Building a Prompt Set to Track Your Brand in AI

Tracking your brand in AI answers starts with building a fixed set of prompts — the real questions your buyers ask engines like ChatGPT, Perplexity and Gemini — and running them on a regular cadence to log whether you’re cited, merely mentioned, or absent. There’s no rank tracker for AI citation, so this manual prompt set is your measurement instrument. Keep it fixed month to month, run each prompt several times, and you get an honest read on whether your AI visibility is improving. This post shows how to build and run one.

Why you need a prompt set at all

Traditional SEO has rank trackers that tell you your position for a keyword every day. AI citation has no equivalent — the engines don’t publish rankings, outputs vary between runs, and much of the traffic is invisible to analytics. A fixed prompt set is the workaround: by asking the same real questions repeatedly and logging what the engines say, you build your own measurement instrument for a channel that otherwise can’t be measured.

Without it, AI visibility is pure guesswork — you’re optimising blind, unable to tell whether your work is helping. With it, you have a repeatable read on presence, citation and mention over time, which is what lets you connect specific content changes to specific movements in visibility.

What to track: four states

For each prompt, on each engine, log which of four states describes the result. Cited: your brand is named and linked as a source. Mentioned: your brand is named but not linked — still valuable for exposure, invisible to analytics. Competitor-cited: a competitor is named or cited but you aren’t — the most actionable gap. Absent: neither you nor an obvious competitor appears, or the engine doesn’t surface brands at all.

Tracking all four, rather than a simple yes/no, tells you not just whether you’re winning but where the opportunities and threats are. A prompt where a competitor is consistently cited and you’re absent is a precise target for content work.

How to build the prompt set

• Start from real buyer questions. The prompts must be questions your actual customers would type, not keywords. “Which digital marketing agencies in India specialise in AI visibility?” not “AI visibility agency India.” Draw them from sales calls, your inbox, People Also Ask, and the questions prospects genuinely ask.
• Cover the buying journey. Include category-level prompts (“what is generative engine optimization”), comparison prompts (“best GEO agencies”), and specific solution prompts (“how do I get my brand cited by ChatGPT”). Different stages surface different competitors and different opportunities.
• Include prompts where you should win and where you don’t yet. A set that only contains prompts you already win tells you nothing about growth. Mix in aspirational prompts where a competitor currently dominates, so you can watch yourself break in.
• Keep it to a manageable size. Twenty to thirty prompts is the practical sweet spot for manual tracking. Enough to be representative, few enough to run several times each across multiple engines without the process collapsing.

How to run it: the discipline that matters

• Keep the prompt set fixed. This is the rule that makes tracking meaningful. Once you’ve built the set, don’t change the wording month to month. If you rewrite prompts, you can’t compare this month to last, and comparability is the entire purpose. Add new prompts to a separate list rather than editing the core set.
• Run each prompt several times. AI outputs are non-deterministic — the same prompt can cite different sources on different runs. Running a prompt once and recording the result is close to noise. Run each three to five times and record how often you appear, so you’re measuring a rate, not a coin flip.
• Run on a set cadence. Monthly is a sensible default. Retrieval-based citation shifts within roughly four to eight weeks of content changes, so monthly checks are frequent enough to catch movement without drowning you in noise.
• Use fresh sessions. Personalisation and chat history can skew results. Run prompts in logged-out or incognito sessions where possible, so you’re seeing a representative answer rather than one shaped by your own history.

Logging and reading the results

Record results in a simple spreadsheet: one row per prompt, columns for each engine, and the state plus how many of the runs cited you. Over months, this becomes a trend line — your citation rate per engine, rising or falling, and the specific prompts where you’re gaining or losing ground.

Read it two ways. Aggregate: is your overall citation rate improving across the set? Granular: which specific prompts flipped from absent to cited after a content change, and which remain stubbornly won by a competitor? The granular view is what turns measurement into a content roadmap.

Frequently asked questions

The bottom line

A fixed prompt set is the closest thing to a rank tracker that AI citation allows. Build 20 to 30 real buyer questions spanning the journey, run each several times monthly across the engines that matter, and log the four states. Keep it fixed so the trend is honest. It’s manual work, but it converts an invisible channel into something you can measure, target and improve.

We build and run your prompt set every month as part of our AI Visibility service — so you see exactly where you’re cited and where the gaps are.