What is a prompt?

A prompt is the question or instruction a person types into an AI assistant such as ChatGPT, Gemini, Microsoft Copilot or Google AI Mode. In AI visibility tracking, a prompt is also a buyer question that a brand runs on these engines on a schedule, to see whether the answers name it.

Why prompts matter for brand visibility

Prompts are the new search queries, and many of them ask for advice. In OpenAI's study of 1.5 million ChatGPT conversations, published in September 2025, about half of messages (49%) were "Asking": people seeking information or advice rather than getting a task done. When the question is "which tool should I pick?", the answer is a shortlist of brands. For a brand, the prompt is the unit of competition, as the keyword is in SEO: one question from a buyer, and one answer that names you or someone else.

How a tracked prompt works

A tracked prompt is a prompt you run again and again to measure your visibility, the basis of prompt tracking.

Write the question as a buyer would, with their context: team size, budget, use case.
Run it on each engine. Engines that search the web often split it into several searches, a process called query fan-out.
Record the answer: which brands are named, in what order, and which pages are cited.
Repeat it on a schedule, because the same prompt gets a different answer from one run to the next (see AI answer variability).

Over time, a set of tracked prompts gives you a mention rate and an AI visibility score.

Example: one buyer question

I'm opening a physiotherapy clinic with two treatment rooms. Which booking software should I use if I want automatic SMS reminders?

An engine reads the criteria (a small clinic, healthcare, SMS reminders), searches for clinic booking software and for reminder features, then writes a shortlist with a few lines on each tool and links to its sources. A vendor that offers SMS reminders but never says so in plain text, on its own site or on the review pages the engine reads, gives the engine nothing to work with.

How to choose the prompts to track

People word the same need in very different ways. SparkToro asked panelists to write a prompt about choosing headphones for a traveling family member: across 142 prompts, the overall semantic similarity was just 0.081. Track intents, not exact wording:

cover each stage: discovery ("best..."), comparison ("X vs Y") and validation ("is X reliable?");
mix unbranded and branded questions (see branded vs unbranded prompts);
borrow your buyers' words from sales calls and support tickets.

Hawkscan shows the full answers, word for word, from ChatGPT, Microsoft Copilot, Gemini and Google AI Mode for every prompt you track.