You publish a page, it starts showing up in ChatGPT and Perplexity answers, and a few weeks later it quietly stops. Nothing on the page changed. A competitor’s page did, or the facts on the ground moved and yours did not. This is the question we hear most from marketing teams building an AI visibility program: how often should you refresh content to stay cited by AI? There is no single number of months that answers it. The right cadence depends on the page type, how fast the underlying facts move, and whether you are already losing citations right now. We treat this as part of running living content rather than a fixed publishing calendar, because AI systems reward pages that stay accurate and complete, not pages that were simply touched recently.
This post walks through a working cadence by page type, the trigger events that should pull you off that default schedule, and how to tell if your refresh pace is actually protecting citations or just keeping your team busy.
Why publish frequency is the wrong benchmark for AI citations
Most “how often should I update my blog” advice comes from an SEO world where freshness was one of many ranking signals among hundreds, and updating a date stamp could sometimes nudge a page in the right direction. AI answer engines do not work that way. A large language model is not scanning for a recent modified date. It is retrieving passages that answer a specific question well, checking whether the claims still hold up, and picking the source that states the answer most clearly and completely. Recency matters only when it changes the accuracy of the answer.
That distinction changes what “refresh” should mean. A page can be five years old and still get cited constantly, as long as the facts, numbers, and recommendations on it are still true. A page can be published last month and lose citations within weeks, if a competitor publishes a clearer answer or the underlying data shifts. Refresh cadence built around a calendar (update everything every quarter, or every year) treats every page as equally volatile. It is not. Some pages need attention every few weeks. Others are stable for a year or more. Applying one schedule to both wastes effort on pages that do not need it and leaves the volatile ones exposed.
In practice, we see teams fall into one of two traps. Either they refresh on a fixed calendar regardless of whether anything actually needs to change, which produces cosmetic edits that do not move citation rates, or they only refresh reactively after a page has clearly dropped out of AI answers, which means you are always a step behind. The cadence framework below is built to avoid both: a sensible default interval by page type, plus a short list of events that should move a page up the queue regardless of when it was last touched.
What actually determines how fast a page loses AI citations
Before you can set a cadence, it helps to understand what makes some pages decay faster than others in AI answers. A handful of factors do most of the work.
- How fast the underlying facts change. Pricing, feature sets, statistics, and regulations shift on their own timelines. A page tied to any of these decays as fast as the facts do, independent of how well it was written.
- How competitive the query is. If ten other sites are actively updating content that answers the same question, your page has to keep pace with the fastest mover, not with some average.
- Whether the page states a clear, extractable answer. Pages that bury the answer in narrative text lose citations faster than pages that state a definition, a number, or a recommendation plainly near the top. Extraction difficulty is itself a form of decay, even if the facts have not changed.
- How often the source pool for that topic gets refreshed. Topics tied to annual reports, seasonal data, or product release cycles see the whole citation landscape reshuffle on a predictable rhythm.
This is why the cadence question can’t be answered with a single number for a whole site. A glossary definition and a “vs” comparison page decay at completely different rates, because the first depends on a stable concept and the second depends on two companies that both ship changes. Setting cadence by page type, rather than by site-wide policy, is the fix.
A refresh cadence framework by page type
Here is the default schedule we use as a starting point for clients running a living content program, along with the trigger events that should move a page up regardless of the calendar. Treat the intervals as a floor, not a ceiling. If nothing has changed, a light verification pass is enough. If something has changed, refresh immediately rather than waiting for the interval to come around.
