AI Visibility

Tracked Prompts

Choosing the questions worth measuring, organising them by intent, and adding your own.

A tracked prompt is a question phrased the way a buyer would type it, measured across the models a scan covers.

Choose for Revenue, Not Volume

There is no search volume for prompts, so the set is a judgement call rather than a report. Track the questions that decide deals, not the ones that would generate impressions.

The kinds of question below are a way of thinking about coverage, not a fixed taxonomy in the product. Categories on a profile are yours to define.

Kind of questionExampleWhy track it
Category"What tools route inbound leads to the right rep?"Are you in the consideration set at all?
Problem"How do I stop losing inbound leads to slow follow-up?"Buyers who do not yet know the category name
Comparison"Chili Piper vs Surface" · "alternatives to Calendly for teams"Highest intent, and where the answer engine is doing the shortlisting
Evaluation"How long does it take to implement lead routing?"Where a wrong answer costs you a deal quietly
Pricing"Which lead routing tools do not charge per seat?"Often the prompt with the clearest right answer: yours

Add prompts as sales hears new phrasing, and keep the set tight enough that you can still tell whether something you did moved it.

Categories, Clusters and Stages

Prompts carry a category and group into topic clusters. One overall number tells you little, while "strong on category, invisible on comparison" tells you what to write next week. list_tracked_prompts takes category and topicClusterId, so you can read the score by either.

In the app, the Prompt Insights table lists prompts with the models they were tested against, a recommendation, and the current AEO leader for that prompt.

Adding Your Own

The starting set comes from your positioning and competitor context. It will miss the prompts only your sales team knows, so add them. Pass the categoryId of a category already on the profile.

create_tracked_prompt {
  "profileId": "...",
  "text": "Which inbound scheduling tools do not charge per seat?",
  "categoryId": "..."
}

Three sources worth mining:

  • Sales calls. The question a buyer asked on a first call is a tracked prompt, word for word.
  • Support tickets. Pre-sale questions that reached support are questions your site failed to answer.
  • Evidence board. A claim you have already written down is one rephrasing away from a prompt a buyer types.

Scans

A scan runs your tracked prompts against the models and records the result. get_scan_status reports running, queued or idle; trigger_scan starts one. Run one before you publish, so the after has a before to compare against.

Model answers are not the same every time. Treat one scan as one sample, not as evidence that you dropped off a prompt. Read the score before drawing a conclusion from one number.

On this page