The GEO citation horizon: where recency beats authority

We spent three months tracking 1,200 brand citations across Perplexity, ChatGPT, and AI Overviews. The results contradicted our assumptions. High domain authority and comprehensive llms.txt files mattered far less than one unexpected variable: publication recency. Content older than 90 days saw citation rates drop off a cliff.

The 90-day citation cliff

We partitioned our tracked queries by content age. Pages published within 30 days accounted for 62% of all citations. The 31-to-90-day bracket took another 24%. Content older than 90 days captured just 14%. This pattern held regardless of whether the source was a major publisher or a niche blog.

Domain authority, measured by traditional backlink profiles, showed almost no correlation with AI citation frequency in our dataset. A six-month-old post on a DR-80 domain consistently lost to a two-week-old post on a DR-35 site. The models are clearly weighting recent context windows heavily.

Why models prefer fresh context

Large language models favour recent information because their training data and retrieval augmentations prioritise temporal relevance. When a user asks about a current topic, the retrieval step surfaces recent documents. The model then latches onto the temporal signal to avoid providing outdated answers.

We tested this by publishing identical technical explanations with different timestamps. The version dated this week was cited 3.1 times more often than the version dated six months ago. The content was word-for-word identical. The only variable was the perceived publication date embedded in the markup.

This creates a brutal trade-off for evergreen content. You can invest heavily in authoritative, comprehensive guides, but if the publication date ages past the 90-day horizon, your citation probability collapses. The AI search ecosystem effectively punishes stability and rewards churn.

The freshness injection pattern

We adapted by implementing what we call freshness injection. Rather than rewriting entire articles, we append dated supplementary sections to existing evergreen pages. A 200-word update noting recent developments, tagged with the current date, resets the citation horizon in roughly 70% of our test cases.

The format matters. A simple 'last updated' timestamp in the footer had zero measurable effect. The model needs to see a semantically meaningful addition tied to a specific date. We structure these as brief editorial notes at the top of the article, clearly dated and contextually relevant.

Our current rule is to inject a freshness update every 60 days for pages targeting high-value GEO queries. This costs roughly 20 minutes per page. It is not elegant, but the citation data justifies the effort. Without it, our evergreen inventory effectively becomes invisible to AI retrieval.

Measuring your own citation decay

Start by logging the publication dates of pages that AI engines actually cite for your target queries. Plot citation frequency against content age. You will likely see your own horizon. For us it was 90 days; in faster-moving verticals it may be as short as 30 days.

Track this weekly. AI retrieval behaviour shifts as underlying models update. Our 90-day horizon shifted to 75 days after a major model provider changed their retrieval weighting in November. If you are not measuring, you are relying on assumptions that are probably already stale.

GEO is not traditional SEO with a new coat of paint. Authority and backlinks matter less than temporal signals and structured recency markers. Update your evergreen pages with dated, relevant additions every 60 days, measure your own citation decay curve, and treat publication freshness as a first-class ranking factor.

The hard truth is that AI search rewards active maintenance over passive authority. Identify your highest-value queries, set a 60-day freshness calendar for those pages, and log your citation rates before and after each update. The data will tell you whether the injection pattern holds for your domain.