Monitoring
Measure how AI engines answer your buyers: visibility, sources, query fan-out, paid placements and brand perception.
Monitoring is the measurement half of Geonimo. Five screens, each answering a different question about how AI engines treat your brand.
| Screen | The question it answers |
|---|---|
| Overview | Am I visible, and is the change real? |
| Sources | Which domains and URLs do the engines actually read? |
| Query Fan-out | What did the engines search for behind my prompts? |
| Ads & Shopping | Who owns the paid and shopping slots in AI answers? |
| Brand Perception | What do the engines say about me, beyond naming me? |
Configuration — prompts, competitors, products, pages and the traffic tracker — lives in Workspace.
The date range
Every Monitoring tab shares one date range, carried in the URL. Pick a range on the Overview and it is still the range when you land on Sources.
Presets are 7, 30 and 90 days, and any custom range is allowed. The real limit is where your data starts — a project can always be asked for its entire history.
Default is 30 days. Shorter ranges are noisier: fewer answers means a wider confidence interval, and a 7-day swing is often inside the noise floor.
How to read a Monitoring screen
- Start with the KPI row, and read the noise floor next to each number. A change smaller than the floor is not a change.
- Then the trend. Charts plot you against tracked competitors on one axis, with a shaded confidence band on your own series.
- Then drill down. Every figure resolves into the answers, sources or placements that produced it — including the verbatim answer text.
Drill-down is the point. Any number you cannot trace to an answer you can read is a number you should not act on.
What is measured, and what is not yet
The collection engine runs daily against the real consumer surfaces of ChatGPT, Google AI Overviews, Perplexity and Google AI Mode. Mentions, citations, fan-out queries and placements are extracted from those captures.
The enrichment pass that scores tone per mention is not yet running, so sentiment-dependent panels are either captioned as sample data or hidden rather than estimated. The Citation Index on the Overview and the engine, location and tag filters are likewise still sample-backed. See Product status.
An unmeasured value renders as a dash, never as a zero. The distinction is deliberate: "we have not measured this" and "we measured this and found nothing" are different claims, and only one of them is bad news.