There's no universal refresh frequency for AI citations. Set the cadence by how fast the facts behind each page change: monthly for prices, tools and model comparisons, quarterly for tactical guides, and every 6-12 months for evergreen explainers. Only make updates that change substance, like new data, corrected facts or new sections. Ahrefs' analysis of about 17 million citations found AI assistants cite content that is 25.7% fresher than organic search results, with ChatGPT showing the strongest preference.
Key Takeaways
- Citation half-life is how long a page keeps getting cited before it loses its place. It varies by topic and by platform.
- ChatGPT prefers fresher sources the most. Google AI Overviews showed virtually no freshness preference over organic results in Ahrefs' July 2025 study.
- Use three refresh tiers based on how fast the facts change, not one site-wide schedule.
- A real refresh changes facts, examples or structure. Changing only the date is noise and can backfire.
- Track citations per page monthly, and trigger refreshes when citations drop, not only by the calendar.
- Some widely shared freshness stats have no public methodology. Rely on primary studies.
What the evidence actually says
Many freshness articles repeat claims like "50% of AI citations are under 13 weeks old", usually credited to Lily Ray and Amsive. The articles I checked while researching this didn't link to the original methodology, so I'd treat that figure as directional until you've read the source yourself.
The most transparent primary data I found is Ahrefs' July 2025 study of 16.975 million citations across seven AI platforms, using its Brand Radar data:
| Platform | Average age of cited URLs | Compared with organic results |
|---|---|---|
| ChatGPT (citations) | 958 days | 458 days newer |
| Copilot | 1,056 days | 360 days newer |
| Gemini | 1,118 days | 298 days newer |
| Perplexity | 1,166 days | 250 days newer |
| Google AI Overviews | 1,432 days | 16 days older |
Two things stand out. First, the average cited page is still close to three years old, so freshness is a preference, not a hard rule. Second, the preference is uneven: strong in ChatGPT, close to nil in AI Overviews. That uneven picture is why one cadence for every page wastes effort.
Defining citation half-life
I use "citation half-life" as a working term: the time it takes for a page's citation rate on its target prompts to fall to half its peak. It isn't an industry-standard metric. It's a practical way to decide when a page needs attention.
Things that shorten half-life:
- Facts with dates (prices, versions, rankings, "best of 2026")
- Fast-moving categories (AI models, ad platforms, regulations)
- Strong competition publishing newer takes
Things that lengthen it:
- Original data nobody else has
- Definitions and fundamentals
- Pages widely referenced by third parties
The three-tier refresh cadence
| Tier | Page types | Refresh cadence | What to check |
|---|---|---|---|
| 1: Volatile | Tool comparisons, pricing, AI model guides, platform policies, "best X 2026" | Monthly, plus on major news | Prices, versions, feature lists, launches |
| 2: Tactical | How-to guides, playbooks, checklists | Quarterly | Screenshots, UI steps, examples, new tactics |
| 3: Evergreen | Definitions, frameworks, case studies | Every 6-12 months | Accuracy, new examples, internal links |
Example from this month: any page comparing AI models needed updating after OpenAI released GPT-6 "Astra" on 3 September 2026 and Anthropic released Claude Fable 5.1 on 1 September 2026. A Tier 1 page left alone through a month like that goes stale right away.
What counts as a real refresh
A refresh should give a reader something they didn't have before. My checklist:
- Facts: re-verify every number, price and date. Replace or remove anything you can't confirm.
- Answer capsule: rewrite the opening 40-70 words so it reflects the current answer.
- New section or example: add at least one real addition, such as a new case, a new tool or a new step.
- Remove dead weight: cut outdated tactics rather than leaving them in with a note.
- Links: fix broken outbound links and add links to newer related posts.
- Dates: update "last updated" and
dateModifiedin schema, and only after the changes above.
Google's own guidance on helpful content warns against changing dates to make pages look fresh when the content hasn't really changed. Treat that as the rule for every engine.
Calendar-based vs trigger-based refresh
Calendars keep you disciplined. Triggers keep you efficient. Use both:
Calendar: Tier 1 monthly, Tier 2 quarterly, Tier 3 twice a year.
