# How stable are AI citations between runs?

Less stable than most reporting assumes. Asked the same question minutes apart, a single engine's citation list reproduced only about 58–62% of itself; domains that several engines agreed on held up far better, which is why agreement, not presence, is the unit worth reporting. The practical consequence: one check cannot establish a baseline, and a change in a single number is not yet movement.

## Key points

- Reproducibility is measurable and should be stated next to the figure it qualifies. A citation count without a repeat measurement is a single observation.
- Prefer multi-engine agreement as the unit. It is the smallest thing that survived a repeat measurement in our own corpus.
- Diff the agreeing set, not the whole set. Raw citation sets churn between checks (a real check of one subject moved 13 domains in and 15 out between two runs), so the raw diff measures noise first and direction second.
- Treat any single-run change as unconfirmed until the next check agrees with it — including changes that look like wins.
- State the interval. Two checks a day apart and two checks a month apart support different conclusions, and the interval belongs in the report.

## Frequently asked questions

**Q: Does that mean AEO measurement is not worth doing?**
A: It means single measurements are not worth much. Repeated measurement, engine-agreement counts and compared sets are exactly what make the answer usable, and they are the part that cannot be reconstructed after the fact.

**Q: How many engines should a check use?**
A: Enough that agreement means something, and few enough that the check stays affordable: two to three engines per check is where our own measurements put the cost-benefit line, with the rest rotated across checks.

## Sources

- https://autoaeo.work/aeo/answers/which-pages-ai-engines-cite.md
- https://autoaeo.work/aeo/answers/share-of-answer.md
- https://autoaeo.work/aeo/product.md

Updated 2026-10-07. Canonical source: https://autoaeo.work/aeo/answers/citation-stability.md