AI Citation Freshness: The 30-Day and 13-Week Windows That Decide Who Gets Cited
Every AI search guide says freshness matters. This Lureon report shows how much, with numbers: the age thresholds that decide citation eligibility, how fast citations decay once earned, how each engine treats recency differently, and the exact refresh system that keeps a page in the answer.
Last updated: July 2026 · Lureon citation tracking + 10 published primary studies · Reviewed quarterly (we practice what this report preaches)
Free 30-minute call · No retainer required to talk
76%of AI citations in Lureon's client tracking come from content updated in the last 30 days
50%of AI search citations industry-wide come from content less than 13 weeks old (Amsive)
25.7%fresher: the average AI-cited page vs Google's organic top 10, across 17M citations (Ahrefs)
4.5 wksmedian half-life of an AI citation before it drops out of answers (Scrunch/Stacker, 3.5M events)
Quick answer: how fresh does content need to be to get cited by AI? Inside 30 days for the best odds, inside 13 weeks to stay competitive, and never past 12 months untouched. Roughly half of all AI citations go to content under 13 weeks old, AI-cited pages run 25.7% fresher than Google's organic results, and in Lureon's own client tracking 76% of citations come from content updated in the last 30 days. Freshness is manufactured, not born: most "fresh" cited pages are old URLs that were substantively updated, which is why a refresh cadence, not constant new publishing, is the winning system.
The Thresholds
How fresh is "fresh"? The age windows that decide citations
Freshness in AI search is not a vague preference. It is a set of measurable windows, and every major dataset lands on the same shape: a steep cliff after the first months, and near-exclusion after two years.
Age of the most-cited pages in AI answers
Share of top citations by content age, across 200M+ citations (ChatGPT, Perplexity, Gemini, Claude)
Source: Lantern AI Citation Content Visibility Report, February 2026, analysis of 200M+ citations.
The corroborating numbers stack the same way from every methodology. Ahrefs measured nearly 17 million cited URLs and found AI-cited content averages 1,064 days old versus 1,432 for Google's organic top 10, a 25.7% freshness gap, with ChatGPT the extreme case at 33% fresher. Amsive's citation analysis puts half of all AI citations inside 13 weeks. AirOps found 83% of commercial-query citations go to pages updated within a year, and that pages left untouched for three months are over 3x as likely to lose their citations entirely. And the most important nuance, from Seer Interactive's July 2026 study of 47,097 citations: the freshness engines reward is mostly manufactured by updates, with more than a quarter of "fresh" cited pages first published two or more years ago. Old URLs win. Stale URLs lose.
Citation Decay
How long a citation lasts once you win it
Earning the citation is the easy half. Survival-curve analysis of 3.5 million citation events across 120,000+ domains shows won citations decay fast, and at different speeds per platform.
Median AI citation half-life, by platform
Weeks until 50% of a cohort's citations have dropped out of answers
Source: Scrunch and Stacker survival-curve analysis of 3.5M citation events, 120,000+ domains, Sep 2025 to Mar 2026.
Two implications follow. First, cadence beats campaigns: a page refreshed once and abandoned re-enters the same decay curve it just escaped, which is why the winning teams run refreshes as a standing system rather than a quarterly project. Second, the volatility is an opportunity: AirOps found only 30% of brands stay visible in back-to-back AI responses, so a competitor's citation is rarely locked in. The next refresh cycle can be yours. This is also why Lureon reports citation share monthly rather than as a one-time audit; you can read how that connects to pipeline in our AI Recommendation Gap report.
Engine by Engine
Every engine reads freshness differently
The freshness rule is universal, but its strength is not. Match your refresh effort to where your buyers actually ask.
ChatGPT: the recency extreme
Cites content roughly 33% fresher than Google's organic results (958-day average vs 1,432) and churns sources fastest, with a 3.4-week citation half-life. It is also where owned content is most likely to earn citations, so this is where a tight refresh cadence pays most. Practical rhythm: refresh priority pages roughly biweekly to monthly.
Perplexity: kindest to evergreen
Leans on blogs, guides, and comparisons but reaches furthest into older reference material, rewarding depth and authority rather than pure recency, and holds citations longest at 5.8 weeks. Strong, substantive older content still wins here. Practical rhythm: a six-week refresh cycle is usually enough.
Gemini and Google AI surfaces
Gemini's citation mix skews heavily recent, with about 90% of citations from content updated within two years, driven by the marketplaces, aggregators, and comparison pages it favors. AI Overviews behave most like classic Google, blending freshness with accumulated authority. Practical rhythm: monthly updates on commercial pages.
Claude: freshness via Brave
Claude retrieves through Brave Search and favors verifiable, well-maintained deep pages, so its freshness signal arrives indirectly: keep the page current, keep Brave crawling it, and keep dated claims verifiable. Our companion report on winning Claude recommendations covers the full mechanics. Practical rhythm: track alongside your ChatGPT cadence.