| Page type | Default refresh interval | Trigger events that override the default |
|---|---|---|
| Definition / glossary pages | Every 6-9 months, light verification pass | The concept’s common usage shifts, a competitor’s definition starts getting cited instead of yours, search intent for the term changes |
| Comparison pages (X vs Y, alternatives) | Every 3-4 months | Either product changes pricing, features, or positioning; a new competitor enters the comparison; review sentiment shifts materially |
| Stats-heavy pages (benchmarks, “state of” reports, data roundups) | Every 60-90 days, or immediately when a cited source updates | A source you cite publishes a new version, gets corrected, or gets retracted; your own first-party data collection cycle completes |
| Product / feature pages | Every 60 days, or on any feature or pricing change | A feature ships, changes, or is deprecated; pricing changes; a competitor repositions against you directly |
Notice the pattern: none of these intervals are longer than nine months, and the highest-volatility page types (stats-heavy and product) sit closest to two months. That is intentional. If a page type touches numbers or claims that change on someone else’s schedule, your review cadence has to be tighter than your instinct says, because you are not the one controlling when the underlying facts move.
One practical note on scope: “refresh” here does not always mean a full rewrite. Most cycles are a verification pass, confirming the facts, links, and examples still hold, with edits only where something has actually moved. Full rewrites are reserved for pages where the core claim itself needs to change. Treating every refresh as a rewrite is how teams burn a quarter’s editorial capacity on pages that only needed a five-minute check.
Trigger events that should move a page up the queue immediately
Calendar intervals are a safety net, not the primary control. The primary control is a short list of events that should send a page to the top of the queue the day they happen, regardless of when it was last refreshed.
- Lost citation. If you are tracking a fixed set of prompts and a page that used to get cited stops appearing, that is your clearest and most actionable signal. Investigate within days, not at the next scheduled review.
- Competitor takeover. A competing page starts appearing in the same answers where yours used to show up. This usually means their content got more current, more specific, or easier to extract than yours, and it is worth a direct comparison of the two pages.
- Stale claim discovered. Someone on your team, a customer, or a reviewer flags a statistic, price, or recommendation that is no longer accurate. Treat this the same way you would treat a factual error report on any page: fix it immediately, not on the next cycle.
- Source material update. A study, report, or dataset you cite gets a new edition. If your page still references the old numbers, it is now visibly out of date to anyone (human or model) that cross-checks it.
- Category-wide news event. A regulatory change, a major product launch in your space, or a widely reported shift in best practice can make an entire cluster of pages stale at once, even if none of them individually triggered a normal review.
These triggers matter more than the calendar because AI citation behavior is not smooth. A page does not slowly fade out of answers the way a search ranking might drift down over months. It can drop out abruptly once a better, more current source exists, and it may not come back just because you eventually get around to a scheduled refresh. Watching for triggers is how you catch the drop close to when it happens instead of a quarter later.
Building refresh cadence into a living content system
A cadence table is only useful if someone actually owns running it. This is where refresh intervals stop being a spreadsheet idea and become part of how your content operation works day to day. Our broader living content strategy treats every published page as an asset with an owner, a review interval, and a defined trigger list, the same way you would manage a piece of software that needs patching rather than a document that gets filed away once it is done.
In practice, that means three things. First, every page gets tagged with a page type from the framework above, so the interval is assigned automatically rather than argued about each time. Second, someone (usually a content or SEO lead, sometimes a dedicated content engineer) owns a standing queue of pages due for review, sorted by interval and by any active triggers. Third, the trigger events themselves need a monitoring source: a fixed prompt set you check periodically for citation drops, a lightweight alert when a cited competitor page changes, or simply a habit of checking review sentiment on comparison pages once a month.
None of this requires new software if you are running a small site. A shared tracker with page, type, last-reviewed date, and next-due date covers most teams. What it does require is treating the review as a real, recurring task with an owner, not an occasional cleanup project that happens when someone notices traffic has dropped.
Aligning cadence with seasonal and hub-level review cycles
Individual page cadence is one layer. The other is making sure your refresh schedule lines up with how your content hubs move as a whole. Our guide on seasonal content refreshes for evergreen hubs covers the broader pattern: hub pages and their supporting spokes often have a natural rhythm tied to industry events, reporting cycles, or seasonal demand, and it is worth reviewing the whole cluster together rather than page by page in isolation.
