Gemini Visibility Tracker: How to Track Your Brand

A Gemini visibility tracker asks Gemini the same buyer questions on a schedule and records whether your brand is named, how it is described, which sources are cited and which competitors appear. To set one up, fix a prompt set, run each prompt several times, and log mentions, citations, sentiment and share of voice.

Key takeaways

• Measure four things per prompt: mention (named or not), citation (your URL listed or not), sentiment (the words used about you) and share of voice (your mentions divided by all brand mentions).
• Freeze a prompt set of 20 to 50 unbranded buyer questions and version it. Changing prompts mid-stream makes every trend line meaningless.
• Run each prompt several times. Repeated runs of the same prompt on AI engines return different brand lists, so one run is an anecdote, not a measurement.
• Record which Gemini surface you tested: the Gemini app, the Gemini API with Google Search grounding, or Google's AI Overviews and AI Mode. They are not the same measurement.
• Report every figure with its numerator, denominator and a margin of error, and call a change only when it exceeds that margin.

What is a Gemini visibility tracker?

A Gemini visibility tracker is a measurement instrument. It puts a fixed list of questions to Google's Gemini, stores each full answer, and turns those answers into numbers you can compare week to week. Without that, you are left with screenshots from whoever happened to ask last.

It is different from a rank tracker. Gemini does not return ten blue links in positions one to ten. It writes an answer that names some brands, omits others and sometimes lists sources. The question is not "what position are we?" but "are we in the answer, how are we described, and who is named alongside us?"

Several vendors publish pages on this, and they agree on the basics: mentions, cited sources, competitor comparison and share of voice. What the generic pages tend to skip is the method: how many prompts, how many runs, which Gemini surface, and how to know a change is real. That is what the rest of this guide covers. For a version of the same job across every engine, see how to track your brand across AI engines.

Why track Gemini separately from ChatGPT?

Gemini draws on Google's systems, so the pages it relies on and the brands it names can differ from ChatGPT's. Averaging engines together hides that. The reasons to give Gemini its own line in your report:

• Scale. Mission Growth, 2026 reports that Google said on 11 August 2026 that the Gemini app passed 1 billion monthly users.
• Google's search surfaces. Tech Insider, 2026 reports Google AI Mode passing 1 billion monthly users at I/O 2026. Gemini models also power AI Overviews: Engadget, 2026 reported Gemini 3 as the default model for AI Overviews.
• Grounding. Google's Gemini API documentation, October 2026 describes grounding with Google Search as a tool that connects the model to real-time web content and returns citations. When Gemini searches the web, your visibility depends partly on what Google Search can find.

Google's own guidance points the same way. In its Search Central guide to generative AI features, 2026, Google says the way Search finds and processes your pages "remains the core of how our AI systems access your data." Ordinary crawlability and indexing still matter for Gemini visibility. You can check yours with the free AI Crawler Checker.

Engines also disagree with each other. Why ChatGPT, Gemini, Claude and Perplexity disagree about your brand explains why a good Gemini result does not predict a good result elsewhere.

Which Gemini surface are you actually measuring?

This is the choice most tracking posts skip. "Gemini" is at least three measurement targets, and mixing them produces numbers nobody can reproduce.

Pick one as your primary series, label it in every report, and add the others as separate series. Do not blend them into one "Gemini score".

What should a Gemini visibility tracker measure?

Four metrics cover the question, each with a formula you can state out loud. Define them before you collect anything.

Mentions and citations are separate outcomes and need separate work; citation vs brand mention in AI search explains why. A brand can be named with no link, or linked without being recommended. For the share of voice calculation in detail, see what AI share of voice is and how it is calculated.

How do I track my brand's visibility in Gemini, step by step?

1. Write the claim you want Gemini to make. One sentence: what you are, who you serve, what you are known for. It tells you which prompts matter and what "accurate" means later.
2. Build the prompt set. Collect 20 to 50 unbranded buyer questions across stages: category discovery ("best X for Y"), comparison, problem-led and pricing. Use the free AI Prompt Finder, which shows the questions buyers ask about your category, tagged by intent. Keep branded prompts in a separate group; they measure accuracy, not discovery. The method is in which prompts to track.
3. Fix the competitor list. Choose three to five brands that actually appear next to you in buyer decisions. Write down alternate spellings and product names so a mention is not missed.
4. Choose the surface and settings. Pick from the table above. Record the model name, whether web grounding was on, language and country, and the date of every run.
5. Run every prompt several times. Use fresh sessions, not one long conversation, so earlier answers do not colour later ones. Five runs per prompt is a workable minimum for a weekly cadence.
6. Store the full answer text and sources. Save the raw answer, the cited URLs and the date. You will want to re-code it when your definitions improve.
7. Code and compute. Mark each answer for mention, citation and sentiment, then calculate the four metrics with numerator and denominator shown.
8. Review on a fixed rhythm. Weekly is enough to catch a model change; monthly is enough for a board slide. Version the prompt set whenever you edit it and start a new series if you do.