Triggers (refresh straight away):
- Citations on the page's prompt panel drop by half month over month
- A competitor publishes a clearly better or newer page on the same query
- The underlying fact changes (price change, product launch, new law)
- Search Console shows impressions steady but clicks falling sharply on the page's main queries
- Server logs show AI user-triggered fetches (ChatGPT-User, Perplexity-User) falling on the URL
Measuring citation decay
A simple setup:
- For each important page, write 3-5 prompts it should answer.
- Run them monthly in ChatGPT, Perplexity, Gemini and Google AI Mode.
- Log "cited / mentioned / absent" per prompt per platform.
- Chart the citation rate per page over time.
- Mark refresh dates on the chart to see if updates bring citations back.
AI answers vary from run to run, so run each prompt two or three times, or use a tool that samples repeatedly. Don't react to a single missing citation.
Prioritising when you have 300 posts
You can't refresh everything. Score each page:
- Business value (1-3): does it drive leads or sign-ups?
- Volatility (1-3): how fast do its facts change?
- Citation gap (1-3): is it losing citations, or never had them for prompts it should win?
Refresh pages scoring 7-9 first. When I've run this on edtech blogs, roughly a fifth of posts carried most of the value. The long tail usually needs pruning or merging more than refreshing.
Refresh vs rewrite vs retire
- Refresh when the structure is sound and the facts are dated.
- Rewrite when search intent has moved (for example, a "how to" that now needs a comparison format).
- Retire and redirect when the topic is dead or duplicated elsewhere on your site.
Merging two thin, overlapping posts into one strong page often brings back citations faster than refreshing both.
Team workflow
- Owner per Tier 1 page, with a monthly calendar slot
- Change log at the bottom of each updated post ("Updated September 2026: added GPT-6 pricing; removed discontinued tool")
- Editor sign-off on facts before the date changes
- Quarterly review of the tier assignments themselves. Pages move between tiers as topics settle or heat up.
FAQ
How often should I update content for AI search?
Base it on volatility: monthly for pages with prices, tools or AI model details, quarterly for tactical guides, and every 6-12 months for evergreen explainers. Also refresh straight away when a key fact changes.
Does ChatGPT prefer newer content?
Ahrefs' 2025 study found ChatGPT cited content that was on average 458 days newer than organic search results, the strongest freshness preference of the platforms studied. Older pages still get cited when they're the best source.
Does updating the publish date help AI citations?
Changing only the date without changing the content isn't a real refresh, and Google advises against it. Update the date after substantive changes, and reflect that in the visible "last updated" line and the schema.
Do Google AI Overviews favour fresh content?
In Ahrefs' July 2025 study, AI Overviews showed virtually no freshness preference over organic results. Cited URLs were slightly older on average. Quality and relevance seem to matter more there than recency.
What is content citation half-life?
It's a working term for how long a page keeps being cited by AI assistants before its citation rate falls to half its peak. It isn't a standard metric, but it's useful for deciding when a page needs a refresh.
Is the "50% of AI citations are under 13 weeks old" stat reliable?
It's widely repeated and usually credited to Lily Ray and Amsive, but many articles quoting it don't link the methodology. Treat it as directional and check the original source before building strategy on it.
Should I refresh old blog posts or write new ones?
If an old URL has links and history, refreshing it usually beats starting a new page on the same topic. Write new pages for genuinely new queries, and merge overlapping old posts.
How do I track whether a refresh worked?
Compare the page's citation rate on its prompt panel for 4-8 weeks before and after the refresh, alongside Search Console clicks and AI referral traffic. Mark refresh dates on your tracking chart so the cause and effect are clear.
Want a refresh system that keeps you cited?
I'm Younus Fardeen, and in 4+ years of marketing for edtech and startup brands I've found that a good refresh process often beats publishing more. If your older content is losing ground in AI answers, I can help you set up tiers, triggers and tracking that fit your team. Explore my work and get in touch via the contact form at younusfardeen.in.