Content Decay: Why Pages Fade From AI Answers, and the 13-Week Refresh That Holds Them

Content decay is the loss of AI citations that follows when a page stops being updated. AI assistants cite pages that are on average 25.7% fresher than Google's organic results, and 76.4% of the pages ChatGPT cites more than any others were updated within the last six months (Ahrefs, 2025). Pages left untouched for a quarter are three times more likely to lose their AI visibility (AirOps, 2025). The fix is a quarterly refresh of the pages that matter.

A guide can rank on page one of Google for three years and still be cited by Perplexity exactly zero times. The ranking says the page is good. The silence says it is stale. Those are two different judgements, made by two different systems, which is why ranking on Google and being cited by an AI engine can come apart.

This is the gap a content library falls into when it is left alone. The pages were written well, ranked early and were picked up by the AI engines in their first weeks. Then nobody touched them.

A content calendar without a refresh schedule is half a GEO strategy.

A note on this update. The June 2026 version of this guide quoted a figure that half of AI-cited content had been updated within 13 weeks. We could not trace that figure to a primary source, so it has been removed. Every number below links to the study it comes from.

What is content decay?

Content decay is the decline in a page's effectiveness over time. In classic SEO it meant rankings slipping as fresher competitor pages displaced yours. In AI search it shows up in citations: the page stops being one of the sources an assistant quotes.

The mechanism is simple. An engine choosing between two comparable sources treats the more recently updated one as the lower accuracy risk. A statistic from 2023, a market description from 2022, a list of current best practices written before AI search was mainstream: each one lowers confidence in the whole page.

That creates a structural problem for any business that built a content library without a refresh schedule. Its best pages lose ground to newer ones every quarter they sit still.

What does the evidence say about freshness and AI citations?

Three studies carry the weight:

• Ahrefs, July 2025, 17 million cited URLs: AI assistants cite pages that are on average 25.7% fresher than the organic results for the same topics. ChatGPT cites pages 458 days newer and lists the newest sources first.
• Ahrefs, October 2025, the 1,000 pages ChatGPT cites more than any others: 76.4% of the dated pages had been updated within the previous six months.
• AirOps, December 2025: pages not updated within a quarter were three times more likely to lose their visibility in AI answers. AirOps does not publish its full method, so read it as directional.

Freshness does not weigh the same in every engine. The same Ahrefs study found that Google AI Overviews cite pages 16 days older than the organic results, and that the average page cited by any assistant is still 2.9 years old. Decay is sharpest in ChatGPT and Perplexity and mildest on Google's own surfaces. A page does not fall off a cliff at a fixed age. Its odds shorten as newer pages arrive.

Why a quarter?

Thirteen weeks is our working cycle, not a threshold found in data. We chose it for three reasons. The one study that measured a refresh interval used a quarter as its cut-off. The platforms themselves change every quarter, so statistics and tool content go out of date on roughly that rhythm. And a quarterly cycle is one a small team can keep: for a library of 30 articles it means about ten refreshes a month, a little over two a week.

Google adds a constraint worth respecting. Its guidance on publication dates asks that a visible date reflect a real change to the page, and its helpful-content guidance says that changing pages to look fresh does not help. A refresh cycle is a schedule for real updates. It is not a schedule for changing the date.

What does content decay look like in practice?

Content decay is invisible unless you measure AI visibility, and a standard SEO dashboard does not.

Picture a professional-services firm that publishes a Cyprus company-formation guide in late 2024. Through early 2025, AI engines cite it and the referral traffic is measurable. By late 2025, with no change to the page, the citations thin out. By early 2026 the guide barely appears in AI answers while still ranking strongly on Google.

The SEO dashboard shows a healthy post. Rankings are flat. Organic traffic is flat. Nobody sees the decline, because nobody is tracking it.

When freshness fades, an engine falls back on the other signals a page offers, and on this island those are thin. In September 2026 Axenor Consulting checked the homepages of 20 accounting, law and corporate-services firms in Cyprus. Nine carried no structured data at all, 11 gave an engine no organisation record to read, none carried FAQ markup, and one of the 20 firms had a Wikidata entity. In June the same check on 10 firms found seven without proper structured data. When the freshness signal weakens and the structural signals were never there, the page has nothing left to be cited on. That is the case for an entity record the engine can confirm, and our AI visibility checker tests a site for the same gaps.

What do our own numbers show?

This guide is the page of ours that Google's AI features show more than any other article, which is why it was refreshed first. In the 90 days to 18 September 2026, Search Console recorded 16 impressions for it inside AI Overviews and AI Mode. The other eight guides earned 5 between them.

It ranked first for a question on why a static content strategy decays faster in AI search, third for one on training a team to spot content decay, and between seventh and eighth for two questions on how quickly AI citations decay. The volumes are small: 77 impressions in total. The pattern is the useful part. Every one of those queries is long, conversational and phrased the way someone types into an assistant, and each maps to a heading in this guide.

We will report what this refresh does. It went live on 24 September 2026, with a correction to the refresh arithmetic on 25 September, and the page is tracked in the month-end review described in our piece on setting a blogging cadence for SEO.

