Citation tracking

Each term page displays five citation badges (ChatGPT, Perplexity, Claude, Copilot, Gemini), each showing one of three states: untested, not cited, or cited. A Last checked date sits below the badges. Today most badges read untested because the site is new; states flip as evidence accumulates from the citation-tracking workflow below.

How badge state moves

Four signal sources flow into the citation badges, weighted by confidence:

• Manual multi-probe sweeps. For each entry, a fixed set of probe queries (the term itself, common phrasings of the question the term answers, adjacent queries practitioners would actually type) is run against each engine. The probe set is stored so probes are repeatable; this is the load-bearing signal for badge state transitions.
• Server log + Search Console cross-validation. Vercel server logs are filtered for AI-engine crawler user agents (ChatGPT-User, GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot) and external visitor referrers. Google Search Console data is cross-checked to separate organic SERP clicks from AI-surface citation clicks (a non-trivial distinction in 2026; some AI surfaces are bundled into Google Search Console without a separate breakdown).
• Acknowledged measurement gaps. Some 2026 AI surfaces (notably Google AI Mode) are bundled into Google Search Console Web Search totals without a separate breakdown, per Google's own Search Central documentation. AI Mode clicks contribute to GSC click counts but cannot be isolated from traditional organic results at the vendor-reporting level. Direct manual probes on the AI Mode tab in incognito sessions are the only first-party way to confirm AI Mode citation; combined GSC totals are tracked at the URL level without attempting to subtract AI Mode out.
• SaaS-tool brand-monitoring crawlers as leading indicator. Declared SEO and AI-visibility crawlers (AhrefsBot, SERankingBacklinksBot, Amazonbot, GoogleOther) periodically scan the site. Their attention is not a citation event but a leading indicator that the term has entered SaaS brand-monitoring scope. In our tracking, this signal has sometimes preceded vendor-side citation reporting by weeks; we treat it as a weak leading indicator, not as evidence of citation, and entries attracting multiple distinct SaaS bots are flagged for closer probe attention rather than promoted in badge state.

What the discipline looks like in practice

A single Vercel referrer or a single server-log hit is never enough to promote a badge state. Founder self-clicks, scraper traffic, and Claude Code background shells have all been mistakenly counted in earlier passes; each such over-claim is documented and the disambiguation framework is tightened. We use a 5-axis disambiguation methodology (foreign edge / cache state / path pattern / UA-plus-referer / non-scraper UA pattern) to separate genuine external AI-engine traffic from the four common false-positive sources. Read the per-entry editorial footer for the audit trail.

The badge state is also entry-scoped, not site-scoped: a cited status on one entry does not transfer to a sibling entry. Each entry has to earn its own citation evidence on each engine. This rule has caught us once already (a same-day publish where we incorrectly inherited cited from a cluster sibling; caught during a same-day post-publish audit and corrected within minutes, with the failure pattern added as an explicit pre-commit checklist item).

Why we publish the gaps

Cross-engine measurement infrastructure in 2026 is incomplete. Google Search Console bundles AI Mode into Web Search totals. ChatGPT, Claude, and Perplexity do not provide publisher-facing citation dashboards. Microsoft is the only major engine offering an AI Performance dashboard in Bing Webmaster Tools (public preview, launched February 10, 2026). For everything else, third-party tracking tools (Profound, Otterly, AthenaHQ, Peec AI, Ahrefs Brand Radar) provide partial coverage with different counting rules. We name the gaps explicitly so readers can calibrate trust: where we say cited, the evidence chain is in the editorial footer; where we say untested, we mean we have not run the probe yet, not that the engine has refused to cite.

How this connects to other parts of the site

• The editorial-methodology page describes the upstream workflow that writes each entry; this page describes the downstream workflow that measures how each entry is cited.
• The ai-citation-metrics pillar is the subject-side reference for the six measurement dimensions (attribution rate, citation share, citation match rate, cite-ability, citation velocity, citation rotation) that the per-entry citation tracking instantiates.
• The external-traffic-disambiguation entry codifies the 5-axis methodology mentioned above as a reusable framework.