How to Track AI Citations: Mentions, Share of Voice, and What Counts

Why Your Rank Tracker Can't See This

Traditional SEO measurement is mature — rank, traffic, CTR, conversions. None of it tells you whether ChatGPT cited your client, how Claude describes them, or which Perplexity answer used their page as a source. AI search doesn't have rankings; it has citations, mentions, and entity associations. Different scoreboard, different measurement layer.

And most teams aren't measuring it at all: 71% of marketing teams have no AI visibility tracking in place — CapstonAI's 1,240-brand report. If you start now, you're ahead of seven out of ten competitors by default.

The Three Layers (Don't Conflate Them)

Mentions — the engine names your brand, no link. Baseline existence signal. Mention rate = % of prompts where you're named.

Citations — the engine links a URL on your domain as a source. Citation rate = % of prompts where you're linked. This is the evidence-source tier.

Cited URLs — which pages get cited. The operational layer: are pillars, posts, or product pages earning the citations? And which third-party domains win when you don't?

The gap between mention rate and citation rate is the diagnostic: mentioned but not cited = structure problem (content isn't extractable). Not mentioned at all = entity problem (the engines don't know you). Two different problems, two different fixes — which is why measuring them separately is the load-bearing step.

The Metrics That Matter

• Mention rate — brand presence in answers
• Citation rate — evidence-source presence
• Share of voice — your citations ÷ all citations on your query set, vs competitors. The metric clients understand instantly
• Coverage — prompts where you appear at all (mention or citation)
• Trend — 6–12 month direction, because single-month snapshots are noise

Manual Tracking: The Irreplaceable Baseline

Low-tech, unglamorous, and still the most accurate method in existence.

Step 1 — build the query list. 20–40 questions your ideal customers actually ask AI engines. Rephrase your GSC queries conversationally, add People-Also-Ask and Reddit questions, steal from support tickets. For the four prompt categories worth tracking — discovery, accuracy, money-keyword, competitor-gap — see the framework in our AEO pillar.

Step 2 — run them everywhere. ChatGPT (web search on), Perplexity, Claude, Google AI Overviews. Per engine per query: mentioned? cited? which URL? in what context? Perplexity's transparent citations make it the easiest — our Perplexity guide covers why.

Step 3 — log it. Spreadsheet: query, engine, date, mentioned Y/N, cited Y/N, cited URL, competitor cited, notes. Weekly for top 20, monthly for the full set. A 20-query audit takes under an hour; re-runs take 15 minutes.

Step 4 — calculate. Monthly: mention rate, citation rate, share of voice, coverage. Look for patterns — which engines cite you, which queries you're winning, which competitors eat your citations when you lose.

When Manual Breaks: The Tool Market

Thirty queries × four engines × ten clients = 1,200 checks a month. Manual dies there. The market as it stands:

The honest framing for the whole category: trackers tell you whether you appeared. Almost none tell you whether the appearance produced revenue — that gap is why our attribution guide exists.

Turning Data Into Decisions

• Find the gaps. Queries you should win but don't → optimization priorities.
• Diagnose correctly. Mentioned-not-cited = restructure (answer-first, FAQ blocks). Not-mentioned = entity work (entity SEO) and third-party presence.
• Watch competitors. Who gets cited instead of you, with what content, corroborated where. Their playbook is your audit.
• Report share of voice. One "AI Search Visibility Score" aggregating citations, share, and mention accuracy — clients get a number they can watch, not a lecture on RAG pipelines.

Frequently Asked Questions

Can I track AI citations in GA4?

Partially — referrer traffic from perplexity.ai and chatgpt.com shows up, but a median ~34% of AI traffic lands in Direct. GA4 is a signal, not the picture. Query testing is still the reliable measurement.

How often should I track?

Weekly for top 20 queries per client, monthly for the full set. Six-to-twelve-month trendlines are the signal; single snapshots are noise.

Mention vs citation — why does the distinction matter?

Mentioned-not-cited means the engines know you but can't extract you — fix structure. Not mentioned means they don't know you — fix entity. Treating them as one number means treating the wrong problem.

What's the minimum viable tracking setup?

A spreadsheet, 20 queries, four engines, weekly. Free, an hour a week, and ahead of the 71% of teams tracking nothing.

The citation-monitoring agent is one of Altyzo's agents in the Altyzo platform — $30K license, runs on your infrastructure. For brands, Altyzo Managed does the tracking and the fixing.