AI citation churn is a repeating fact, not a one-off: 43.7% of cited domains vanished in seven days — and multi-engine domains survived at 87%

Method, stated before the numbers

Same instrument as the rest of the citation-economy series: 12 pinned small-business buyer questions × 4 engines (ChatGPT, Gemini, Claude, Perplexity) × 3 identical samples = 144 production answers per run, citations read from each engine's returned citation payload, domains normalized hostname-minus-www — the same rule as every dated cut in the series, so the diffs are like-for-like. Three dated runs now exist: 2026-07-28 (538 distinct cited domains), 2026-08-03 (520), and 2026-08-10 (540). A domain "survives" a window if it is cited anywhere in both runs; it "churns out" if it is cited in the earlier run and nowhere in the later one.

The churn replicated — and so did the multi-engine protection effect

Roughly 45% of the AI citation pool turns over every week, two weeks running. 247 of this week's 540 cited domains (45.7%) were cited nowhere in the corpus seven days earlier. Only 225 domains — 41.7% of the current corpus — have appeared in all three dated runs. A citation observed once is more likely to be a flicker than a foothold, which is the same conclusion our within-run stability cut reached from a different direction.

Engine breadth is the strongest survival predictor we can measure

Last week's headline — half of AI-cited domains vanish in six days, unless two engines cite them — replicated almost exactly: single-engine citations remain a coin flip week to week, while a second engine's citation nearly doubles survival odds. 34 domains climbed into the multi-engine tier this week (cited by ≥2 engines on 2026-08-10 after 0–1 engines on 2026-08-03). This week's breadth histogram: 439 single-engine domains, 64 at two engines, 30 at three, 7 cited by all four.

Full disclosure: our own domain entered the corpus this week

We publish this panel's read on our own domain every week, zeros included, and this week it is not zero for the first time: Perplexity cited askedabout.com in 3 of 3 identical samples of one panel question ("What is a good Profound alternative for a small business?" — a question where we publish an indexed head-to-head comparison). The other 141 answer cells remain zero for us, and ChatGPT, Gemini, and Claude cite us nowhere on the panel. By this post's own base rates, a first-week citation is the start of a test, not a result: single-engine entrants survive the week at 49%, and the 3-of-3 sample stability our new citation shows is the pattern only 28.9% of citations exhibit. The falsifiable claim we are registering: whether this citation survives to the 2026-08-17 run will be published either way.

What this means if you are chasing AI citations

A single week's citation win is mostly noise — 51% of single-engine domains lose it within seven days. The durable asset is breadth: being cited by a second engine moved week-over-week survival from 49% to 87% in both measured windows. If you are auditing your own AI visibility, ask not just "am I cited?" but "by how many engines, and did it repeat?" Both dated cuts behind this post are machine-readable JSON in our public repo; the method is fixed and the next re-run is 2026-08-17.