Live data · build-in-public
Citation Observatory
Transparent first-party data on whether AI search engines cite this glossary. 94 terms × 5 engines × multi-probe sweeps run as capacity allows. Site launched 2026-05-14; first scheduled full review 2026-06-19.
Methodology
Each term carries a sidecar set of probe variations (typically 2-4 phrasings covering the term itself, the question it answers, and adjacent practitioner queries). Every variation is run against each of the five engines; a cell is marked cited if aisearchglossary.comappears in any variation's source list across that engine, partial if the domain is mentioned but our specific entry URL is not the surfaced source, and not cited if no variation surfaces the domain. Untested cells are simply not yet covered in the current sweep. See editorial methodology for the full protocol including signal sources beyond direct probes.
Cells probed
469 / 470
99.8% of the 470-cell matrix
Cells cited
167 / 469
35.6% citation rate among probed cells
Last sweep
Probes run as capacity allows; full sweeps target monthly review milestones.
Citation over time
Distinct terms cited by any of the five engines, per probe round. The solid line is the cumulative count ever cited; the dashed line is this round alone. As it nears the size of the corpus, the cumulative line becomes a lagging indicator — the leading signals are depth and rotation, below.
Terms probed per round (n): 13 / 30 / 13 / 18 / 44 / 21 / 31 / 31 / 36 / 19 / 21 / 17 / 37 / 16 / 19 / 17 / 32. Elicited-probe data — we ask each engine a fixed set of questions and record whether aisearchglossary.com appears in the cited sources, not engine telemetry. The frozen panel began on 2026-06-09; earlier rounds used rotating panels, so the amber gap still mixes true citation rotation with terms simply not re-probed. As more frozen-panel rounds accumulate, the gap becomes a clean rotation signal.
Depth & rotation
Breadth (being cited at all) saturates as the corpus fills in; the signals that keep moving are how deep a citation runs and whether it holds. Depth here is cross-engine consensus: the running count of terms that at least one round saw cited by three, or four, of the five engines at once.
Rotation — of the terms cited in an earlier round and re-probed on 2026-09-07, 17/22 (77%) were cited again. The rest rotated out this round — citation rotation is real and bounded, not decay: cumulative depth keeps rising while individual cells churn.
Depth is single-round consensus, accrued: a term counts toward a tier the first round it clears it, so the lines only rise (a term reaching 4-of-5 once stays counted). This makes depth robust to the panel-size swings that drive the round-by-round line above. Rotation is the honest complement — small early denominators are noisy; the frozen-panel rounds (2026-06-09 onward) are where the percentage means something.
By engine
| Engine | Cited | Not cited | Untested | Cited rate |
|---|---|---|---|---|
| ChatGPT | 48 | 46 | 0 | 51.1% |
| Perplexity | 55 | 38 | 1 | 59.1% |
| Claude | 20 | 74 | 0 | 21.3% |
| Copilot | 23 | 71 | 0 | 24.5% |
| Gemini | 21 | 73 | 0 | 22.3% |