AI Summary
The AI search analytics stack that actually matters is a seven-metric framework spanning citation share, share of voice, source diversity, crawl-to-citation lag, AI-referral conversion, branded search lift, and citation velocity. Each metric maps to a different layer of the measurement problem, and together they form a closed decision loop. Our May 2026 analysis of 153,425 citations found that 76.95 percent of URLs cited by AI engines do not appear in the organic top-10, which means your existing rank-tracking stack is missing the majority of what matters.
The sister post on AI brand visibility tracking covers the four core tracker metrics, the competitive intelligence tools roundup covers which platforms instrument them, and the share of voice vs search volume post covers the concept shift. This post is the instrumentation layer: how to collect and act on each metric.
Why seven metrics and not one
Each metric captures a different failure mode. Citation share tells you whether you are present. Share of voice tells you whether you are winning relative to competitors. Source diversity tells you whether your visibility is structurally fragile. Crawl-to-citation lag tells you whether your content pipeline is healthy. AI-referral conversion rate tells you whether the traffic you do earn is worth anything. Branded search lift tells you whether your AI exposure has real downstream demand effects. Citation velocity tells you whether your authority is improving or eroding in real time.
Run one metric in isolation and you get a misleading picture. A brand with high citation share but low source diversity is one editorial decision away from losing half its AI visibility. The framework works because the seven metrics check each other.
Metric 1: Citation share by query
Citation share by query is the percentage of tracked prompt runs where your domain appears in the answer set for a specific query, measured per AI engine. It is the closest structural analogue to a traditional ranking position. Unlike rank, it is probabilistic: the same query run ten times on the same day will produce different citation appearances.
How to instrument it. Define a stable prompt panel of 50 to 200 priority queries, covering branded, comparison, and category-intent variants. Run each prompt against ChatGPT, Perplexity, Gemini, Copilot, and AI Overviews. Log presence or absence per run. Average across runs to get a rate, not a binary. The open-source GEO/AEO Tracker handles this across six engines with scheduled runs and CSV export. Run weekly per engine for every query in the panel.
Metric 2: Share of voice vs named competitors
Share of voice normalises citation share against your competitive set on the same prompt panel. If you appear in 23 percent of runs for a query and the top competitor appears in 41 percent, that gap is your optimization target. Share of voice is more actionable than absolute citation count because it is unaffected by query volume shifts or engine update cycles.
The SOV vs search volume post covers the strategic context. The short version for measurement: track share of voice per query cluster, not just brand name. A brand can dominate general awareness queries and be absent from all buying-intent queries. That gap is where competitors close deals.
Metric 3: Source diversity
Source diversity measures which URLs and domains AI engines use to support mentions of your brand, not just whether your own pages are cited. In our May 2026 study of 153,425 citations, YouTube appeared in 9,868 citations and Reddit appeared in 6,595 citations. AI engines frequently cite your brand through third-party content rather than directly from your site.
How to instrument it. When logging citation runs, capture the full cited URL list per answer, not just your domain. After four weeks, group cited URLs by domain and calculate what percentage of your brand’s AI mentions come from each. Concentration risk is real: if 60 percent of your AI mentions come from a single third-party source, one editorial or moderation change collapses your visibility. Healthy source diversity means no single external source accounts for more than 25 percent of your brand’s AI mentions.
The source map is also your off-site content roadmap. If Reddit threads power your category answers, you need a Reddit presence strategy. If Tier-1 publications dominate, you need earned media in those outlets.
Metric 4: Crawl-to-citation lag
Crawl-to-citation lag is the elapsed time between publishing a new page and its first appearance as a cited source in AI answers. It is a leading indicator of AI authority. A domain with high authority earns citations within days of publishing. A domain with structural problems may publish strong content that never gets cited at all.
How to instrument it. Log every published URL with a precise timestamp. Run weekly prompt suites covering queries the content is targeting. Record the first date each URL appears as a cited source per engine. Calculate days-to-first-citation per URL, then average across the cohort published in the same month. Track the monthly trend, not the absolute number.
Interpreting the trend. Lag getting shorter after a topical authority push means your structural changes are working. Lag getting longer month over month is an early warning of declining technical health: blocked AI crawlers, broken schema, or thinning content. Check server logs for GPTBot and PerplexityBot hit rates before assuming a content problem.
Metric 5: AI-referral conversion rate
AI-referral conversion rate measures what percentage of sessions arriving from AI platforms complete a goal event. It is the commercial translation of citation share into business outcomes. Adobe Analytics data from the 2025 holiday season showed AI-referred traffic converting 31 percent higher than other traffic sources, with AI-referred users spending 45 percent more time on-site and viewing 13 percent more pages per visit. The user arriving from an AI assistant has already been through a research and recommendation process before clicking, which compresses the consideration stage.
