Last updated September 2026. Two studies anchor almost every number below. Seer Interactive tracked the publish year behind citations on three engines, using a citation feed built on Peec AI. Ahrefs measured average content age across 16.975 million citations, engine by engine.
Where a figure comes from one of those two, it carries a verified study tag. Where a figure comes from a single vendor reporting on its own dataset, it carries a vendor-reported tag, and you should read it as directional. Full definitions for the terms used throughout, including citation share, live in the glossary.
Generative engines rarely retrieve every candidate page from scratch at answer time. Most blend a static training corpus with a live retrieval layer, and that retrieval layer ranks partly on recency signals: publish date, last-modified date, and how often a crawler has seen new content on a domain. A page that never changes eventually reads as a weaker retrieval candidate than a newer page saying the same thing, even when the older page is still accurate. That is the mechanism behind almost every number on this page.
Freshness by publish year, across ChatGPT, Perplexity, and Google AI Overviews
Not every engine leans on new content the same amount. Seer Interactive tracked the publish year behind citations on three platforms. Perplexity draws the strongest lean toward the current year. Google AI Overviews sits close behind it. ChatGPT leans hardest on older, established pages.
| Engine | 2025 | 2024 | 2023 | 2023-2025 combined |
|---|---|---|---|---|
| ChatGPT | 31% | 29% | 11% | 71% |
| Perplexity | 50% | 20% | 10% | 80% |
| Google AI Overviews | 44% | 30% | 11% | 85% |
Source: Seer Interactive, Study: AI Brand Visibility and Content Recency, June 2025 (5,000+ URLs with extractable publish dates, citation tracking via Peec AI). Verified study.
The pattern carries a lesson for content strategy. If Perplexity is your priority engine, a stale page has less time to earn a citation before the model favors something newer. If ChatGPT is your priority, well-established pages keep earning citations for years, so a refresh matters less than earning that first citation.
Average age of cited content, engine by engine
Ahrefs measured the same question a different way: how many days old is the average page an AI engine cites? Across 16.975 million citations, the answer varies more by engine than the year-bucket data suggests.
| Engine | Average age of cited content |
|---|---|
| ChatGPT | 958 days (2.6 years) |
| Microsoft Copilot | 1,056 days (2.9 years) |
| Gemini | 1,118 days (3.1 years) |
| Perplexity | 1,166 days (3.2 years) |
| Google AI Overviews | 1,432 days (3.9 years) |
| Organic Google search (baseline) | 1,416 days (3.9 years) |
Averaged across every AI platform, cited pages run 1,064 days old, about 2.9 years, which is 25.7% fresher than the 1,432-day average for organic search results. ChatGPT pulls that average down: its citations are the freshest of any engine in the study. Google AI Overviews pulls the other way. Its citations run almost as old as plain organic search, the least fresh of any AI engine Ahrefs measured.
Source: Ahrefs, study by Ryan Law and Xibeijia Guan, July 2025 (16.975 million cited URLs). Verified study. Ahrefs itself flags the finding as correlational, not causal, and notes that the study’s own average age for AI-cited pages is still 2.9 years: old, established content still wins most citations.
In practice, a 2.6-year average for ChatGPT still means most of its citations go to pages published one to three years ago, not pages published last week. Freshness helps at the margin. It does not replace a page that already ranks and already earns links.
Ahrefs also found that organic Google rankings show a much stronger age gradient than AI citations do. Older pages dominate the top organic positions, while AI-cited pages cluster closer in age to each other, regardless of which position in the answer they are quoted from.
Why fresher pages tend to win the citation
Three mechanisms show up across the research above, even though none of the studies proves causation directly.
Retrieval favors recency signals. Search-grounded engines such as Perplexity and Google AI Overviews run a live retrieval step before generating an answer, and that step weighs publish and last-modified dates alongside relevance and authority.
Training-data engines default to older, established sources. ChatGPT leans more on its training corpus for stable topics, which favors pages that have already accumulated links, mentions, and time in the index, exactly what the year-bucket data above shows.
A stale page loses the freshness signal without losing the authority signal. That is the gap a refresh closes. The page keeps its links and its history, and it picks back up the recency boost it had lost.
The freshness decay curve
Seer Interactive also logged raw AI bot traffic, not just citations, hitting the same 5,000-plus URLs. The pattern looks like a decay curve: most activity sits in the first year or two, then falls off fast.
| Content age | Share of AI bot activity |
|---|---|
| Published within the last year | 65% |
| Published within the last 2 years | 79% |
| Published within the last 3 years | 89% |
| Published within the last 5 years | 94% |
| Published more than 6 years ago | 6% |
These figures track raw crawler hits across all bots combined, not citations broken out by engine. Read them as a shape, not a per-engine benchmark: activity concentrates early, then thins out for years before it drops close to nothing. That shape mirrors what search marketers already know about content decay in organic search, just compressed into a shorter window.
Source: Seer Interactive, same June 2025 study. Verified study.
More proof: the 30-day update pattern
A few more figures point the same direction as the two verified studies above, though they come from a single vendor and carry a lighter evidence grade. Treat them as supporting evidence, not the backbone of your case.
- Content updated within the last 30 days earns roughly 3.2 times more AI citations than older content, per ConvertMate’s 2026 AI Visibility Study. The dataset spans more than 80 million citations across upward of 10,000 domains, but the methodology is not independently published. Grade: vendor-reported.
- 76.4% of ChatGPT’s most-cited pages had been updated within the prior 30 days, per the same ConvertMate study. Grade: vendor-reported.
- One client saw AI-referred traffic rise 300% after a page refresh, per a single-client case study from Seer Interactive. The industry was not disclosed, and one client is not a benchmark. Grade: vendor case study.
Track the recency of your cited pages
Knowing that freshness matters is different from knowing which of your own cited pages are aging out. A handful of GEO platforms track citation activity closely enough to spot that gap.
Ahrefs Brand Radar draws on a real-query corpus of 370M+ monthly prompts, which makes it useful for spotting when a competitor’s fresher page starts pulling citations away from yours. Profound adds citation source attribution built for reporting up to leadership, useful when a freshness gap needs budget to fix. Temso is an all-in-one option, from $89/mo, that pairs citation tracking with the content brief needed to act on what it finds, including an AI agent that can draft the refresh itself. Peec AI runs a large index of commercial-intent prompts and is a common pick for teams that want deep coverage on buying-stage queries specifically.
None of these tools publish a page-level freshness score today. Each one does, however, log the date it last saw your page cited, which is a reasonable proxy: a page that has not been cited in months is either losing the freshness race or losing on some other signal entirely, and it is worth checking the publish date either way. The workaround: pull your list of most-cited pages from whichever platform you use, then check each one’s last substantive update against the 30-day and one-year windows in the tables above. See how platforms like these are scored in the methodology.