rather than the number of prompts. A large prompt set checked weekly costs less than a small one checked daily.
Prompts and keywords draw on one weekly allowance with no separate quota for each.
Each prompt runs once per selected provider in each scheduled collection, and the report stores that one answer rather than an aggregate. Credits are checked when the weekly assignment is saved rather than spent call by call, so a failed collection has no separate charge or refund.
What is captured from each answer
The collected result includes the answer text, the sources it showed, the web results it used, the follow-up queries it reported, and the brand entities it identified. Nothing re-reads the answer afterward to find brand names.
So a report shows what the service returned at that moment rather than a later interpretation of it. Read the follow-up queries as what the service reported it would search, not as proof of what it privately executed.
Reading the result
Assistants return different answers to the same question at different times. A single run is one observation of a variable system, and acting on it is the most common mistake made with these reports.
What remains consistent across those answers is consistent presence, consistent absence, direction across weeks, and comparison against competitors facing the same conditions. Compare results across a month.
The most actionable output is usually the cited sources. For a question where you want to appear, those pages are demonstrably being used to answer your buyers, which makes them a placement list with named targets.
Keeping a series comparable
The prompt is the measurement instrument. Change its wording and the series restarts, because a different question is now being asked. Keep the wording fixed and record the date whenever you deliberately change one.
Items can be archived, which stops collection and keeps the history.
What this cannot do
It cannot report a position, because there is usually no ordered list to hold one in. It cannot attribute traffic or revenue, because the path from an answer to a visit is frequently indirect or absent. And it cannot explain why an assistant selected its sources, which is not observable from outside.
Before recommending work from a tracking change, compare the same question and provider across dates and read the answer in context.
See how collection runs, how to choose the prompts, or why repeated answers vary.