An AI visibility tracker that shows its sample size

What it measures, and how

No single source sees every AI answer, so the tracker reads five and never adds them into one score. Each answers a different question, and they're expected to disagree: an answer can cite a page nobody clicks, and a crawler can read a page no answer cites.

Probe sweeps

Each day, questions a site should answer go to Perplexity, ChatGPT, Gemini and Claude; the sites each answer cites are recorded. Each question counts once per sweep, as cited if any answer cited the site. Questions that name the brand are left out of the rate, because asking about yourself measures the question, not the market.

Bing grounding citations

Bing Webmaster Tools reports how many times Microsoft's AI answers cited the site and, for each question, the share of grounded answers that cited it. Of the five, it's the only citation count an engine reports to the site owner.

Search Console AI features

Search Console has a separate report for Google's generative AI features. It gives impressions only, with no clicks, no queries and no API. Growth Radar reads it from the Search Console screen and keeps it beside the ordinary search totals, never inside them.

AI-referred visits

When someone clicks a cited link, the visit arrives with the assistant as its referrer. Growth Radar recognizes those referrers from published lists of AI assistant hosts and counts the visits in the site's own analytics over the last 28 days. A visit is proof that a citation existed.

AI crawler hits

Server-side hits from GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended, counted only when the request is verified as coming from the operator the user agent names. A request that merely borrows the name is dropped as an impersonator.

Why a rate needs its n

The same question asked twice can return a different set of sources. A tracker that asks each prompt once and prints a bare percentage turns that noise into a number that looks precise. Growth Radar treats every answer as one draw: it reports how many questions were asked, how many answers cited the site, and the range that sample allows.

A site with no reading is shown as not measured, never as zero. Not connected, no traffic, a failed import and an import that hasn't run yet are four different states, and the tracker keeps them apart.

When you ship a change aimed at AI citations, it's recorded with the reading at ship time and graded 28 days later the same way a change aimed at clicks is.

What it can't see

• AI engine coverage is narrow: Perplexity, ChatGPT, Gemini and Claude, checked daily.
• Probe questions come from Bing's report of what people asked AI assistants where the site could have been cited, then from the site's niche keywords, then from its top Search Console queries. They're a sample of at most eight per sweep, not a census of what people ask.
• No share of voice, mention position or sentiment score: they need the full text of many answers naming competitors, and estimating them from a few would be a guess.
• Bing's AI citation report and Search Console's AI-features report are AI reports: they come with Pro and Agency, and with the 14-day free trial. Free keeps 1 site and no AI reports.

Growth Radar or another tool

Dedicated prompt trackers cover more engines and more prompts than Growth Radar does. Each claim below comes from the vendor's own page, linked. Checked October 28, 2026.

Sources: Peec AI pricing · Peec AI Actions · Profound pricing · Profound pricing, archived 14 September 2026 · Profound Prompt Volumes · Otterly.AI pricing · Semrush blog: AI visibility tools for agencies