Real vs Cosmetic
What counts as an update (and the shortcut that fails)
The most tempting tactic in this entire field is changing the date and touching nothing else. It does not work, and every source that has tested it agrees.
Date-washing fails. AI crawlers compare the fetched page against their cached version: if the body is unchanged, the new timestamp is discounted or ignored. Google's John Mueller has explicitly warned against re-dating content without substantial changes, and modern retrieval systems treat contradictory date signals as a trust problem. The freshness signal is earned by the edit, not the timestamp.
What registers as substantive
Replacing outdated statistics with current, sourced figures. Adding a section that answers a question the page missed (fan-out queries are the best source of these). Correcting or removing stale claims. Refreshing examples, screenshots, and product names. Adding in-text temporal anchors like "as of July 2026". A visible changelog noting what changed and when.
The technical signal stack
Once the edit is real, sync every machine-readable signal to it: dateModified in JSON-LD Article schema, an honest sitemap lastmod (per-page, never the sitemap generation date), the Last-Modified HTTP header, OpenGraph modified_time tags, a visible "Last updated" line near the top, and an IndexNow ping so crawlers return sooner. When these signals contradict each other, crawlers trust the most pessimistic one.
The System
The refresh cadence that keeps pages cited
Refreshing beats republishing: established URLs integrate into AI answers roughly 40% faster than new pages at about half the editorial cost, so most teams should run roughly 70% refresh, 30% new. This is the tiered system Lureon operates for clients.
1
Tier the library by volatility and value
Tier 1: pages earning citations now, plus commercial, comparison, and statistics-heavy pages. Tier 2: cornerstone guides and explainers. Tier 3: the stable long tail. Volatility, not content type, sets the tier: pricing and tool comparisons decay fastest, foundational concepts slowest.
2
Set the clocks: monthly, quarterly, annually
Tier 1 gets a substantive touch every 30 days, keeping it inside the strongest citation window. Tier 2 gets a quarterly review. Tier 3 gets an annual pass, because pages that cross 12 months untouched largely fall out of citation pools regardless of authority.
3
Refresh substance first, then signals
Update the stats, add the missing section, fix the stale claims, and only then sync the date signals (schema, sitemap, visible date, changelog). A signal update without a substance update is date-washing; a substance update without signal sync is invisible.
4
Watch decay signals, not just calendars
A lost citation on a tracked prompt, a sharp impressions drop, a competitor's fresher page entering the answer, or a year reference going stale all trigger an off-cycle refresh. Signal-driven beats schedule-driven once tracking is in place.
5
Measure the before and after
Log citation presence on your revenue prompts before each refresh and re-check within 30 to 60 days. Over a few cycles this shows which update types produce the biggest citation lifts for your site, turning the cadence from industry defaults into your own tuned system.
One honest caveat on the evergreen question: freshness is not a death sentence for older content. Engines split by query type, sending recency-implying prompts to fresh sources while stable informational queries still reward established authority, and Perplexity in particular keeps citing strong older pages. The "publish once, rank forever" playbook is what died. Evergreen content survives as a maintained living asset: even Wikipedia's compound interest page, covering a 4,000-year-old concept, was updated within the last three months. The platform-by-platform playbook is in our guide to getting recommended by ChatGPT and other LLMs.
Proof
What a freshness system produces
Refresh cadence is one pillar of the AEO program Lureon runs. Here is what the full system delivers.
Lureon's work is led by a senior operator with 12+ years in search, everyone else on the team has 6+ years, no juniors, no outsourcing. For SaaS companies, scope lives on the B2B SaaS AEO service page and pricing is public.
FAQ
Frequently asked questions
How fresh does content need to be to get cited by AI?
Inside 30 days of a substantive update gives the best odds, inside 13 weeks keeps you competitive, and past 12 months untouched a page largely falls out of citation pools. Half of all AI citations go to content under 13 weeks old, 62% of top-cited pages were published within six months, and in Lureon's client tracking 76% of citations come from content updated in the last 30 days. The window is a refresh window, not a publish window: updated old URLs qualify as fresh.
Do AI assistants really prefer fresher content?
Yes, measurably. Ahrefs analyzed nearly 17 million citations across seven AI platforms and found AI-cited URLs average 1,064 days old versus 1,432 days for Google's organic top 10, a 25.7% freshness gap. ChatGPT is the extreme at 33% fresher. Ahrefs frames this as a citation pattern rather than a causal multiplier, and the widely shared "4.3x more citations for fresh content" figure traces to no primary study and should not be repeated.
How long does an AI citation last once earned?
The median half-life is about 4.5 weeks: half of a cohort's citations drop out of AI answers within that window, per Scrunch and Stacker's survival-curve analysis of 3.5 million citation events across 120,000+ domains. ChatGPT churns fastest at 3.4 weeks and Perplexity holds longest at 5.8. Earned third-party coverage that corroborates your facts on other domains roughly doubles citation persistence.