The connection to AI citations is direct. If your comparison pages and your definition pages both sit under the same hub, and the hub gets a seasonal review every quarter, that is a natural checkpoint to also run the page-level checks from the cadence table above. You are not adding a second parallel process. You are making sure the two schedules meet at the same point, so a hub-level review always includes a look at citation status for the pages under it, not just traffic and rankings.
This also helps with prioritization when you cannot review everything at once. If a hub-level check flags that overall organic visibility or engagement is softening across a cluster, that is a signal to move up the individual page reviews inside that cluster, even if none of them have hit their default interval yet. Cadence by page type gives you the floor. Hub-level signals give you the reason to move faster than the floor when the whole area is under pressure.
Pairing cadence with citation measurement, not just a schedule
A refresh calendar without a way to measure citation impact is just busywork with good intentions. You need to know whether refreshing on this cadence is actually protecting citations, or whether you are refreshing pages that were never at risk while the ones actually losing visibility go untouched. This is the piece that trips up teams who treat AI visibility the way they treated old-school SEO: publish, wait, hope the dashboard moves.
The practical version is a small, repeatable check: a fixed set of prompts tied to your key pages, checked on a regular interval, logging whether each page gets mentioned, cited with a link, or dropped entirely. Our post on evergreen content and AI search visibility goes into why “evergreen” does not mean “immune to decay” in AI answers, and that distinction is exactly why measurement has to sit next to cadence rather than replace it. A page can look evergreen by every traditional signal and still be quietly losing citations because a more current competitor page took its place.
If you are already tracking the broader visibility metrics that sit around this (mention rate, citation rate, share of voice inside AI answers for your core topics), tie the refresh log directly to that data. When a page’s citation rate drops two checks in a row, that is a trigger event on its own, on top of the ones listed earlier. Teams that already run an AEO measurement roadmap have a natural home for this: the refresh cadence becomes one of the levers you pull when the metrics say a page needs attention, not a separate process running on its own clock.
Common mistakes teams make with refresh cadence
A few patterns show up often enough to call out directly.
- Treating every page the same. A one-size cadence, whether monthly or annual, ignores that stats-heavy and product pages decay far faster than definition pages. You either over-refresh the stable pages or under-refresh the volatile ones.
- Confusing “updated” with “verified.” Changing the publish date or adding a sentence does not fix a stale statistic. AI systems (and careful readers) can tell the difference between cosmetic freshness and an actual correction.
- Waiting for a traffic drop before checking citations. Traffic and AI citation status do not always move together, especially early in a decline. By the time organic traffic clearly drops, you may have lost citations weeks earlier.
- No owner, no queue. Cadence tables that live in a strategy document but never turn into an actual assigned task list do not get executed. Someone needs to see “this page is due” on a specific day.
- Refreshing content in isolation from the hub. Fixing one page’s stats without checking whether the same claim appears (and is now wrong) on three related pages leaves obvious inconsistencies for anyone cross-checking your site.
Set cadence by what protects citations, not by the calendar alone
The real answer to how often you should refresh content to stay cited by AI is: often enough to catch the trigger events before they cost you a citation, and no more often than that for pages that are not moving. A page type framework gives you a sensible floor. Trigger events and citation measurement tell you when to move faster than that floor. Neither one works well without the other.
If your team is refreshing on instinct, or only reacting once a page has clearly dropped out of AI answers, that is usually a sign the cadence and the monitoring are not connected yet. Building both into a working system, with clear ownership and a real queue, is what turns “we should update our content more” into a process that actually protects the citations you have already earned.
We help marketing teams build exactly this: a page-type cadence, a trigger list tied to real monitoring, and a review queue with a named owner, so refresh effort goes where it actually protects visibility.
Content refresh cadence questions for AI citations
Quick answers on how often to refresh different page types, what trigger events should move up a review, and how this fits into a living content system.
How often should you refresh content to stay cited by AI?
There is no single interval that works for every page. Definition and glossary pages tend to hold up for six to nine months before they need attention, because the underlying concept rarely changes. Comparison pages need a look every three to four months, since the products on either side of the comparison keep shipping changes. Stats-heavy pages and product pages move fastest, often needing a check every 60 to 90 days.
