How to choose the right prompts to track your AI visibility

AI prompt tracking is the repeated monitoring of a defined set of questions across AI platforms, to measure brand visibility, mentions, competitors, citations, and changes in generated answers over time.

Here are the three most common mistakes.

Guessing. A prompt list generated at random, rather than built from actual search behavior, produces data that does not reflect how buyers ask questions.

Relying on a single category. Tracking only branded terms, or only comparison prompts, produces a prompt list that does not cover the full range of how buyers ask questions, leaving the data incomplete.

Phrasing prompts as keywords rather than questions. A prompt written as a keyword string with a question mark added does not match how a buyer actually asks an AI engine anything, and the tracked result reflects that mismatch, not the brand's actual visibility.

What prompt tracking reveals

Subsig's prompt tracking runs a defined list of prompts and records, for each one, whether a brand is mentioned, its sentiment, its position when ranked against competitors, and the mentions percentage over a set period.

In each prompt, Subsig further breaks down visibility trends over time, brand ranking against named competitors, and the specific domains and URLs being retrieved and cited. Every individual AI response feeding into that prompt is also logged, with the date, the engine, and the sources it pulled from.

This turns a single visibility score into a diagnostic view: which prompts a brand wins, which sources competitors are winning through, and where a brand has zero visibility on a question it should be answering.

How to choose the right prompts

A prompt list built from five categories covers the different ways a buyer asks about a category, rather than testing one type repeatedly.

Comparison and alternatives.
A buyer already aware of specific brands asks an AI engine to weigh them against each other. This tests whether a brand appears in direct comparisons, and whether it is named when a competitor is the anchor point.

Branded terms.
A buyer who has already encountered a brand name asks an engine to describe or validate it. This tests how an engine describes a brand when asked about it directly, including accuracy and tone.

Problem-led.
A buyer who has not yet identified a specific solution describes the underlying issue in their own words. This tests whether a brand's content is structured well enough to show up in response to an unstructured problem statement.

Qualification and buying criteria.
A buyer who knows they need a solution asks an engine to narrow the field by a specific constraint. This tests whether a brand is recommended for the segment it actually serves.

Unbranded discovery.
A buyer has not yet framed their need as a category and asks a general question with no brand or product term. This tests whether a brand's content reaches a buyer at the earliest point in their search.

Choose the right prompts, then see where you stand

Choosing prompts across five categories, and testing them against ChatGPT, Perplexity, Google AI Overview, and Microsoft Copilot, is not something to do once. A prompt list needs revisiting as a brand's category shifts, competitors change, and buyer language evolves.

Running this by hand means testing the same prompts across four engines, on a repeating schedule, and manually logging mentions, sentiment, ranking, and sources for each one. A handful of tools now exist to help with this, Semrush, Profound, and Ahrefs among them, each tracking some version of prompt-level visibility.

Subsig runs this automatically, breaking down every tracked prompt into visibility trends, brand ranking against competitors, and the specific sources being cited, showing where a prompt list is working and where it needs attention.

Ready to track your AI visibility?