A method for finding out what Google Gemini says about your brand: which prompts to run, why the app and AI Overviews have to be recorded separately, and what to log each time.
This is the step that makes Gemini different from the other engines, and the one most teams get wrong. “What does Google’s AI say about us” is two separate questions:
They are different systems with their own retrieval, and they do not reliably name the same brands for the same question. Measuring one and reporting it as “our Google AI visibility” will eventually embarrass you.
Record them as two separate columns from the beginning. Merging them later is impossible; splitting them later means re-running everything.
Start with the questions someone asks before they know you exist. Prompts containing your brand name only tell you what Gemini knows about you once prompted, which is a much easier test than whether it brings you up unprompted.
Ten to twenty is enough. Write them down and keep the wording fixed — the method depends on asking the same questions every time.
Gemini is attached to a Google account, and account history influences what you see. If you have spent months reading about your own category from your work profile, your results are not what a prospect gets.
Gemini leans heavily on material it can retrieve from the web, which means the pages written about your category often matter more than the pages on your own site. For each response, record:
The source column is where the actionable work comes from. If the same review site or roundup keeps appearing, that page is deciding your category, and influencing it beats editing your homepage again.
Run the full list again on a fixed interval and keep results side by side. One pass gives you a snapshot; several give you a mention rate per prompt and the direction it is moving.
Google ships model and retrieval changes continuously, so a position held last month is not evidence of a position held now. This is also where doing it by hand stops being realistic: twenty prompts across two surfaces is forty answers per cycle, and the numbers only compare if the same judgment is applied every time.
Do one manual round regardless. It teaches you what the answers look like in your category, which makes everything you automate afterwards more useful.
If you appear in AI Overviews but not the Gemini app, you have search-side authority that is not translating into the assistant. Those are different problems and the fix is not the same.
Appearing in a third of answers sounds weak in isolation. If the strongest competitor appears in a fifth, it is a lead. Visibility in a generated answer is relative by nature.
The prompts where you never appear are the conversations happening without you. Nothing in your analytics will ever report them, which is why they go unnoticed for months.
Automate when the question shifts from “does Gemini know us?” to “is our presence improving or eroding, and against whom?” That question needs a trend, and a trend needs the same prompts run the same way on a schedule.
InfluenceLayer runs this method for you: prompt clusters executed against Gemini on a recurring schedule, every response analyzed for brand mentions, competitor mentions and cited sources.