AI Mode rank tracking is the practice of measuring how often Google AI Mode names and cites your brand across a fixed set of prompts, run on a fixed cadence. There is no position 1 to 10 to track: AI Mode returns a single synthesised answer with a few inline citations, not a ranked list. The four numbers that replace a position are mention rate, citation rate, share of voice against competitors, and the set of cited URLs. Track those weekly over an unchanging prompt set and they move the way rankings do — and unlike rankings, the cited-URL list tells you exactly which page to fix.
The short version: stop looking for a position. Freeze 30-100 buyer prompts, pin your locale, run them weekly, and log mention, citation, competitors and sources for every answer. The trend is the metric; a single answer is a sample.
Why "rank tracking" breaks down in AI Mode
Classic rank tracking works because a search results page is a stable, ordered list. You ask the same question, you get ten results in an order, and you read off where you sit. Every reporting habit in SEO is built on that one property.
AI Mode removes it. Ask a question and you get one answer, synthesised by a custom Gemini model, with a handful of inline citations attached to specific claims. There is no second result, no third, no ordering to read. A brand is either in that answer or it is not.
It gets less stable still, because AI Mode does not run one search. It uses query fan-out: your question is split into several related sub-queries, each fired at Google's Search index, Knowledge Graph and Shopping Graph in parallel, and the answer is assembled from the strongest passages retrieved across all of them. Change the phrasing slightly and the fan-out changes, so the retrieved passages change, so the cited brands change. Ask the same question from a different country and the same thing happens.
This is why a screenshot proves nothing and why "we checked, we're in there" is not a measurement. Two searches an hour apart can cite different domains. Anything you want to call AI Mode rank tracking has to survive that variance, which means sampling across many prompts rather than inspecting one.
The four metrics that replace a position
Every workable AI Mode tracking setup converges on the same four numbers. Together they answer the questions a position used to answer: am I visible, am I gaining, and who is beating me.
| Metric | What it counts | What it tells you |
|---|---|---|
| Mention rate | Share of prompts where the answer names your brand at all | Whether the model knows you exist in this category |
| Citation rate | Share of prompts where it names you and links one of your URLs | Whether your own content is doing the work, or someone else's page is |
| Share of voice | Your citations as a proportion of all brands cited on the same prompts | Competitive position — the closest thing to a ranking you get |
| Cited URLs | Every source URL the answer pulled, yours and everyone else's | The actionable half: which page to write, fix, or outrank |
The fourth one is the one teams underuse. A position tells you that you lost; a cited-URL log tells you which competitor page took the citation and what it says. See AI share of voice for how the competitive benchmark is computed, and AI visibility for how these roll up into a single score.
How to set up AI Mode rank tracking: five steps
- Freeze a prompt set. Write 30 to 100 buyer-intent prompts describing the problem your category solves, without naming your brand — "best AI visibility tool for agencies", not "is Rankio good". Then freeze the list. Comparability week to week is the entire point, and a prompt set that keeps changing produces a chart that means nothing. Thirty is the floor; 60 to 100 is where the numbers get stable enough to show a board.
- Pin locale and session conditions. AI Mode answers shift with country, interface language and signed-in history. Pin the locale you actually sell into — if you sell in France, track in French from France, not in English from a US IP — and run logged-out or memory-disabled sessions so personalisation does not inflate your own mention rate. Then keep those conditions identical between runs.
- Log four things per answer. Is your brand named. Is it cited with a link. Which competitor brands appear. Which source URLs the answer pulled. Three booleans and a list — that is the whole schema, and everything else is derived from it.
- Aggregate into rates, never positions. Roll the log up across the whole prompt set into mention rate, citation rate and share of voice. Resist the urge to synthesise a fake "average position" from where your brand appears in the paragraph: it looks familiar to SEO stakeholders and it measures nothing.
- Re-run weekly and read the trend. Weekly is the minimum useful cadence; daily if you are actively shipping content and want to attribute movement to it. Below weekly, the retrieval layer refreshes several times between measurements and you lose the ability to connect a change in citation rate to anything you did.
Checking by hand vs using an AI Mode rank tracker
Manual checking is a reasonable way to find out whether you have a problem. It is not a way to measure one. The arithmetic makes the case on its own: 60 prompts run weekly across one engine is 3 120 answers a year to open, read and log by hand — and that is before you add the other engines your buyers use.
| By hand | With a tracker | |
|---|---|---|
| Sample size | A handful of prompts, whichever came to mind | The full frozen set, every run |
| Conditions | Your own logged-in browser, your own location | Pinned locale, clean sessions, identical between runs |
| Variance | Invisible — one answer looks like the truth | Averaged out across the set |
| Cited sources | Rarely logged, never aggregated | Every URL captured and ranked by frequency |
| Trend | None — each check starts from zero | Week-on-week movement you can attribute |
If you want the engine-side view of how AI Mode picks what it cites — and what to change on your pages to be picked — that lives on the Google AI Mode visibility page. This page is about measuring; that one is about earning.
AI Mode rank tracking for French and European brands
Google is still the dominant search entry point in France and across Europe, which makes AI Mode the largest AI answer surface most European brands have — but tracking it in English from a US IP will tell you nothing about your actual market. French-language fan-out retrieves different sub-queries, different sources and different brands. Locale is not a setting to leave at default; for a French brand it is the measurement.
Two things follow. First, AI Mode alone is an incomplete picture in France: Le Chat (Mistral) pulls from a different source mix, and a brand can be well cited in AI Mode and absent from the assistant a French buyer actually opens. AI Mode tracking belongs inside one cross-engine share of voice, not in its own dashboard.
Second, the data itself deserves a look before you sign anything. AI Mode monitoring stores raw generated text, and answers about smaller companies routinely name founders and executives — which puts the stored output close to personal-data territory. Ask any vendor where raw model outputs live, how long they are retained, and whether processing happens inside the EU. Rankio keeps all measurement data in the EU with a transparent DPA; see our RGPD position.
AI Mode rank tracking checklist
- 30-100 buyer-intent prompts, written without your brand name, and frozen
- Locale pinned to the market you sell into, in that market's language
- Logged-out or memory-disabled sessions, identical between runs
- Mention, citation, competitor set and source URLs logged per answer
- Reported as rates and share of voice, never as an invented position
- Weekly cadence minimum, with the prompt set unchanged
- AI Mode read alongside ChatGPT, Perplexity, Gemini, Claude and Le Chat
- Raw output storage location and retention confirmed with your vendor
Frequently asked questions
Track your AI Mode visibility weekly
A frozen prompt set run on a fixed cadence across Google AI Mode, ChatGPT, Perplexity, Gemini, Claude and Le Chat — with every cited source logged. EU-hosted, RGPD-native.