Which content decays fastest?

Not all content decays at the same rate. Knowing which types go first lets you set the schedule by page type instead of by guesswork.

How do you refresh a piece of content?

A refresh is not a rewrite. It is a targeted update that restores accuracy and makes the page easier to quote. For a 1,500-word article it takes 60 to 90 minutes. The libraries we keep on this cycle are on our work page: one Limassol client had 34 of its 57 pages appear inside Google's AI features in August 2026, according to Search Console.

Step 1. Update every statistic. Find each figure. Check whether newer data exists. Update the number and the year, and link the source. If the original source has nothing current, find an equivalent recent one. If no source can be found at all, remove the figure. That is what happened to the statistic this guide used to open with.

Step 2. Rewrite or strengthen the direct answer. The 40 to 90 words at the top of the page should still be the best current answer to the core question, with a current data point in it. If the page has no direct answer, add one. It is the highest-impact change in a refresh, and the same block does the heaviest lifting in answer engine optimisation.

Step 3. Update the FAQ section. Add one question that has emerged since the page was published, and correct any answer that has gone out of date.

Step 4. Check every link. Sources move and studies get superseded. A dead link under a statistic undoes the work of citing it.

Step 5. Show the date, and move it only when the content has changed. Put the updated date where a reader can see it, and set the modified date in your CMS so the AI crawlers your robots.txt lets in reassess the page. A date change with no content change is cosmetic, and Google says as much.

How do you build a content refresh calendar?

A refresh calendar is a spreadsheet with five columns: URL, original publish date, last refresh date, next refresh date, and a priority keyed to AI visibility.

To decide what to refresh first:

1. List your top 10 pages by organic traffic.
2. Open the generative AI report in Search Console and note which of them appear in Google's AI features.
3. Ask ChatGPT, Gemini and Perplexity the core question each page answers, and note whether you are cited.
4. Pages that rank well on Google and are not cited by AI are priority 1.
5. Pages already cited are priority 2: hold them on the 13-week cycle.
6. Low-traffic pages with no AI visibility come last.

For a library of 30 articles, that works out at about ten refreshes a month, or 10 to 15 hours of maintenance at 60 to 90 minutes each. Refreshes on this calendar are part of every SEO and AI search retainer we run.

Why is refreshing better than writing new content?

A refresh has a structural advantage new content cannot match: it inherits what the page has already earned. A page that has been indexed for two years has links, a crawl history and a record of being useful. A new page starts with none of that.

There is no published study of how fast a refreshed page returns to AI answers. The closest evidence covers new pages. Semrush found in December 2025 that Google AI Mode cited 36% of new pages within 24 hours, with citations coming and going from day to day, while ChatGPT cited 10% on day one and 42% within 30 days and kept citing them once it had started. That study used a domain far stronger than a typical Cyprus business site, so expect slower.

So the GEO play for a business with an existing library is refresh first, create second. Create new content to fill real gaps, usually the ones that complete a topic cluster. Refresh existing content to defend the visibility you already earned.

Key takeaways

• AI assistants cite pages 25.7% fresher than Google's organic results. ChatGPT leans hardest on recency, Google's own AI features least.
• 76.4% of the pages ChatGPT cites more than any others were updated within six months.
• A quarterly cycle is a working rule, not a threshold. What matters is that each refresh is a real change.
• Refresh first, create second. A refreshed page keeps the links and history a new page has to earn.

Frequently asked questions

How do I know if my content is being cited by AI engines? Start with the generative AI report in Google Search Console, which shows impressions inside AI Overviews and AI Mode for each page. Bing Webmaster Tools reports citations in Copilot. For ChatGPT and Perplexity, ask the core question each page answers once a month and record whether you are cited.

Does every page need a 13-week refresh cycle? No. Stable topics such as definitions and fundamental concepts can run on a 26-week cycle. Fast-moving topics such as AI search, platform features and market statistics belong on 13 weeks or tighter. Service pages change when the offer changes.

Does refreshing content hurt SEO rankings? No. Updating the data, the direct answer and the structure improves a page for Google as well as for AI engines. The real risk is changing the angle of the page or cutting the sections that earn its rankings, which a targeted refresh does not do.

What is a direct answer block, and why add one to every page? It is a passage of 40 to 90 words at the top of a page that answers the core question completely and makes sense when lifted out by itself. It is the passage an engine is likeliest to quote. Adding one to a page that lacks it is the highest-impact change in a refresh.

Does updating the date without updating the content help? No. Google's guidance on publication dates asks that a date change reflect a real change to the page, and its helpful-content guidance says that making a site look fresh does not improve rankings. Engines also read the content: an old statistic under a new date is still an old statistic.

How fast does a refreshed page come back into AI answers? No published study measures refreshed pages. For new pages, Semrush found in December 2025 that Google AI Mode cited 36% within 24 hours and ChatGPT cited 42% within 30 days, on a strong domain. We are tracking this guide after its September 2026 refresh and will publish what we see.