Do ChatGPT, Perplexity, and Gemini treat freshness differently?
Yes. ChatGPT shows the strongest recency bias and the fastest churn, making it the platform where a tight refresh cadence pays most. Perplexity reaches furthest into older evergreen material and rewards depth and authority, holding citations longest. Gemini skews fresh through its citation mix, with roughly 90% of citations from content updated within two years, while Google AI Overviews blend freshness with classic authority signals. Cadence should follow the platform your buyers use.
How often should I update content for AI search?
Tier it. Commercial, comparison, and statistics-heavy pages: every 30 days. Cornerstone guides and explainers: quarterly. The stable long tail: at least annually, since pages untouched past 12 months largely exit citation pools. Layer decay signals on top of the calendar: a lost citation, a stale year reference, or a fresher competitor entering the answer triggers an immediate off-cycle refresh regardless of schedule.
Is it better to update old content or publish new content?
Update, in most cases. Established URLs carry accumulated authority, backlinks, and retrieval history, and integrate into AI answers roughly 40% faster than new pages at about half the editorial cost. A practical split is 70% of resources toward refreshing and consolidating, 30% toward genuinely new coverage. Publish new only when the topic is uncovered or the existing page is beyond repair, and redirect the old URL if you rebuild.
Does changing the publish date without updating content work?
No. AI crawlers compare the fetched page against their cached version, and an unchanged body with a bumped date is treated as unchanged; some systems discount the signal entirely, and Google's John Mueller has explicitly warned against artificial re-dating. The honest pattern is the one that works: make a substantive edit, keep the original publish date, update the visible "last updated" line and dateModified schema, and log what changed.
Which technical signals tell AI crawlers a page is fresh?
Six, kept in sync: dateModified in JSON-LD Article schema (the most explicit machine-readable signal), an accurate per-page sitemap lastmod, the Last-Modified HTTP header, OpenGraph modified_time tags (used by Perplexity and ChatGPT for metadata), a visible "Last updated" date near the top of the page, and an IndexNow ping after publishing the change. Contradictory dates make crawlers trust the most pessimistic one, so consistency matters as much as recency.
Does evergreen content still get cited by AI?
Yes, with maintenance. Engines split by query type: recency-implying prompts get fresh sources, while stable informational queries still reward established authority, and research across 485,000+ ChatGPT citations found authoritative sources winning those queries. Perplexity is the friendliest engine for strong older content. What died is publish-and-forget: a well-maintained pillar can stay cited for 18+ months, while an identical page left static drops out within a couple of refresh cycles.
What is the fastest freshness win?
Pick your five highest-value pages, replace every outdated statistic with a current sourced figure, add one section answering a question the page missed, then sync the visible date, dateModified schema, and sitemap lastmod, and resubmit the URLs. This is a days-not-months fix that moves pages back inside the 30-day window where citation odds are strongest, and it is typically the first sprint in any Lureon engagement.
How do I audit my content freshness for AI visibility?
Inventory your library with last-substantive-update dates, flag every Tier 1 page older than 90 days and anything past 12 months, verify the technical signals (dateModified, lastmod, visible dates) agree with reality, and baseline citation presence on your buyers' real prompts across ChatGPT, Perplexity, Claude, and Gemini. Lureon runs this as a free initial audit: you get the stale-page list, the signal fixes, and the citation baseline before any engagement decision.
Your best pages are quietly going stale
We'll audit your library against the 30-day and 13-week windows, show you which citations you're losing to fresher competitors, and hand you the refresh plan that wins them back.
Limited client roster: a senior-only team means capped capacity.
Methodology: the 76% figure is from Lureon's internal citation tracking across client accounts, measuring the share of observed AI citations pointing to pages substantively updated within the prior 30 days; it describes our tracked sample, not a universal rate. Industry figures are from published primary research: content-age distribution from Lantern's AI Citation Content Visibility Report (Feb 2026, 200M+ citations); freshness gap from Ahrefs' analysis of 16.975M cited URLs across 7 platforms (Jul 2025); the 13-week finding from Amsive (2026); citation half-life from Scrunch and Stacker's survival-curve analysis of 3.5M citation events (Sep 2025 to Mar 2026, 120,000+ domains); commercial-query update windows and the 3x stale penalty from AirOps' State of AI Search (2026); manufactured-freshness and per-engine patterns from Seer Interactive's study of 47,097 citations (Mar to Jun 2026, published Jul 2026); update-vs-create economics from published RAG integration analyses. Lumanu results are from Lureon client reporting. Sample definitions and conversion events vary across the cited studies.
Lureon
Data Report · Citation Freshness
Covers ChatGPT, Perplexity, Claude, Gemini, Grok & AI Overviews