How to instrument it. Set up a custom channel group in GA4 with an AI source regex placed above Referral in the ordering. The GA4 AI search referral attribution guide covers the exact regex and setup steps. Once the channel is clean, build a free-form Explore report filtered to AI Traffic. Track conversion events, goal completions, and revenue per session month over month. Sessions arriving via utm_source=chatgpt.com already carry source attribution; referrer string parsing catches the rest.
Metric 6: Branded search lift
Branded search lift measures the increase in Google Search volume for your brand name following a period of elevated AI citation activity. Users who encounter your brand in a ChatGPT or Perplexity answer frequently verify via Google before taking action. This creates a measurable downstream signal in GSC and Google Trends data even when direct attribution from AI to conversion is impossible.
How to instrument it. Pull weekly GSC data filtered to branded queries (exact brand name and variants). Overlay it against your citation share trend from the same period. A consistent 3 to 6 week lag between a citation share increase and a branded search increase is the expected causal pattern. Google Trends confirms directional movement; GSC provides absolute impression counts. This is the most reliable indirect proxy for AI exposure when referral attribution is incomplete.
Metric 7: Citation velocity
Citation velocity combines crawl-to-citation lag with citation count growth rate to measure the acceleration of your AI authority over time. A domain gaining citations faster month over month is in the compounding zone: each new citation increases the authority signal that makes the next citation easier to earn. A domain with stable citation share but slowing velocity is a plateau risk.
The citation velocity measurement framework covers the five-step tracking protocol in full. For this metric’s place in the stack: velocity is the leading indicator that citation share is the lagging indicator of. Watch velocity first. If velocity is accelerating, citation share will follow in four to eight weeks. If velocity is flat or declining while you are publishing actively, the problem is structural rather than one of content quantity.
Our May 2026 study found 74.9 percent of cited sentences appear in the first half of the document, with a mean cited position of 37 percent. Content that front-loads its key facts earns citations faster, because retrieval models do not parse deep to find the extractable claim. Velocity responds directly to positional placement.
Instrumenting all seven in one stack
The instrumentation stack does not require seven separate tools. Most of the data comes from three sources: a prompt panel runner (for metrics 1, 2, 3, 7), server analytics with referrer parsing (metric 5), and GSC/Google Trends (metric 6). Crawl-to-citation lag (metric 4) comes from correlating your CMS publish timestamps with the first appearance dates from your prompt runner.
| Metric | Primary data source | Minimum cadence | Leading or lagging |
|---|---|---|---|
| Citation share by query | Prompt panel runner (GEO/AEO Tracker) | Weekly | Lagging (4-6 weeks) |
| Share of voice vs competitors | Same prompt panel, all cited domains | Weekly | Lagging (4-6 weeks) |
| Source diversity | Prompt panel URL log | Monthly | Structural (slow-moving) |
| Crawl-to-citation lag | CMS timestamps + prompt panel first-cite dates | Monthly cohort | Leading (2-4 week signal) |
| AI-referral conversion rate | GA4 custom channel group | Monthly | Lagging (business outcome) |
| Branded search lift | GSC + Google Trends | Weekly | Lagging proxy (3-6 weeks) |
| Citation velocity | Prompt panel + CMS timestamps | Monthly trend | Leading (4-8 weeks ahead, our estimate) |
The decision loop
The seven metrics form a closed decision loop rather than a passive dashboard. Each combination of metric signals maps to a specific action:
- Lag high, share flat, velocity slow: structural problem. Audit crawler access, schema coverage, and sentence-level structure. The atomic sentence format and readability scoring are the first fixes.
- Share rising, velocity accelerating, branded search up: compounding zone. Document the content pattern (topic, structure, length, internal linking) and replicate it across adjacent clusters.
- Share high, velocity good, conversion flat: landing page and internal linking problem. AI traffic lands on editorial pages without a clear path to commercial ones. Fix interlinking from citation-heavy pages to service pages.
- Source diversity concentrated: off-site authority risk. Diversify the source map regardless of other metric health.
For teams starting from zero: set up GA4 referral attribution first (metric 5 data is immediate), stand up a prompt panel in the GEO/AEO Tracker (metrics 1, 2, 3, 7 in one step), and run a first citation velocity cohort after one month. That baseline is enough to run the decision loop.
The GEO KPI measurement framework and the AI citation tracking service extend this stack to enterprise query volumes once the DIY setup has scaled as far as it goes.
You can pull most of these analytics automatically with a purpose-built GEO tool.