Key takeaways :
- 76.4% of pages cited by ChatGPT were updated in the last 30 days (Ahrefs, 2025).
- Freshness is a confirmed ranking factor across 7 AI models tested in October 2025.
- Refreshing a strategic page can increase its traffic by 106% (HubSpot).
- Prioritization method: GSC (positions 5 to 20, CTR below 3%), monthly cadence, real refresh not cosmetic.
Why evergreen content no longer exists in 2026
Evergreen content used to promise one thing: write once, rank for years. That promise no longer holds. Generative platforms (ChatGPT, Perplexity, Gemini, Claude) massively favor recent content.
The Ahrefs analysis published in 2025 on millions of AI citations quantifies this bias: 76.4% of pages cited by ChatGPT were updated in the last 30 days before their citation. On Google, the same content can sit at the top for 18 months. In LLMs, it disappears within 60 days if not maintained.
Freshness is not an editorial preference. It is a measured ranking factor. Independent research by Metehan Yesilyurt published in October 2025 on 7 AI models (GPT-4o, GPT-4, GPT-3.5 and several LLaMA and Qwen variants) confirms that freshness scoring ranks among the top 3 extraction factors across all tested models.
AI platforms prefer content that is 25.7% fresher than what is cited in classic organic search. Evergreen is not dead because it no longer delivers value. It is dead because generative engines no longer see it.
What “refresh” really means (and what it is not)
A refresh is not a publication date update. Changing “2023” to “2026” at the top of an article changes nothing in the ranking, nor in LLM citation. Platforms analyze substance, not display.
A real refresh touches at least 3 dimensions:
- Numerical data: stats, studies, benchmarks. An article citing a 2022 study is penalized compared to an article citing the same source in its 2026 version.
- Examples and use cases: replace outdated examples (discontinued tools, acquired brands, removed features) with current cases.
- Semantic structure: add new sections on emerging angles, remove obsolete parts, update named entities.
Two secondary levers add to these 3 main axes: updating outbound links (dead or redirected), and adjusting the overall angle if the topic has evolved.
The most common mistake is to refresh form without touching substance. LLMs detect this gap. The ranking doesn’t move, and the LLM citation moves even less. An effective refresh modifies at least 20% of the original text.
The GSC prioritization method to identify pages to refresh first
You can’t refresh every page. The question is not “what to refresh” but “what to refresh first”. I systematically apply the following prioritization method, based on Google Search Console, across client accounts.
Filter by position between 5 and 20
Pages ranking between positions 5 and 20 are the priority pool. They are already recognized by Google as relevant, but they don’t reach the top 3 that captures 75% of clicks. A well-targeted refresh can propel them to position 3 or better.
Pages beyond position 30 generally require heavier structural work than a simple refresh: they sometimes warrant a complete rewrite, or even a merger with another page.
Filter by CTR below 3%
Among pages in positions 5 to 20, filter those with a CTR below 3%. This signal indicates either an unengaging title, a meta description misaligned with intent, or a SERP saturated by other formats (featured snippet, carousel).
These pages are where a combined refresh (title, meta, content structure) produces the fastest gains. Expect 30 to 60 days between the update and the first observable results.
Cross with volume and business potential
The final step is to rank the filtered pages by two criteria: search volume of their main keyword, and proximity to your offer (a top-of-funnel page weighs less than a bottom-of-funnel page).
The top 10 or top 20 that emerges from this crossing constitutes your refresh roadmap for the quarter. You work through it in order, at a pace of 2 to 4 pages per month depending on your editorial bandwidth.
A HubSpot study cited in 2025 documents that a well-executed refresh can increase a page’s traffic by 106%. The result is replicable across an entire portfolio of pages provided this prioritization logic is respected.
The ideal refresh cadence by content type
Cadence depends on page type. A news article dies in 3 months, a technical guide lives 18 months, a product page must be reviewed at every offer change.
| Content type | Refresh cadence | Nature of the refresh |
|---|---|---|
| Statistics / benchmarks article | 1 to 3 months | Update figures and sources |
| News / trend article | 2 to 3 months | Partial rewrite |
| Technical guide / tutorial | 6 to 12 months | Verify procedures and add content |
| Pillar article / evergreen | 4 to 6 months | Add sections, refresh examples |
| Product / pricing page | At every offer change | Complete update and monthly maintenance of social proof |
| Case study | 6 to 12 months | Update the results figures |
Rule to remember: the more a piece of content depends on numerical data or external entities (brands, tools, people), the tighter the refresh cadence must be. A conceptual guide on on-page SEO can live 12 months. An article on “best SEO tools 2026” must be refreshed quarterly to remain citable.
How to measure the real impact of a refresh on your LLM visibility
An effective refresh is measured on two complementary fronts. The first is the classic Google ranking, tracked in Search Console. The second, and most important in 2026, is citation within generative engines.
To measure LLM impact, you need before-and-after data. I apply the following method:
- Before the refresh: capture the page’s AI Share of Voice across 20 to 30 strategic prompts, via a tool like Cockpyt AI. This provides a measurable baseline.
- After the refresh: wait 30 to 60 days for LLM indexes to integrate the new version, then rerun the same prompt panel.
- Comparison: you obtain a citation delta directly attributable to the refresh. No speculation, no reasoning by analogy.
Without this before/after measurement, you refresh blind. You may know that it works. You don’t know how much it works, on which prompts, or against which competitors.
FAQ on content refresh and LLM citations
How often should I refresh a blog article?
Cadence depends on content type. A statistics article requires a refresh every 1 to 3 months. A technical guide can live 6 to 12 months. General rule: the more a piece of content contains dated data, the tighter the cadence must be.
Does a simple date change count as a refresh?
No. Modifying the publication date without touching the content produces no effect on Google ranking or LLM citation. Algorithms detect the absence of real text modification. A valid refresh modifies at least 20% of the content.
Does Google penalize refreshes that are too frequent?
No. Google values freshness, including monthly refreshes, as long as each update brings new value (updated figure, added section, replaced example). Penalty risk only appears if you manipulate dates without touching substance.
Can a refresh degrade a page’s ranking?
Yes, in two cases. First case: a rewrite that radically changes the targeted search intent. Second case: a modification of title and H1 tags that disrupts signals Google had already indexed. A successful refresh keeps the page’s main semantic anchor intact.
What’s the difference between refresh and cannibalization?
A refresh updates an existing page. Cannibalization occurs when you create a new page targeting the same keyword as an existing one, diluting authority between the two. Before creating a new article, always check whether refreshing the old one would suffice.
How do I know if a refresh worked?
Two KPIs to track: evolution of the average position in Google Search Console on target keywords (wait 30 to 60 days), and evolution of the AI Share of Voice across a strategic prompt panel. The second KPI is harder to measure but more informative in 2026.
Should product pages and landings be refreshed?
Yes. These pages carry most of the conversions and face the strongest competition both in SERPs and in LLMs. Recommended cadence: monthly audit, complete refresh at every offer change, continuous update of social proof (reviews, client figures).
Sources
- Position Digital, AI SEO Statistics 2026, https://www.position.digital/blog/ai-seo-statistics/
- Ahrefs, Fresh Content and AI Search (analysis of 17 million AI citations), 2025, https://ahrefs.com/blog/fresh-content/
- Metehan Yesilyurt, Independent AI Ranking Factor Research on 7 LLM models, October 2025 (relayed by Position Digital).
- HubSpot, Historical Content Optimization Study (relayed by SearchAtlas in 2025), https://searchatlas.com/blog/seo-statistics/
