How-to

📊 Measure whether AI is actually mentioning you

Track AI visibility with a monthly scorecard of 10 buyer questions across every engine — scoring mentions, citations, and sentiment in a sheet — plus GA4’s AI-referral channel to catch the traffic that lands.

🎯 Why it matters

You cannot improve what you do not measure, and because the same prompt returns a different answer day to day, AI visibility needs a monthly scorecard, not a rank tracker. Here is the sheet, the prompt, and the GA4 setup to run it.

💬 My take

Build a monthly AI-visibility scorecard, not a rank tracker

AI does the sweep here: it asks your buyer questions across every engine and tallies who got mentioned. What it cannot do is watch the trend for you. That is a simple sheet you keep month over month, plus GA4 to catch the traffic that actually lands.

Fastest path: one prompt, end to end

Run this once a month, once per engine (ChatGPT, Perplexity, Gemini, Google AI Mode). Their sources barely overlap, so one combined check hides most of the picture:

🤖 AI prompt — paste into ChatGPT / Claude

You are running a monthly AI-visibility sweep for [MY BRAND] in [CATEGORY]. Competitors: [COMP A], [COMP B], [COMP C].
1. Ask these 10 buyer questions with web search ON: [PASTE 10 REAL BUYER QUESTIONS].
2. For each, record ONLY what actually appeared: which brands were named, which domains were cited (with links), and the sentiment toward [MY BRAND] (positive / neutral / cautionary / not mentioned).
3. Output ONE markdown table: Question | [MY BRAND] mentioned? | [MY BRAND] cited (linked)? | Sentiment | Competitors named | Domains cited.
4. Below it, give three totals: mention rate (% of 10 questions naming me), citation rate (% linking me), and share of voice (my citations divided by all brand citations across the 10).
Do not guess, estimate, or fill gaps. Every row must reflect this session's real answers. If you cannot browse, stop and tell me. Record only what appeared.

Paste the totals into a sheet, one column per month, and watch the three numbers move.

Or do it in 5 steps

Write 10-15 real buyer questions (start small; grow to 50-200 once the habit sticks).
Run the prompt monthly, per engine. Citation sources overlap little between engines: in one audit only about 1 in 10 domains cited by ChatGPT were also cited by Perplexity.
Log mention rate, citation rate, and share of voice per engine in a sheet. Share of voice (your citations divided by all category citations) is the one that shows how you are doing against competitors.
In GA4, turn on the native "AI Assistant" channel (launched May 2026, auto-tags ChatGPT, Perplexity, Gemini, Claude, Copilot via medium=ai-assistant). It misses roughly 30% of sources and is not retroactive, so also build a custom channel group: Admin, then Channel Groups, then new group; add an "AI Traffic" channel matching Source by regex for chatgpt.com, perplexity.ai, gemini.google.com, claude.ai, copilot.microsoft.com, and drag it ABOVE Referral so it evaluates first (source: OrganikPI, 2026).
Treat every number as directional. Independent tests found commercial trackers undercounting real mentions, and note that AI Overview and AI Mode clicks inside Google Search count as normal Google organic in GA4, not AI Assistant. Done = you have two consecutive months to compare.

Sample scorecard (format example, not real data)

Question	Mentioned?	Cited (linked)?	Sentiment	Domains cited
best [category] for [use case]	yes	no	neutral	reddit.com, g2.com
[competitor] alternatives	yes	yes	positive	yourdomain.com, capterra.com
is [product] worth it	no	no	not mentioned	reddit.com, youtube.com

Mention rate 2/3, citation rate 1/3. Fill these with your own real answers.

Review monthly.

⚖️ Do & Don't

Do

Build a fixed panel of 50-200 real buyer prompts for your category and re-run it on a schedule (weekly is typical) across ChatGPT, Perplexity, Gemini, and Google AI Overviews separately.
Track share of voice (your citations divided by total category citations across your panel), not just raw mention count, so you can see how you're doing against named competitors.
Turn on GA4's AI Assistant channel to segment ChatGPT/Gemini/Claude referral traffic, and manually tag Perplexity and Copilot referrers since GA4 doesn't cover those automatically.
Spot-check any tool's numbers by manually running a sample of the same prompts yourself — independent tests have found real undercounting in commercial AI-visibility trackers.

Avoid

Don't expect a single "AI rank" number — the same prompt can return a different answer on the same engine from one run to the next, so treat every reading as a sample.
Don't track only one engine — citation sources overlap so little between engines that single-engine tracking hides most of the real picture.
Don't assume "no referrer" traffic in GA4 is direct traffic — a large share of AI-driven visits arrive with no referrer header and get silently misclassified.
Don't choose a tracking tool on citation-count claims alone — ask what prompt set and methodology it uses, since backward-looking, keyword-based tools miss real prompts your buyers actually type.

💡 Quick tips

Start manually and free: run your 10 highest-intent buyer questions in ChatGPT, Perplexity, and Google AI Mode yourself, once a week, before paying for a tool.
When you do pick a tool, run its numbers against your own manual spot-check for the first month, and treat the gap between the two as its real margin of error.

🏢 Brand in focus

BrandBrand Radar tracks visibility across 405M+ prompts and 6 AI platforms

Ahrefs — turned AI-visibility measurement into a product, warts and all

Ahrefs · Why it fits: Ahrefs built Brand Radar as a dedicated AI-visibility tracker, modeling share of voice, mentions, and citations across ChatGPT, Perplexity, Gemini, Copilot, and both Google AI surfaces at a scale (hundreds of millions of modeled prompts) most brands couldn't run themselves. Did well: it published its own methodology and a stream of large-sample studies, including one showing AI Overviews now cut top-ranking-page click-through by 58%, that became reference data for the whole GEO industry. Watch-out: independent testers found Brand Radar undercounting real mentions badly in side-by-side checks — a reminder that even a leading vendor's numbers are modeled and backward-looking, not a live read of what AI engines actually said to your buyers, so verify against your own manual spot-checks.

📚 Go deeper — resources

article

What is Brand Radar and how to use it

article

AI Visibility Toolkit

article

GA4 AI search referral attribution

🏷️ Tags

geoai-searchmeasurement

🔗 Related