AI visibility tracking for agencies
optmz asks AI engines your clients' buyer questions every week and reports how often each client is mentioned, recommended and cited, with a 95% range.
What AI visibility tracking measures
AI visibility tracking measures what generated answers say about a brand. For each client, optmz keeps a set of questions their buyers would plausibly ask an assistant, such as “best accounting software for a 10-person agency” or “is [made-up brand] worth it”, and collects the answers every week.
Each answer is read for three things: whether it names the client (mention rate), whether it suggests the client for what was asked (recommendation rate), and whether it links to one of the client's own pages as a source (citation coverage). Competitors get the same three rates from the same answers, which gives a fair share of voice.
Which AI engines optmz tracks
optmz tracks ChatGPT, Claude, Gemini and Perplexity by default, through their APIs with web search turned on. Grok, DeepSeek and Meta AI can be added per client. Google AI Overviews and AI Mode are read from Google results pages, and the ChatGPT and Gemini apps can be checked as a signed-out user sees them.
| Surface | How answers are collected | Samples per question |
|---|---|---|
| ChatGPT, Claude, Gemini, Perplexity | API with web search | 3 a week by default |
| Grok, DeepSeek, Meta AI (opt-in) | API with web search | 3 a week by default |
| Google AI Overviews and AI Mode | Google results page | 1 a week |
| ChatGPT and Gemini apps (opt-in) | Consumer app, signed out | 1 a week |
Every answer carries a label saying how it was collected. An API answer is not the same thing as what a person sees in the app, so the two are never mixed silently. The methodology explains each label.
Why every rate has a sample size and a range
AI answers change from one run to the next. Independent studies in 2026 found that two runs of the same prompt share only about 40–60% of the brands they name, and SparkToro's test saw the same brand list come back in under 1% of runs. A single answer is noise; the rate across many answers is the stable unit.
So optmz pools answers across questions, runs and a rolling four-week window, and shows each rate with the number of answers behind it and a 95% confidence range. A question and engine with fewer than 20 answers shows “Collecting” instead of a percentage, and a week-over-week change is called a change only when a direct test on the difference says it is unlikely to be noise. The guide on measuring AI visibility without fooling yourself explains why.
What you can do with the numbers
- See which buyer questions name a competitor but not the client, ranked by how clear the gap is.
- Open any answer to read the full text, the sources it cited and the web searches the engine ran.
- Compare the client with each competitor on the same answers, per engine or across all of them.
- Turn gaps into ranked fixes in the Action Center, then send a client report.