When LLMs pick a source, they weigh recency in citation selection alongside authority and relevance—but "fresh" doesn't mean "published yesterday." A 2020 guide on CRM software might stay cited after one update; a Python how-to from 18 months ago may never recover. The threshold depends on query type, engine training cutoff, and whether the content has actually changed.
Why Recency Matters to ChatGPT (But Not Always to You)
Here's the honest version: language models are trained on snapshots of the web frozen at a cutoff date, and answer engines add live retrieval on top. ChatGPT with search, Perplexity, and Claude with web search can all pull current pages when they answer. So a preference for content freshness citations isn't a conspiracy—it follows from how they work.
But recency alone doesn't win citations. A brand-new blog post from an unknown site often won't beat a five-year-old post from HubSpot on the same topic, even though it's fresher. The engines are balancing three signals: relevance (does the page actually answer the query?), authority (who wrote it?), and LLM recency signals (when was it last updated?).
For SEO folks, this is the shift. Google weighs freshness too, mostly for trending topics and news. An answer engine has reason to care more broadly, because it's composing advice, and stale advice makes a bad answer. A "best CRM for 2024" guide needs updating every year. A conceptual piece on "how to write an effective email" doesn't age the same way. The engines are learning to tell the difference—or at least, they should be.
The practical implication: if you haven't touched your cornerstone content in 18 months, check it. On topics that imply currency, an untouched page can lose ground in AI answers even when its information is still right, because a fresher source looks like the safer citation.
The Citation Staleness Cliff: When Does Old Content Get Dropped?
There's no published threshold, and b/cited hasn't measured one. What is structurally true is that freshness matters far more on some topics than others: a software roundup, a regulatory guide or a market trend piece carries an implicit claim to be current, and an engine has good reason to prefer a newer source for those. A conceptual explainer carries no such claim and ages far better. Before you rewrite anything on a schedule, ask which kind of page it is — the answer decides whether a date is a signal or noise.
But here's the qualifier: the authority vs. freshness tradeoff is real. A widely-linked, high-authority page from 2022 on "JavaScript frameworks" can still beat a fresher page from a no-name blog, because the fundamentals haven't changed. Recency becomes a tiebreaker, not a disqualifier.
The tricky part is detecting when you've hit the cliff for your content. Some pages have a natural half-life. A post on "Top SaaS Trends 2024" is worthless on January 1, 2025. A post on "How to configure Stripe webhooks" will stay relevant for years unless Stripe changes the API. Most content lives in the fuzzy middle.
One signal worth watching: a page whose citations fall over a few weeks while its content hasn't changed. Recency is one possible cause; a competitor publishing something better is another. Use tracking citation rank over time to watch this. A drop across several engines points at the page itself; a drop on one engine may be a change in that engine's sources.
Do ChatGPT, Claude and Perplexity Treat Freshness Differently?
They may not weigh it the same way. A common impression is that ChatGPT leans on established sources while Perplexity surfaces newer pages and leans harder on topical freshness signals. b/cited hasn't measured that, so treat it as a hypothesis to check: run the same query across ChatGPT, Claude, and Perplexity and compare the dates of what each one cites.
For example, on a query like "best AI writing tools," one engine might cite established review roundups while another cites recent posts from the tools themselves or newer review sites. Neither is wrong: one is favoring authority and longevity, the other currency.
For your content strategy, this means tracking which engines matter for your business and how each one treats your topic. If the engines your audience uses cite older sources on your queries, a quarterly update may be enough. If they cite recent pages, treat your content more like a news site and keep core pages current monthly or bi-weekly when the market moves that fast.
Timing differs too. An engine that searches the web when it answers can pick up an update once it re-reads the page; what a model learned in training changes only when the model is retrained. How quickly each engine reflects an update isn't published, so measure it on your own pages rather than assume a number.
Query Type Matters: News vs. Evergreen vs. How-To Recency Rules
News-driven queries ("Apple earnings 2026," "latest EU AI regulations") require currency. Older coverage rarely makes the cut. If you're chasing citations on breaking news, you're racing against journalists and the companies themselves.
Evergreen how-tos ("how to write a LinkedIn headline," "how to structure a SQL JOIN") age slowly. A post from 2019 can still dominate citations because the fundamentals are stable. But here's the catch: if the tool or platform changes, you need to signal that you know about it. A how-to on Instagram that doesn't mention Reels in 2024 reads as out of touch, even if the core advice is sound.
Trend and roundup content ("best CRM software 2024," "emerging Python frameworks") needs yearly updates, at minimum. Software rankings change; new competitors emerge; pricing shifts. A roundup is expected to be re-evaluated. A post that hasn't been touched in 18 months screams "I forgot about this."
How-to for stable tools (Git, CSS Grid, PostgreSQL fundamentals) can stay fresh longer because the concepts don't drift. But if your how-to cites outdated best practices or deprecated syntax, recency signals won't help.
