Two pages, same topic. Which one does the AI answer quote?
STALE
CITED
A content refresh used to be housekeeping: something you did to a tired blog post when the calendar was empty. That math changed in 2025, when Ahrefs analyzed 17 million citations across ChatGPT, Perplexity, Gemini, Copilot, and Google’s AI Overviews and found that AI assistants cite content that is, on average, 25.7% fresher than what ranks in Google’s organic results. Updating what you already published is no longer maintenance. It is one of the few levers that measurably improves your odds of being the page an AI answer quotes, and it costs a fraction of writing something new.
Below is what the data actually says about freshness and AI citations, how often different kinds of pages need updating, how to pick refresh targets without wasting a quarter on the wrong ones, and, because this part gets skipped constantly, what counts as a real update versus a cosmetic date change that Google has explicitly said it ignores. Every figure is attributed to a named, dated source; nothing here is modeled or estimated.
Does updating content help AI citations?
Yes: the largest citation study run to date found a clear freshness preference across every major AI assistant. Ahrefs extracted 16.975 million cited URLs from ChatGPT, Perplexity, Gemini, Copilot, AI Overviews, and the organic Google SERP in July 2025, then compared how old the cited pages were. URLs cited by AI assistants averaged 1,064 days since publication. URLs in the organic SERP averaged 1,432 days.
The pattern holds when you look at update dates instead of publish dates. Pages cited by AI assistants had been updated an average of 909 days ago, against 1,047 days for organic results: a smaller but still meaningful 13.1% difference. And two platforms go further: Ahrefs found that both Perplexity and ChatGPT appear to order their in-text references from newest to oldest, meaning a recently updated page doesn’t just get cited more often, it gets cited earlier in the answer, where people actually read.
Worth pausing on what this does not say. The average AI-cited page is still 2.9 years old. Assistants are not chasing yesterday’s hot take; they favor established pages that have been kept current. That is exactly the profile a disciplined refresh program produces, and it is why we treat refresh cadence as a core part of content marketing strategy rather than an afterthought: the compounding asset is the old page, maintained, not the new page, abandoned.
Which AI engines prefer fresh content the most?
ChatGPT, by a wide margin, and Google’s AI Overviews barely care at all. Using the organic SERP as a baseline, Ahrefs measured how much newer each platform’s cited pages were. ChatGPT’s citations run 458 days newer than organic results, and its in-text references 393 days newer. Copilot came in at 360 days newer, Gemini 298, Perplexity 250. AI Overviews were the outlier: the pages they cite are actually 16 days older than the organic SERP average.
The practical read: if ChatGPT and Perplexity matter to your category (and for anyone selling to buyers who research in an assistant, they do) recency is a real ranking input you can influence on demand. If your visibility problem is specifically AI Overviews, freshness will not save you; AIO behaves like classic Google, where authority and relevance dominate. Ahrefs’ March 2026 follow-up work found only 38% of AI Overview citations come from pages ranking in the top 10 for the query, down from 76% a year earlier, with citations increasingly pulled from fan-out sub-queries instead.
Different engine, different lever.
One more platform note from the same research thread: YouTube has become the single most-cited domain in AI Overviews, which is why video deserves a lane in this plan too. We broke that down in our guide to earning AI citations with YouTube video.
How often should you refresh your content?
There is no single correct interval: cadence should follow how fast each page’s subject actually changes. That answer is unsatisfying, so here is the working schedule we use, consistent with Ahrefs’ December 2025 guidance on content freshness:
- Statistics pages, tool comparisons, “best X” lists: every quarter, and always before your category’s buying season. These decay fastest and get cited most.
- Seasonal and recurring topics: roughly three months before peak. Ahrefs’ analysis of pages ranking for seasonal queries found top results are updated ahead of the demand spike, then again mid-season — not after.
- Evolving topics (AI search being a perfect example): whenever the underlying facts change, which in practice means a review every one to two months and an edit when something material moved.
