LLM rank tracker: why there is no rank, and the five numbers to track instead

An LLM rank tracker is a tool that asks a fixed set of questions in ChatGPT, Gemini, Perplexity, Claude and Google's AI surfaces on a schedule, then records whether your business was named, where in the answer it appeared, and which sources were cited. It does not report a position from one to ten, because an AI answer has no ranked list to read a position from. The metric that replaces rank is a mention rate across repeated runs, and this page covers the five numbers worth tracking, how to build the tracker yourself in a spreadsheet, and what a paid one adds.

Why there is no rank to track

A search results page is a list in a fixed order, so position is a real property of it. An AI answer is written at the moment you ask. It names nobody, or one company, or three, in prose, and the same question asked an hour later can return a different set. There is no slot ten to be in.

That changes the unit of measurement in three ways, and every one of them has a practical consequence.

• The reading is a frequency, not a place. You are measuring how often you appear across many runs, which means a single run tells you very little and twenty runs tell you a lot.
• Each engine is its own market. The five surfaces run no shared ranking, so a reading from one assistant is a reading about one assistant. More on that in why one AI app is not enough to measure with.
• The citation list is half the data. An answer that names a competitor and cites three directories has just handed you a target list. Keep it.

The five numbers worth tracking

Metric	Definition	Why it earns a column
Mention rate	Share of runs on a prompt where your business is named, reported with a range	This is the headline. It replaces position
Share of voice	Your mentions as a share of all company mentions across your prompt set	Shows whether you gained or the category simply got noisier
Ordinal position in the answer	First named, second, third, or named only in passing	Being named first is worth more than being named at all
Citation share	Share of answers citing a page on your domain	The number that tracks your pages rather than your reputation
Source set	The list of domains cited across all answers, counted	Your off site work plan, written for you by the engines

Report mention rate with a range around it and compare ranges rather than single figures. When two ranges overlap, the honest reading is that the number has not moved yet. The reasoning is set out in what a mention rate is and why it needs a range, and the question of what counts as good is handled in what is a good AI mention rate.

Build one yourself in a spreadsheet

Before you buy anything, run two months by hand. It costs an hour a month and it teaches you what the paid dashboard is doing.

• One sheet, one row per run. Columns: date, prompt, engine, run number, named yes or no, ordinal position, every company named, every source cited, and a note.
• Freeze the prompt set. Ten to twenty buyer questions, written in your buyers' words, with your company name in none of them. Changing the prompts mid quarter resets your history.
• Three runs per prompt per engine. Three is the minimum that shows variation at all. More runs narrow the range.
• Temporary chats only. Memory off, custom instructions off, signed out where the product allows it. Your own account is the least representative place to run a test.
• Same day of the month, same hour. Consistency in when you measure is what makes two months comparable.
• Then compute three figures. Mention rate per engine, share of voice across the set, and a counted list of cited domains.

On cadence, monthly is the right default for most businesses, with a weekly round only on the handful of prompts tied to a live campaign. We cover the reasoning in how often you should measure AI visibility.

What a paid tracker adds

The spreadsheet gives you the truth about one month. A tool earns its invoice on four things.

• Volume. Fifty prompts times five engines times several runs is thousands of answers a month, which is past the point of doing it by hand.
• History. Stored runs let you see a trend line rather than two snapshots, and a trend is what a board meeting needs.
• Competitor rows. The same prompt set scored for five rivals at once, which turns a mention rate into a share.
• The citation ledger. Every source from every answer, counted and ranked, so your off site target list maintains itself.

Since there is no rank, the honest version of a tracker is a counter: of the questions a buyer would ask, in how many does the reply name you, and how does that differ between assistants. That number moves for reasons you can act on. AI Knows Us keeps the question set fixed so the count is comparable over time, and stores the cited sources so a change has an explanation.

Three mistakes that make a tracker useless

• Naming yourself in the prompt. Ask about your own company and you will find yourself everywhere. The reading is about the prompt, not the market.
• One run per prompt. A single answer is a sample of one. Two runs can disagree completely, and reporting the first one as a result produces a chart that moves for no reason.
• Tracking the head term. Broad category terms are often answered from model training with no sources attached. Prompts that describe a buying situation are the ones that trigger a search and name companies.

This page covers tracking: the metrics, the method and the tooling. For the work that changes the numbers, see LLM optimization. For choosing which questions to track, see what a buyer question is and how to pick yours.

Asked and answered: LLM rank tracker

Is there such a thing as a ChatGPT rank?

No, and the useful substitute is a mention rate across repeated runs, plus whether you were named first. ChatGPT writes prose rather than publishing a list, so frequency and order inside the answer are the properties that exist to be measured.

How many prompts should I track?

Ten to twenty for a single business. Enough to cover the real buying questions, few enough that you can run each one several times every month. Fifty or more makes sense for an agency reporting across several clients or a brand with many product lines.

How often do AI answers change?

Enough that two runs of the same prompt on the same day can name different companies, which is exactly why the method uses several runs and a range. Treat a change as real when the ranges stop overlapping.

Can I track this in Google Analytics?

Analytics shows referrals from assistant domains when somebody clicks through, which is a small and well qualified slice of the picture. Most AI answers end without a click, so the mention itself has to be captured by asking the question.

Do GEO rank trackers and AI visibility tools differ?

They are the same category under different names. GEO, AEO and LLMO all describe the same work, and the tools built for them measure mentions and citations in generated answers. The naming is sorted out in LLMO and how it relates to GEO, AEO and SEO.

What to do first

• Write ten buyer questions in your buyers' words, with your company name in none of them.
• Open a spreadsheet with the nine columns above and run each question three times in two assistants today.
• Compute a mention rate per engine and a counted list of every domain cited.
• Pick the three most cited domains you are missing from and get properly listed on them this month.
• Diarise the same round for the same day next month before you look at any paid tool.