The rule of thumb: if your content includes comparisons, rankings, product recommendations, or version-specific syntax, treat it as high-freshness. If it's conceptual or architectural, you have more runway. But even evergreen content benefits from a "last updated" marker and occasional signal refreshes.
The Authority-Freshness Tradeoff: When Newer Loses to Established
This is where the rubber meets the road. You're competing with high-authority sites that update slower, or low-authority sites that update faster. Who wins?
Often, authority wins. b/cited hasn't measured how often, but a New York Times piece on remote work from 2022 can still get cited over a 2024 blog post from a random startup, because The New York Times is trusted. Recency tends to decide it only when authority levels are close.
This is actually good news for smaller sites with strong expertise. If your domain authority is in the same ballpark as your competitor's—say, both are industry publications or well-known blogs—then a more recent, substantive update can tip the citation your way. You become the fresher expert.
The trap: dumping old content to chase freshness signals. A 2020 guide on "remote work best practices" that you refresh with a 2026 dateline isn't better if the core advice hasn't changed. A new date on old advice gives an engine nothing new to cite; see how LLMs decide what to cite for what does make a page citable.
Crawlers can read dates from HTML metadata (Last-Modified headers, schema.org dateModified) as well as from the page itself. A new date on unchanged content gives a reader, human or machine, nothing new; refreshed data, re-reported stats, and retested examples do.
Testing Recency Impact: A/B Update Timing and Citation Win Rate
You can test this. Pick a cornerstone piece of content you know is getting cited (use tracking citation rank over time to establish baseline). Note your citation count in ChatGPT, Perplexity, and Claude for a specific query.
Update the content substantively—refresh stats, retest examples, add new research. Don't just change the dateline. Give the engines a few weeks to re-read the page, then re-run the same queries and check citation count.
You're looking for a delta. Did you jump from 3 citations to 5 in Perplexity? Did ChatGPT still show you but not in the top 3? That tells you something about how each engine weights freshness vs. your authority.
Repeat this quarterly on different pieces. Over time, you'll build your own model of how long a page holds its citations in your industry, at your authority level, on each engine. That's your update frequency AEO baseline, measured on your own pages.
One caveat: if you're in a fast-moving niche (AI tools, crypto, SaaS pricing), your baseline will be shorter. If you're in a stable niche (finance fundamentals, design principles), it's longer. Don't force a monthly update cycle if your content doesn't need it—that's busywork that signals the opposite of authority.
Update Signals Engines Can See
Several freshness signals are visible to an engine:
Publication date and last-modified headers. These are the obvious ones: the Last-Modified header and the schema.org dateModified field both say when a page last changed. But also check: if your CMS auto-timestamps when you edit anything, you might be gaming this accidentally. Don't count on a trivial edit reading as fresh.
Content changes. An engine that retrieves pages live reads what the page says today, not just its date. How it treats the difference between versions isn't published, but a substantive change gives it something new to cite; a changed date alone doesn't.
Inbound links. New links from other sites show a page is still being referenced, which search engines have long treated as a sign of relevance. How much answer engines weigh it isn't published.
The implication: a small, real refresh (new research, one fresh study cited, an updated example) plus an honest schema.org dateModified is the cheap way to stay current. Flipping the dateline with no content change adds nothing a reader or an engine can use.
Refresh Without Rewrite: Minimal Changes That Trigger Recency
You don't need to rewrite the whole thing. Refresh without rewrite is a real tactic, and it's useful when you have 50 pieces to maintain and limited budget.
Swap out dated stats. If a 2023 post quotes a survey, find the latest edition of that survey. One or two fresh data points make a difference.
Add a "2026 update" callout box. At the top or bottom, note what's changed in the market, new tools released, or pivots you've seen. This signals you're paying attention without a full rewrite.
Update one section. If your post has 5 sections, pick the one most likely to be stale (trends, tools, pricing) and refresh it. Leave the fundamentals alone.
Retest a core example. If you have a how-to with a code snippet, tutorial, or workflow example, run through it again and note if anything broke or changed. Update accordingly.
Refresh the intro. The opening paragraph and summary box are the first things crawlers and readers see. A fresher intro signals recent attention.
Add a schema.org dateModified field. This tells engines when you last touched the page. Use the actual date you refreshed, not a fake one.
These are small jobs, not rewrites. Spread them across your content cluster on a schedule: high-traffic posts monthly, mid-tier posts quarterly, low-traffic posts annually. That's a content refresh impact without burnout.
Publication Date as a Citation Signal (And Its Limits)
Here's the gotcha: publication date alone doesn't make or break citations. It's a signal, not a law.
A page published two weeks ago on "best AI tools" can get cited if it's comprehensive and authoritative. But it won't automatically beat a page published four months ago if that older page has stronger topical authority, more inbound links, or better writing. Date is one signal alongside relevance and authority, not a sort order.
