Research · Dispatch #1 ·

AI engines cited this page before Google indexed it

Three AI engines cited our citation-precision page while Google Search Console still showed it unindexed. An honest look at the co-occurrence, with caveats.

Welcome to the first GEO Glossary dispatch: a roughly monthly note on what we actually measure about AI-search citation. Every week we probe ChatGPT, Perplexity, Claude, Copilot, and Gemini by hand, with the same prompts, and record which pages each one cites. No vendor spin, no growth hacks. Just the data and one finding.

The data point

Our page on citation precision and recall went live on May 29, 2026. By June 1, three engines (ChatGPT, Perplexity, and Gemini) had each cited it as a source. In our June 3 round, ChatGPT and Perplexity still cited it (at positions 3 and 1); Gemini did not that run, which is the normal volatility of these systems.

Google, meanwhile, has crawled the same page and, as of June 3, still lists it as "Crawled, currently not indexed." Five days after publish, AI engines were citing a page that Google had not yet put in its index.

The finding: cited by AI, absent from Google's index report

The interesting part is the mismatch. AI engines were citing the page while Google Search Console still showed it as not indexed. For a publisher, the implication is direct: if you grade your AI-search performance only by Search Console (indexing, rankings, impressions), this page reads as a non-event, while in AI answers it was already one of our best performers.

Two honest caveats, stated up front. First, part of this gap is almost certainly that Google is slow to index new, low-authority sites, and this page may well get indexed later. Second, and more important for the framing: Search Console's index-coverage report is not real-time and can lag by several days, so we cannot prove a strict sequence ("AI cited it before Google indexed it") from this snapshot alone. The honest claim is co-occurrence (AI cited the page while Search Console showed it unindexed at the time we checked), not a proven temporal ordering. Both caveats are exactly why the follow-up matters: we are tracking whether and when Search Console later reports this page indexed, which is the only way to settle the sequence. Either outcome is a future data point, not a retraction.

Why this happens (the mechanism)

AI engines do not depend on Google's index. Perplexity runs its own crawler and index; ChatGPT's search uses Bing's index plus OpenAI's own crawling; live page-fetching at answer time is common across tools. So an engine can cite a page Google has not indexed because it is retrieving from a different substrate entirely.

The exception worth watching is Gemini, which grounds on Google's own Search. That is exactly why the two engines carrying this finding are the non-Google ones (ChatGPT, Perplexity), and why Gemini's June 1 citation of a Google-unindexed page is the curious case we will keep an eye on rather than lean on.

This month's round (June 3)

Same prompt, web search on, logged out. The live per-term, per-engine view is on the Observatory.

  • Perplexity cited 3 panel terms, all as the #1 source.
  • Claude cited 3 (inline).
  • ChatGPT cited 1.
  • Copilot and Gemini cited 0 this round.

Across the whole glossary, Google has now indexed 62 pages.

Method notes, because they matter

  • We ask each engine to cite its sources, so this measures elicited citation (cited-when-asked), not spontaneous.
  • Counts are small and citation is volatile; one capture is not permanent. Note the distinction above between "cited on June 1" and "cited in the June 3 round."
  • A frozen probe measures one query on one run, not "ever cited." This month a real visitor reached our Gemini citation page from inside Gemini, yet our standard probe of that page did not reproduce it. Fixed-prompt probing has a systematic blind spot: it misses real surfacings that referral logs catch. That is a finding about the method itself.
  • What this is and isn't: the citation-precision case above is one vivid example, not a systematic result. The systematic version is a cross-tab across all 76 pages: of the pages AI engines cite, what share has Google not indexed, and vice versa. We have the indexing data (62 of our pages) and are building that table. Treat the single case as the dispatch-grade signal; the cross-tab is the publishable one.

Next month

Why some of our practitioner-coined terms rank on Google and others do not. The difference turns out to be whether the coined term collides with a meaning people already search for: cite-ability sits near the top of Google's results, while "LLM optimization" sits on page 8, and both are coined terms. Plus the next citation round, and whether Google has indexed citation-precision yet.

This is a dated snapshot, not a permanent claim: the data is current as of 2026-06-03, and the point of the series is to watch it change. GEO Glossary is a free, vendor-neutral reference for AI-search vocabulary, with per-term citation status tracked across five engines on the Observatory. Subscribe to get future dispatches.

More dispatches