The calendar is only a floor. Trigger events, like a lost citation or a competitor page taking your spot in an AI answer, should move a page up the queue immediately, regardless of when it was last reviewed. Treat the intervals as a starting point for planning capacity, not a guarantee that a page is safe until the date arrives.
Is refreshing for AI citations different from refreshing for SEO rankings?
They overlap but are not the same job. Traditional SEO refresh cycles were built around ranking signals, where updating a page could sometimes nudge it up a few positions even with modest edits. AI answer engines are retrieving passages to answer a specific question, and they check whether the claims still hold up rather than how recently the page was touched.
That means cosmetic updates, like changing a date stamp or adding a sentence, do very little for AI citations. What moves the needle is verifying the actual facts, numbers, and recommendations are still accurate, and making the answer easy to extract. A page can rank well in traditional search while quietly losing AI citations, which is exactly why the two need separate tracking, not one shared assumption that “updated recently” covers both.
How do I know if a page has lost its AI citations?
You need a fixed set of prompts tied to your key pages and a habit of checking them on a regular interval, logging whether each page still gets mentioned or cited. Without that log, the only signal you get is an eventual drop in organic traffic, which usually shows up weeks after the citation was actually lost.
Once you have a baseline, a lost citation is easy to spot: a page that used to appear in answers for a given prompt stops appearing, while nothing on your page changed. That is your clearest trigger to move the page to the top of the refresh queue. Pair this with the broader visibility metrics in your AEO measurement roadmap so a dropped citation shows up as a flagged event, not something you notice by accident.
Do stats-heavy pages really need refreshing more often than evergreen guides?
Yes, and it is one of the more counterintuitive parts of cadence planning. A page built around a definition or a stable concept can hold its citations for the better part of a year, because the concept itself barely moves. A page built around benchmarks, statistics, or a “state of” report is only as current as its weakest cited source, and those sources get updated on their own schedule, not yours.
When a study you cited gets a new edition, or your industry publishes fresher numbers, your page becomes visibly outdated the moment that happens, whether or not your calendar says it is due. That is why we set stats-heavy and product pages to a 60 to 90 day default, tighter than definition pages, and why a source update should trigger an immediate check rather than waiting for the interval.
What counts as a trigger event that should move up a scheduled refresh?
Five events matter most: a lost citation on a prompt you track, a competitor page appearing where yours used to show up, a stale claim someone flags (a wrong price, an outdated statistic, an old recommendation), a cited source publishing a new edition, and a category-wide news event that makes a whole cluster of pages stale at once.
Any one of these should send a page to the front of the queue the same week it happens, not at the next scheduled review. The point of a trigger list is to separate “nothing has changed, a light check is fine” from “something changed and this needs attention now.” Teams that only work off the calendar tend to miss the second category until traffic has already dropped.
Does every refresh cycle require a full rewrite?
No, and treating every refresh as a full rewrite is one of the fastest ways to burn through editorial time without protecting more citations. Most cycles are a verification pass: confirm the facts, numbers, links, and examples still hold up, and edit only the parts that have actually changed.
A full rewrite is reserved for pages where the core claim or structure genuinely needs to change, not for routine maintenance. If you find your team rewriting every page from scratch on every cycle, that is usually a sign the cadence intervals are too far apart, or the trigger events are not being caught early enough, rather than a sign that full rewrites are the right default.
How does refresh cadence fit into a broader living content strategy?
Cadence is one operating piece inside a larger system. A living content strategy treats every published page as an asset with an owner, a review interval, and a defined trigger list, rather than a document that gets filed away once it ships. Cadence tables only work when someone actually owns running the queue.
In practice, that means tagging pages by type so intervals are assigned automatically, assigning a standing owner for the review queue, and connecting the trigger events to real monitoring, whether that is a prompt set you check on AI answers or a simple watch on competitor pages and cited sources. Without that ownership layer, even a well-designed cadence table stays a document nobody executes.
