Definition
Prompt tracking records what AI assistants say when a fixed list of buyer questions is asked, run after run, so a brand can see whether it is named. Unlike rank tracking, which follows a keyword’s position, prompt tracking follows whole answers: who gets mentioned, what gets claimed, which sources get cited.
At a glance
- CategoriesMeasurement
- Related fieldsAnalytics
- DifficultyIntermediate
At a glance
- The unit of measurement is a prompt, written as a buyer would type it, not a keyword.
- The list stays fixed. Changing prompts mid-quarter destroys the comparison you were building.
- Every prompt needs repeated runs, because assistants do not return the same answer twice.
- Useful output is three fields per run: were you named, in what terms, and was a source cited.
Watching answers instead of rankings
A rank tracker asks a search engine for a position. Prompt tracking asks an assistant a question and reads the paragraph that comes back.
The mechanics are simple and the discipline is where it goes wrong. A fixed set of questions, a fresh session for every run so nothing personalizes the result, repeated runs, and a consistent record of what came back.
That last part is the one teams skip. A screenshot folder is not tracking. A row holds the prompt, the assistant, the date, whether your brand appeared, which competitors appeared, and the URLs cited.
Once that table exists, the useful questions become answerable. Which prompts have you never appeared in. Which competitor is gaining. Which of your pages an assistant reaches for when it reaches for you at all.
A rank tracker has nothing to point at
There is no position in an AI answer, so every tool built on position reports nothing here. Your rank tracker shows a column of green arrows while an assistant recommends three competitors and never mentions you.
Analytics has the same blind spot from the other side. Someone reads your name in an answer and returns two days later through a branded search. Nothing connects that visit to the answer that caused it.
So the only evidence about the answer layer is the answer itself. Either you read what assistants say on a schedule, or you are guessing.
Which makes the prompt list a strategic document, not an admin task. It encodes what you think buyers ask. Get it wrong and you measure the wrong thing precisely for a year.
Start a prompt list this week
Nothing here needs a platform. It needs a spreadsheet and a recurring hour.
- Write 30 questions from three sources. Sales call recordings, support tickets, and your site search log. Use the buyer’s words, including the ones that are wrong.
- Split them by intent. Category questions, comparison questions, problem questions, brand questions. Synthetic queries fill the gaps where you have no recorded examples.
- Run each prompt three times on two assistants, always in a fresh session. Record brand named, competitors named, and cited URLs.
- Vary the asker, not just the question. Running the same list through persona-based querying shows which buyer type you are invisible to.
- Review monthly and change nothing about the method. The list stays fixed for at least two quarters. Trend is the product here, not any single month.
Frequently asked questions
-
How many prompts should I track?
Around thirty to start, covering your main category, your closest comparisons and the problems buyers describe before they know the category name. Fewer than twenty and one odd result swings everything. More than a hundred and the list stops being maintained, which costs you the trend line that made it valuable.
-
How often should I run the list?
Monthly for most brands. Weekly if you are actively fixing pages and want faster feedback on whether the rewrite landed. Daily is theatre for almost everyone: answers vary run to run, and a daily chart mostly plots that variation rather than any change in how you are doing.
-
Should I track competitors in the same list?
Yes, and it is often the more useful half. Being absent from an answer tells you little on its own. Knowing that the same three competitors appear in every answer where you do not tells you exactly which pages to read and what standard you are being measured against.
-
Can I run prompt tracking in a spreadsheet?
For a first quarter, yes, and it teaches you what you actually want to measure. It stops scaling when you need several assistants, repeated runs and a persona split, because that multiplies into hundreds of manual sessions and the funnel view gets hard to hold together by hand.
-
Is prompt tracking worth it if almost no traffic comes from assistants?
Often not at full spend, no. If your buyers are not using assistants, a monthly manual run of ten prompts keeps you informed. Put the rest of the budget into content and crawlability, which pays in ordinary search either way. Revisit when branded searches start arriving that you cannot explain.