Also, the date matters for the specific query. On "how to write Python" (conceptual), publication date is weak. On "Python 3.13 release notes" (time-bound), publication date is strong. On "best Python IDE" (evolving but not news), it's medium. Understanding what is answer engine optimization means recognizing that recency is query-dependent, not universal.
One more thing: queries carry an implied recency. If someone asks "what's the latest in AI safety," they want 2026 sources. If they ask "how do transformers work," an explanation from 2018 can be fine. A page that matches the query's implied recency is the better citation.
Building an Update Cadence for Your Content Cluster
Start by bucketing your content:
Tier 1 (high-velocity topics): News, market trends, software rankings, pricing guides. Update monthly or every 6 weeks. These lose citation power fast.
Tier 2 (medium-velocity topics): How-tos for evolving tools, best practices in fast-moving fields, case studies. Update quarterly or every 6 months.
Tier 3 (low-velocity topics): Conceptual pieces, foundational how-tos, historical context. Update annually or when a significant change warrants it.
Tier 4 (evergreen topics): Philosophy, principles, timeless advice. Update only if your perspective shifts or a major fact is wrong.
Assign a person or process to each tier. Use content freshness signals from your tracking tool to identify which pieces are losing citation rank and bump them up a tier.
Here's the honest truth: if you're a small team, you can't maintain 500 posts on a monthly cycle. Pick your top 30–50 high-impact pieces. Keep those fresh. Let the rest age gracefully. An old, well-ranked post will still get citations if it's authoritative, even if it's not constantly updated. Your energy is better spent on the pieces that actually move the needle.
Also, consider a seasonal content update strategy AEO—refresh your content in concert with seasonal demand. Update your "best holiday gift guides" in September, not July. Update your "remote work trends" in January when new data drops. Align your refresh schedule to when LLMs are most likely to cite the topic.
Frequently Asked Questions
How often do I need to update content to stay cited in ChatGPT?
There's no single number. An engine that searches the web when it answers can reflect an update once it re-reads the page; what a model learned in training changes only when the model is retrained, and neither schedule is published. A quarterly review suits most content; check high-velocity topics monthly. Measure how quickly each engine picks up your updates rather than assuming.
Does updating just the publish date work, or do I need to change the actual content?
Change the content. A new date on an unchanged page gives readers and engines nothing new, and readers notice. Make a meaningful change—new stats, fresh examples, a rewritten section—then set the dateModified field to the real date. A small genuine refresh beats a fake dateline.
What's the difference between content staleness threshold and a ranking penalty?
Citation staleness threshold is when an LLM stops including your page in answer summaries, not because it's penalized, but because it's filtered out as outdated. A ranking penalty is different—it's an active demotion. You can hit the staleness threshold and still appear in search results if you have authority. The engines are just preferring fresher sources for summaries.
Should I update old content or write new content on the same topic?
Update if the core advice is still sound and the page has authority and traffic. Write new if you have a significantly different angle or if the old page is low-authority and low-traffic. For citation purposes, updating wins because you keep the authority signals (links, domain power) and add freshness. New content starts from zero.
How do I know if recency or authority is hurting my citations?
Test it: update the content substantively and track citations over the following weeks on each engine. If citations rise after the update, freshness was likely part of it; if nothing moves, look at authority and relevance first. Also compare your content to top-cited competitors—if they're older but still cited, your authority gap is bigger than your freshness gap.
Are there topics where freshness doesn't matter for citations?
Yes. Foundational how-tos (SQL basics, HTML structure), conceptual pieces (design systems, team building principles), and reference material (API docs) age slowly. Freshness helps, but it's not make-or-break. News, rankings, regulatory guides, and tool comparisons are the opposite—freshness is critical. Know which bucket your content sits in before setting your update cadence.
Can one update push a 2020 guide back into citation contention?
Often, if the page still has authority. If you update a 2020 CRM guide with 2026 research, re-rank the new leaders, and add current pricing, you can re-enter the citation pool. But the lift depends on how much the market has changed. If five new major competitors emerged, you're not just refreshing; you're rewriting. Also depends on the engine's training cutoff and whether you've lost authority (fewer links to the page).
How does publish date interact with last-modified date for LLM citations?
Both are visible to crawlers: datePublished says when the piece was first written, dateModified when it was last changed. Together they tell a reader, or an engine, that this is original work from 2022, checked in 2026, which is a better story than "written yesterday, no track record." A dateModified that lags far behind the topic's pace can look neglected. Keep dateModified current on active topics; don't sweat it on evergreen pieces.
Bottom Line
Recency in citation selection is real, but it's not a tiebreaker between authority and relevance—it's a tiebreaker when authority is close. A fresh page from an unknown site usually won't beat an established page from a well-known company, but a substantive update to your own established page can beat a competitor's newer one. The key is matching your update cadence to your content type: monthly for news and rankings, quarterly for evolving how-tos, annually for evergreen fundamentals. Test on your own pages across ChatGPT, Claude, and Perplexity. Small, substantive refreshes beat fake datelines.