The citation cliff is the observed pattern, dating to early 2025, that pages older than 90 days lose their AI Overview and Perplexity citations at an accelerating rate. The drop is steeper for pages on competitive commercial queries, where fresh content from competitors keeps arriving. AEO programs that produced a flagship piece in 2024 and stopped maintaining it have watched citations evaporate. The lesson: AI citation share is a continuously-funded asset, not a one-time win.
Why retrievers are weighting recency
Two forces drive recency weighting. First, on factually-evolving topics, the retriever wants to avoid serving stale claims, and publication date is the cheapest available proxy for staleness. Second, the major engines are explicitly biased toward newer pages because that is what their freshness benchmarks reward — a 2024 paper from Ahrefs documented that ChatGPT preferentially cites newer pages even when older pages have higher topical authority. The combined effect is a strong recency tilt baked into modern retrieval.
What the curve looks like
Internal data across several B2B SaaS sites shows roughly the following shape. Pages 0 to 30 days old: full citation share. Pages 30 to 90 days: a 10 to 20 percent decline as competing fresh content displaces. Pages 90 to 180 days: a 30 to 50 percent decline relative to peak. Pages over a year: usually under 30 percent of peak citations unless the page has been substantially updated. The cliff is steepest between 90 and 120 days, which is where most maintenance gets deferred indefinitely.
What counts as keeping a page alive
A genuine update — adding new sections, refreshing entity references, integrating recent data, updating dateModified in the schema — restores citation odds substantially. A cosmetic touch (changing dateModified without changing content) sometimes works briefly but tends to be detected and devalued by the retriever within weeks. The most resilient pattern is to ship a real content update every 60 to 90 days on flagship pages: a new section, a fresh data point, a quote from a recent customer.
Implications for content strategy
The 3-month cliff fundamentally changes how content investment should be planned. A piece written and published once is worth far less than the same piece refreshed quarterly. That argues for fewer flagship pages maintained relentlessly, rather than many pages published and abandoned. The team that keeps 30 evergreen pages updated quarterly will out-cite the team that publishes 100 pages a year and never revisits any of them.
How to operationalize maintenance
Pick your top 20 to 30 highest-citation pages. Set a calendar reminder per page on a 60-day cycle. Each cycle, do one real edit: add a paragraph, refresh a number, integrate a new entity. Update both the visible content and the schema's dateModified. Run a citation re-scan a week after the update. Over a year, this is roughly two days of writing per quarter — modest investment for a major share of your AI search visibility.