# How do you track Google AI Mode?

Source: https://aimodetracking.com/guide · Publisher: AI Mode Tracking

You track AI Mode by asking a fixed set of real buyer questions under fixed conditions, recording every brand and source in each answer, and repeating that on a schedule so you can count rates instead of trusting single screenshots.

## The five-step method
### 1. Write the questions your buyers actually ask
AI Mode is a conversation, not a results page, so you track prompts rather than keywords. Start with twenty to fifty questions in the words a customer would use: problem questions (“how do I stop my dishwasher smelling”), comparison questions (“X or Y for a small office”) and shortlist questions (“best payroll software for a 30-person company”).

Tag each prompt by intent and product line. Keep branded prompts separate from unbranded ones: being named when someone asks about you is expected; being named when nobody did is the visibility you are trying to measure.

### 2. Fix the conditions before you look
The same prompt can produce a different answer depending on who asks, from where, in which language and with which search history. Decide the conditions once and write them down: country and city, language, device, signed-out or a clean test account, and the date and time of each run.

If you change a condition later, start a new series. Mixing conditions is the most common reason a chart shows a jump that never happened.

### 3. Record the whole answer, not a yes or no
For every run, store the full answer text, every brand it names and in which order, every source card and link it shows, and whether your own domain is among them. A screenshot is useful evidence, but a structured log is what lets you count.

Mentions and citations are different things. A brand can be named without its site being linked, and a site can be linked as a source for an answer that recommends a competitor. Track both, separately.

### 4. Repeat the same prompts on a schedule
One run is an anecdote. Generative answers vary between runs, so visibility is a rate: in how many of the runs for a prompt did your brand appear? Re-run the full set on a fixed cadence, weekly at minimum, and run high-value prompts more than once per session.

Keep the prompt wording frozen. If a question needs rewording, retire the old prompt and add the new one as a separate line so history stays comparable.

### 5. Compare against competitors and act on the sources
Report three numbers per prompt group: how often you are mentioned, how often your domain is cited, and how often each competitor is. Then look at the sources AI Mode leans on for the prompts you lose. Those pages, and the questions they answer, are your content and outreach list.

When the log outgrows a spreadsheet, usually once you track more than a few dozen prompts or more than one country, a tracking tool does the same steps automatically. Our tool comparison explains which ones can.

## What to log for every run
| Field | Why it matters |
|---|---|
| Prompt + tag | Groups results by intent and product line |
| Conditions | Country, city, language, device, signed-in state |
| Answer text | Lets you re-check wording and sentiment later |
| Brands named, in order | Mention rate and position versus competitors |
| Sources and links shown | Citation rate and the pages AI Mode trusts |
| Timestamp | Separates real change from day-to-day variance |

## Five mistakes that make AI Mode data useless
- Checking from your own signed-in browser, where history and location shape the answer.
- Counting a single run as a result instead of a rate across several runs.
- Reporting mentions and citations as one number.
- Rewording prompts over time and comparing the new answers with the old.
- Treating AI Mode and AI Overviews as the same surface; they are triggered and laid out differently.