- True evergreen: annually, or when intent drifts. Some pages rank for years untouched; forcing updates onto them is busywork.
If that sounds like a lot, consider the benchmark Ahrefs published in December 2025 after auditing its own blog: roughly 52% of its pages published before 2025 had not been updated at all during 2025, and that is a content team whose entire business is SEO. Most companies sit far below that bar, which is precisely why a systematic refresh program is such an available advantage. Demand for the tactic itself is modest but real (“content refresh” draws about 390 US searches a month and “content decay” another 320 (Semrush, US database, July 2026)) yet the searched questions behind it (“how often should you refresh your content,” “how to decide if content should be refreshed or deleted”) show people are stuck on process, not concept.
How do you decide which pages to refresh, or delete?
Start from traffic and citations, not from age. The pages worth refreshing first are the ones already carrying weight: Ahrefs’ freshness guide recommends focusing on your top 20% of pages by traffic before touching anything else, because updates to pages with existing authority show results fastest.
- Pull every page older than 12 months with its organic traffic, rankings, and — if you track them — AI citation counts.
- Flag decay: pages whose traffic or positions have slid for two-plus consecutive quarters. Those are your refresh candidates.
- Flag the dead weight: pages with no traffic, no rankings, no links, and no strategic purpose. Consolidate them into a stronger page or retire them with a redirect. Refreshing a page nobody ever found is polishing a stone.
- Sort what remains by commercial value — the money pages and the pages AI assistants already cite get the first slots on the calendar.
The refresh-or-delete call trips teams up more than anything else in this process. The tiebreaker is search intent: if real demand still exists for the topic and your page once served it, refresh. If the topic itself expired (an event, a discontinued product, a question nobody asks anymore) deletion with a redirect to the closest living page is the honest move. What you update matters as much as what you pick, and the bar for AI-quotable structure is its own discipline: one we covered in detail in how to write content AI engines actually quote.
What counts as a real content update?
A real update changes the substance of the page: the facts, the examples, the answers. Google has been unusually blunt about the alternative. In an April 2021 Search Central session, Google’s John Mueller said the company “definitely wouldn’t rank those pages differently in search just because you’re changing the date and time on a page,” and recommended updating dates only when making significant changes: advice consistent with his earlier written guidance not to “artificially freshen” a story without adding significant information (Search Engine Journal, April 7, 2021).
So what clears the bar? The changes that both readers and machines can detect:
- Swap aging statistics for current, attributed ones — and date them in the text (“as of July 2026”), which doubles as a freshness signal an LLM can lift verbatim.
- Answer the questions that emerged since you published. Check what now appears in People Also Ask and what top-ranking pages cover that you don’t; close those gaps.
- Update screenshots, pricing, product names, and dead examples. Stale interface shots quietly tell every reader the rest of the page is old too.
- Rewrite the title and opening if the year or framing dates them — the snippet is the first thing decaying click-through rates trace back to.
- Update
dateModifiedin your schema markup and show a visible “last updated” date, so the change is machine-readable everywhere it counts. - Add internal links from and to the refreshed page while you’re in there — refreshes are the cheapest moment to strengthen your linking structure.
The date-stamp shortcut fails on both fronts: Google says date-only changes carry no ranking weight, and the AI freshness data measures actual content age from crawler observation: Ahrefs’ study used the date pages were first seen and last substantively changed, not the date printed on the page. A refreshed timestamp over unchanged text is invisible to the systems it is meant to impress, and mildly corrosive to the readers who notice. If a page is not worth genuinely updating, it is not worth re-dating.
How do you measure whether a content refresh worked?
Set the baseline before you touch the page, then check four numbers on a schedule. Position movement shows up first: typically within one to three weeks. Ahrefs’ December 2025 guide includes a clean example from its own blog: after a substantive refresh, its article targeting “republishing content” jumped from position 21 to position 4. The other three numbers take longer:
- Organic traffic to the refreshed page over the following 60–90 days, against the pre-refresh trend line rather than a single point.
