Measurement & Analytics

Measuring AXO success requires tracking how AI agents discover, process, and reference your content. Unlike traditional web analytics, agent measurement focuses on citation rates, content quality scores, and authority signals.

Search Console's Generative AI Report

Start here. Launched June 3, 2026, this is the first first-party AI visibility data most sites can get, it is free, and it is already sitting in an account you own. For a long time the honest answer to "how much do we show up in AI Overviews?" was "nobody can tell you." This closes part of that gap.

Full walkthrough: the Search Console Generative AI report, explained. Google's announcement: Search Central blog, June 2026.

Citation Tracking

Agent Citation Monitoring

Track when and how AI agents reference your content:

# Nginx/Apache log analysis
grep -E "(GPTBot|OAI-SearchBot|ChatGPT-User|OAI-AdsBot|ClaudeBot|Claude-SearchBot|Claude-User|Google-Extended|Google-Agent|PerplexityBot|Perplexity-User)" access.log

# Count crawls by user agent
awk '$12 ~ /GPTBot/ {count++} END {print "GPTBot crawls:", count}' access.log

// GA4 Enhanced Measurement
gtag('config', 'GA_MEASUREMENT_ID', {
  custom_map: {
    'custom_parameter_1': 'user_agent_type',
    'custom_parameter_2': 'crawler_category'
  }
});

// Track AI crawler visits
gtag('event', 'ai_crawler_visit', {
  'user_agent_type': 'GPTBot',
  'crawler_category': 'training',
  'page_path': window.location.pathname
});

-- Track AI crawler trends vs content updates
SELECT 
  DATE(event_timestamp) as date,
  event_name,
  COUNT(*) as events,
  COUNTIF(user_agent LIKE '%GPTBot%') as gpt_crawls,
  COUNTIF(user_agent LIKE '%PerplexityBot%') as perplexity_crawls,
  COUNTIF(user_agent LIKE '%ClaudeBot%') as claude_crawls
FROM `your-project.analytics_123456789.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20260101' AND '20261231'
  AND event_name = 'page_view'
GROUP BY date, event_name
ORDER BY date DESC;

// Shape of a citation-tracking record. Values are illustrative:
// there is no API that returns these numbers. You assemble them
// from AI-visibility tooling, server logs, and manual spot checks.
interface CitationReport {
  periodStart: string          // ISO date
  periodEnd: string
  totalCitations: number       // from your AI visibility tool
  byEngine: Record<string, number>   // engines diverge widely — keep them separate
  topCitedUrls: Array<{ url: string; citations: number }>
  crawlerHits: Record<string, number> // from server logs, by user agent
}

// Track engines separately rather than summing them. Research in 2026
// found only ~11% of domains cited by both ChatGPT and Perplexity
// overlapped, so a single blended "citation count" hides more than
// it reveals.

Citation Volatility: Why One Check Proves Nothing

Before you act on any citation check, understand that the same prompt does not reliably return the same sources. Research published in July 2026 documented ChatGPT drawing on multiple retrieval pipelines, with roughly 11.6% of prompts switching primary source between runs and producing substantially different URL sets as a result.

AI Share of Voice — and Its Big Caveat

AI Share of Voice is the headline metric practitioners have converged on: roughly, how often your brand appears in AI answers for a defined set of prompts, relative to competitors. It is a reasonable thing to want to know, and it is the number most tools lead with.

A note on vocabulary, since it trips people up here more than anywhere else. “AI visibility” settled as the name for the metric, which is why every tracker uses it, while AEO and GEO stuck as names for the practice. That split is the whole reason the terminology feels unresolved — we unpack it in the post on what to call this.

The AI-Visibility Tool Market

These are categories with examples, not endorsements — we have not benchmarked these tools against each other, and given the lack of a standard Share of Voice formula, a like-for-like benchmark would be difficult to construct honestly. What is worth knowing is the shape of the market: it consolidated hard in 2026, and the independent tool you evaluate this quarter may be a feature inside a larger suite by the next.

Agent Analytics

Separating Agent Traffic

Distinguish between human users and AI agent crawlers:

# Quick grep for AI crawlers
grep -E "(GPTBot|OAI-SearchBot|ChatGPT-User|OAI-AdsBot|ClaudeBot|Claude-SearchBot|Claude-User|Google-Extended|Google-Agent|PerplexityBot|Perplexity-User|Applebot-Extended|meta-externalagent|Amazonbot|Bytespider|CCBot|MistralAI-User|DuckAssistBot)" access.log

Attribution Reality: GA4 Undercounts AI Traffic

Building AI-referral channel groups in GA4 is standard practice and you should do it. But you need to know its failure mode before you report the numbers to anyone: 35–70% of AI-referred sessions arrive with no referrer and therefore land in Direct. Depending on the mix of engines and clients sending you traffic, GA4 can undercount AI traffic by roughly 2x.

The cause is mundane: many AI clients are native apps or strip the referrer on outbound links, so there is nothing for GA4 to attribute. This is not a tagging mistake you can fix — it is a ceiling on what referrer-based attribution can see. Report your GA4 AI number as a floor, not a measurement.

chatgpt.com
perplexity.ai
gemini.google.com
copilot.microsoft.com
claude.ai

Content Quality Scoring

Automated Quality Assessment

Score content based on agent-friendly characteristics:

// Content quality scoring system
function calculateAXOScore(content, metadata) {
  const scores = {
    structure: assessStructuralQuality(content),
    clarity: assessContentClarity(content),
    completeness: assessInformationCompleteness(content),
    trustSignals: assessTrustSignals(metadata),
    technicalImplementation: assessTechnicalQuality(content)
  };
  
  const weightedScore = (
    scores.structure * 0.25 +
    scores.clarity * 0.20 +
    scores.completeness * 0.20 +
    scores.trustSignals * 0.20 +
    scores.technicalImplementation * 0.15
  );
  
  return {
    overallScore: Math.round(weightedScore),
    componentScores: scores,
    recommendations: generateRecommendations(scores)
  };
}

ROI Measurement

Business Impact Metrics

Connect AXO improvements to business outcomes:

Measure Conversion Rate, Not Session Volume

If you judge AI referrals by session count, you will conclude the channel does not matter. It is a small share of sessions for almost everyone. That is the wrong denominator.

Multiple 2026 analyses found AI referral traffic converting several times better than organic search — Semrush put the gap at roughly 4.4x. The intuition behind it is straightforward: someone who arrives after an assistant has already answered their question and recommended you has done their evaluation before they clicked. They land further down the funnel than a search visitor.

Performance Monitoring

Automated Quality Alerts

Set up monitoring for content quality degradation and technical issues:

• Quality score drops below threshold
• Schema markup validation failures
• Citation rate decreases
• Agent crawl errors or blocks

What to Actually Track

You do not need a large stack. You need four instruments that fail in different directions, because each one is blind to something the others catch. Here is the recommended minimum, with an honest statement of what each cannot tell you.