Prompt tracking

Prompt tracking is a monitoring practice that runs a fixed set of prompts on AI assistants such as ChatGPT, Gemini, Microsoft Copilot and Google AI Mode on a schedule, then records which brands each answer names and which pages it cites. It is the AI-search counterpart of keyword rank tracking.

Also called: AI brand monitoring, LLM tracking, AI rank tracking

Why prompt tracking matters

Prompt tracking exists because a single AI answer tells you almost nothing. In SparkToro's January 2026 study, volunteers ran the same prompts through ChatGPT, Claude and Google's AI 60 to 100 times each. Two answers listing the same brands in the same order came up less than once in 1,000 runs.

Presence was far steadier than order. When ChatGPT was asked for the best cancer care hospitals on the U.S. West Coast, City of Hope appeared in 69 of 71 answers, but came first in only 25 of them. SparkToro's conclusion is the case for prompt tracking: "visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric." Checking a prompt once by hand is a coin toss; tracking it every day turns the noise into a trend.

How prompt tracking works

A prompt tracking setup repeats the same loop on every engine:

Choose the prompts. A fixed list of questions your buyers ask, grouped by intent.
Run them on a schedule. Each prompt goes to ChatGPT, Gemini, Microsoft Copilot, Google AI Mode or other assistants, every day or every week.
Store the full answers. The text, the brands named and the pages cited.
Detect the brands. Each brand mention of your brand and your competitors, with the order in which each is first named.
Aggregate over a period. Mention rate, average position and an AI visibility score summarize hundreds of answers.

Two limits are worth knowing. First, a tracker sees a baseline, not what every user sees: OpenAI explains that ChatGPT "may use an approximate location based on your IP address" and, if memory is enabled, "may use relevant saved memories when rewriting a search query." Second, trackers observe answers from the outside. As Google puts it, "No third-party tool has access to our internal ranking or AI systems."

Example: a prompt set for one brand

Imagine a payroll software company that sells to independent restaurants in the United States. Its prompt set might cover five intents:

Intent	Prompt
Category	What's the best payroll software for a small restaurant?
Problem	How do I handle tip reporting for restaurant staff?
Comparison	Gusto vs ADP for a restaurant with 25 employees
Alternatives	Cheaper alternatives to Toast Payroll
Constraint	Payroll app for a restaurant group with three locations in California

Each prompt runs every day on four engines, so the set produces 20 answers a day and about 600 a month. Depending on the prompt, an engine may search the web before answering or answer from what its model already knows. Over a month, the company sees which intents name it, which competitors take its place, and which pages the engines cite on the comparison prompts.

Prompt tracking vs keyword rank tracking

Prompt tracking is often called AI rank tracking, but it differs from keyword rank tracking on almost every point.

	Keyword rank tracking	Prompt tracking
What you track	Short keywords	Full questions, often long
What comes back	A ranked list of pages	A written answer that names brands
Stability	Positions move slowly	The list and its order change from run to run
Main signal	Your page's position	Whether your brand is named, then where
Demand data	Search volume from keyword tools	Estimated at best, since assistants do not publish prompt volumes

Because order is so unstable, treat position as an average over many runs, and presence as the main signal.

How to choose the prompts to track

Pick prompts by intent, not by exact wording. SparkToro asked volunteers to write their own prompt for the same need and found that, across 142 responses, "there were barely two prompts that, if you squinted, looked similar." Yet across 994 answers to those prompts, the same headphone brands (Bose, Sony, Sennheiser and Apple) showed up 55-77% of the time. The engines captured the intent behind very different wordings.

Start from real questions: sales calls, support tickets, reviews, on-site search and question-shaped queries in Search Console.
Keep branded and unbranded prompts apart. Prompts that name your brand will always mention it and inflate your figures.
Track every engine your buyers use, and in their language. Chen et al. found that AI search services "differ significantly from each other in their domain diversity, freshness, cross-language stability, and sensitivity to phrasing."

Hawkscan runs your prompts every day on ChatGPT, Microsoft Copilot, Gemini and Google AI Mode and keeps each full answer, word for word, so you can read what each engine said, not just the score. It suggests prompts when you set up a project, and you choose which ones to run.

Common questions

How many prompts should I track?

There is no agreed number yet: SparkToro lists "How many are needed" among the open research questions. In practice, track dozens of prompts rather than a handful, spread across the intents that matter to your sales. A small set exaggerates the swing of a single answer, while a set grouped by intent shows where you are strong and where you are missing.

How often should prompts run?

Prompts should run often enough to average out the noise, and daily is a common choice. SparkToro estimates that to really know an AI's set of recommendations, you need to ask "usually at least 60-100X". A daily run builds about 30 answers per prompt and engine each month, enough to see a trend over a few weeks.

Can Google Search Console track AI prompts?

Only partly, and only for Google. Search Console's Generative AI performance report shows how your content performs in Google's own generative AI features. It does not show what ChatGPT, Microsoft Copilot or Gemini answer, nor which competitors they name, which is what prompt tracking adds.