We measure citation movement on a ±14-day window around every Action. That number isn’t arbitrary — it’s the sweet spot between “noisy” (too short) and “stale” (too long).
How LLM indexes actually update
Different engines use different update cadences:
- Perplexity — refreshes its retrieval index relatively quickly (days to a week for new content to be citable).
- ChatGPT browsing — grounded retrieval on queries that route through Bing/Search; fast.
- ChatGPT non-browsing — relies on training-cutoff knowledge plus tool calls. New content visible only via grounded retrieval until next training round.
- Claude / Gemini — similar mix; depends on whether the query routes through their search tools.
Why 14 days specifically
Shorter than 14 days and random fluctuation dominates — engines give different answers to the same prompt run twice with no intervening change. Longer than 14 days and other actions / competitor activity / training updates start confounding the signal. 14 days is wide enough to catch real propagation, narrow enough to attribute the movement to the Action.
What won’t show up here
Slow-burn effects (e.g. a new article that gets crawled in week 3, cited in week 6) won’t register on this Action’s Impact row. We don’t expand the window indefinitely because the noise dominates after a month or two. For long-tail attribution, watch the Visibility tab’s 90-day chart instead.