The AI Visibility Measurement Standard from Noma by Nurdd sets out the same discipline: a fixed, versioned prompt set, several runs per engine, and every figure reported with its denominator and margin of error.

Why one run of a prompt is not enough

Generative answers are sampled, not looked up. Search Engine Journal, 2026 reported SparkToro research finding that ChatGPT and Google Search AI Overviews each returned the same brand list less than 1% of the time across repeated runs of the same prompt. The study covered those two products, not the Gemini app, but the mechanism applies to any generative engine. Treat a single Gemini answer as one draw.

That is also why raw positions are a poor metric. Whether you are named first or third in one run tells you little. Whether you are named in 40% of runs, and whether that holds next month, tells you something.

A worked example of the margin of error

Say you track 20 prompts and run each 5 times: 100 answers. Gemini names your brand in 30 of them, a 30% mention rate. A rough 95% margin is 1.96 x the square root of (0.30 x 0.70 / 100), which is about 9 percentage points. So the true rate is plausibly anywhere from 21% to 39%.

Next month you measure 36%. That is inside the margin, so it is not yet a result. It is a reason to keep measuring, not to report a win. This calculation is an approximation: repeated runs of the same prompt are not fully independent, so the real uncertainty is somewhat larger. Treating it as a floor is the safe reading.

How do I measure whether Gemini mentions my brand?

You can run a first pass by hand in an afternoon with a spreadsheet. It will not scale, but it teaches you what the data looks like.

1. Open a fresh Gemini session for each run and paste one unbranded prompt.
2. Copy the full answer and any listed sources into a row.
3. Fill the columns below.
4. Repeat for five runs per prompt, then compute the rates.

Do the arithmetic once by hand: 20 prompts at 5 runs is 100 rows, and the manual pass will take hours. That is the point at which a scheduled tool earns its place. For a comparison of manual, scripted and platform approaches across engines, read how to monitor how often your website shows up in AI chatbot answers.

How do I read the results and decide what to fix?

The pattern of the four metrics tells you which kind of work to do. Use this table as a triage.

The cited sources list is the most actionable output. It shows which pages Gemini leans on for your category, and those pages, not only your own, are where the work happens. How AI engines decide which sources to cite covers what makes a page eligible.

What mistakes make Gemini tracking misleading?

• Tracking branded prompts only. Asking "what is [your brand]" proves Gemini knows you exist. It says nothing about whether buyers find you.
• Editing the prompt set mid-series. Add new prompts as a new version and report the old and new series separately.
• Blending surfaces. App, API and AI Overviews numbers belong on separate lines.
• Reading one run. See the margin-of-error example above.
• Ignoring the wording of mentions. Being named as "a cheaper alternative" and as "the standard choice" are not the same result. Code the phrase.
• Changing the competitor set. Share of voice moves when the denominator moves, regardless of your performance.

What should I look for in a Gemini visibility tracker?

Whether you build it or buy it, check these before committing.

Where does Noma fit?

Noma by Nurdd is an answer engine optimisation platform that tracks 13 AI engines, including ChatGPT, Gemini, Perplexity and Claude, with Gemini included on every paid plan. For Gemini it asks your buyers' questions daily and reports visibility, position and sentiment, competitor share of voice, and the sources each answer cites. See the Gemini visibility tracker page and prompt tracking for how the fixed prompt set works.

See it on your own brand: book a demo or start the 3-day free trial. A card is taken at signup so billing can start on day 3; cancel before then and nothing is charged. Plans are on the pricing page.

How we researched this

We reviewed Google's Search Central and Gemini API documentation, trade press coverage of SparkToro's research on repeat-run variability, secondary reports of Gemini and AI Mode usage, and the public Gemini tracker pages of several AI visibility vendors to see which metrics they cover and what they leave out. Facts, statistics and product details were checked on 2026-10-30. The margin-of-error example is our own arithmetic on illustrative numbers, not data from a study. Usage figures come from secondary reports of Google statements, as labelled above.

What to do next

This week, write the one-sentence claim, build a 20-prompt set with the AI Prompt Finder, and run each prompt five times in a fresh Gemini session. You will have a baseline by Friday. If you would rather have it run daily across every engine, book a demo or start the free trial.