- AI citations: whether ChatGPT, Perplexity, and AI Overviews cite the page for your target prompts, checked before and after. This is the number the freshness data says should move, and it is the one almost nobody tracks.
- Click-through rate in Google Search Console, since a rewritten title on the same rankings often lifts clicks by itself.
Two housekeeping notes that speed up the payoff. Submitting updated URLs through IndexNow gets the change into Bing’s index quickly: relevant because ChatGPT’s browsing draws on Bing, a point Ahrefs flags in its freshness guidance. And keep a simple refresh log, page, date, what changed, because six months from now, attribution will be impossible without it.
The freshness data turned content refreshes from hygiene into strategy. AI assistants cite content 25.7% fresher than organic Google results, ChatGPT’s citations run more than a year newer than what ranks, and Perplexity and ChatGPT push the newest sources to the front of their answers, while Google’s AI Overviews stay indifferent to age. The winning program is unglamorous: prioritize your top pages by traffic and commercial value, update substance on a cadence matched to how fast each topic moves, never fake it with a date change Google has said it ignores, and measure rankings, traffic, and AI citations against a logged baseline. Old pages, genuinely maintained, are the cheapest AI visibility you can buy.
How much of your site is quietly going stale?
We’ll audit your existing content for decay, show you which pages AI engines currently cite (and which they’ve dropped), and hand you a prioritized refresh plan with the expected payoff for each page. Free, no obligation.
Sources
- Ahrefs, “New Study: AI Assistants Prefer to Cite ‘Fresher’ Content,” Ryan Law & Xibeijia Guan, July 28, 2025 (16.975 million cited URLs from ChatGPT, Perplexity, Gemini, Copilot, AI Overviews, and organic Google via Brand Radar): AI-cited URLs average 1,064 days since publication vs 1,432 for organic results (25.7% fresher); last-updated gap 909 vs 1,047 days (13.1%); ChatGPT cites URLs 458 days newer (citations) and 393 days newer (references) than organic results; Copilot 360, Gemini 298, Perplexity 250 days newer; Google AI Overviews cite pages 16 days older than organic; Perplexity and ChatGPT order in-text references newest to oldest; average AI-cited page age 2.9 years
- Ahrefs, “Fresh Content: Why Publish Dates Make or Break Rankings and AI Visibility,” Mateusz Makosiewicz, December 22, 2025: refresh prioritization (top 20% of pages by traffic first), seasonal update timing roughly three months before peak, “republishing content” keyword moving from position 21 to 4 after a refresh, ~52% of Ahrefs’ own pre-2025 blog pages not updated during 2025, IndexNow/Bing relevance to ChatGPT visibility
- Ahrefs, “Update: 38% of AI Overview Citations Pull From The Top 10,” Louise Linehan & Xibeijia Guan, March 2, 2026 (863K keyword SERPs, 4 million AI Overview URLs): 38% of AI Overview citations rank in the top 10 for the same query, down from ~76% in July 2025; citations increasingly drawn from query fan-out sub-queries; YouTube the most-cited domain in AI Overviews, up 34% in six months
- Google Search Central documentation, “A Guide to Google Search Ranking Systems”: freshness systems that show more recent content for queries where it would be expected
- Search Engine Journal, “Google’s John Mueller: Changing Dates on Pages Won’t Improve Rankings,” Matt G. Southern, April 7, 2021: Mueller: Google “definitely wouldn’t rank those pages differently in search just because you’re changing the date and time on a page”; update dates only for significant content changes; earlier guidance against artificially freshening articles
- Semrush Keyword Analytics (US database), retrieved July 2026: “content refresh” 390 searches/mo (keyword difficulty 20); “content decay” 320/mo (KD 21); “content freshness” 110/mo (KD 31); question queries including “how to decide if content should be refreshed or deleted” (50/mo) and “how often should you refresh your content” (10/mo)