Content Freshness

1. What content freshness is

Content freshness describes whether a page’s age and update history help or hurt its chances of being retrieved and used to ground an answer.

Freshness combines two properties that behave very differently:

• Recency concerns the timestamp. It measures how long it has been since the page was published or last changed. This is a machine-readable fact.
• Currency concerns the substance. It describes whether the page’s claims remain true. This is a semantic fact that no timestamp can establish.

The distinction matters because each anti-pattern in §7 attempts to improve recency without improving currency. The practices in §6 correct the substance first and then update the date to reflect that work.

Freshness can affect two stages in the answer loop. At retrieval, the updated version must first be present in the index, which depends on recrawling. At grounding and reranking, the engine chooses among candidate passages. A genuinely updated page may appear unchanged to an engine if the new version has not been fetched. Within GEO, freshness is one signal rather than a complete strategy.

2. Recency versus currency: What each can prove

The relationship is asymmetric: a publisher can assert recency but not currency. A publisher can set dateModified to any value, but cannot simply declare that every claim remains true.

Engines therefore treat stated dates as claims that need corroboration, not as facts in their own right. The practices in §7 can damage trust rather than merely fail to help. In E-E-A-T §4.4, Trustworthiness includes the “accuracy, transparency, currency” grouping. This is separate from citability, which concerns whether a passage can be extracted. A current page must still be extractable, and an extractable page must still be accurate.

3. What the evidence shows and what it does not

Evidence that AI systems prefer recent content comes in two forms. One body of work is observational, while the other is counterfactual. Each supports a different conclusion.

3.1 Correlational evidence: AI systems cite fresher content, but “fresher” does not mean new

The largest public dataset comes from an Ahrefs analysis of 16.975 million cited URLs across ChatGPT, Perplexity, Gemini, Copilot, AI Overviews, and organic Google results (Law & Guan, July 2025). It found that AI-cited URLs were 25.7% “fresher” than organic results.

The 25.7% difference compares an average age of about 1,064 days for cited URLs with about 1,432 days for organic results. In other words, the average AI-cited page is roughly 2.9 years old.

Here, fresher describes a comparison with an old baseline; it does not mean new. The aggregate results contradict the idea that publishers must post every week to retain citations. Across engines, the pattern was consistent: ChatGPT favored the newest content among the major assistants, while Google AI Overviews behaved much more like organic search than the other AI surfaces.

This evidence is observational. Fresher pages may receive more citations because of their age, or because newer pages also tend to be better structured, better linked, and more closely aligned with current questions. Correlation cannot separate those possibilities. The counterfactual experiment in §3.2 isolates the effect of the stated date.

3.2 Causal evidence: Recency bias appears in the model, not just the corpus

Fang, Tao, Chen, Chang, and Sakai (2025) provide counterfactual evidence that isolates recency more directly than observational data can.

The authors prepend artificial publication dates to passages from the TREC Deep Learning collections (DL21 and DL22), then repeat the LLM reranking process. The passage text remains unchanged; only the stated date moves. Any resulting change in ranking therefore comes from the date.

The result has a troubling implication: the bias responds to the stated date, not to an observed change in the content. That makes the stated date open to manipulation. §7 explains the drawbacks of exploiting it.

3.3 How to interpret the evidence

Vendor marketing on this topic often repeats figures from other vendors without adding evidence. Reliable conclusions depend on three types of sources: original-data studies, peer-reviewed or preprint research, and first-party engine documentation. Other sources can indicate a pattern, but they cannot support precise conclusions.

Related evidence also limits how much recency can accomplish. When a model chooses between conflicting sources, topical relevance dominates, while stylistic credibility markers have little effect (Wan et al., ACL 2024). Freshness can influence that choice, but it cannot replace topical fit.

4. Freshness depends on the query

The average cited page in §3.1 is about 2.9 years old, yet vendor studies report that half of AI citations are less than 13 weeks old. Both findings can be accurate because the value of freshness depends on the query, and the two samples included different types of queries.

This mechanism comes from classical search and remains part of generative retrieval. Recency receives more weight when the answer is likely to have changed since publication. Google’s ranking-systems guide calls these Freshness systems and describes “various ‘query deserves freshness’ systems designed to show fresher content for queries where it would be expected” (A Guide to Google Search Ranking Systems). The phrase where it would be expected makes the condition explicit. The system does not apply a blanket preference for new pages.

The review cadence should match how quickly the facts can change. Rewriting a definitional page every month offers almost no benefit and creates the risks described in §7. Leaving a pricing page untouched for a year, however, compromises its currency no matter how well the page is structured.

Freshness also varies across engines and search experiences:

These findings show broad tendencies. The day counts for individual platforms come from one vendor study that does not disclose its query mix, so they should not be treated as fixed benchmarks. Citability §5 applies the same caution to GEO benchmark figures.

5. What actually signals freshness to an engine

Engines assess freshness through five channels. Publishers fully control only three of them.

The two channels outside the publisher’s control determine whether engines believe the other three.

Google documents this verification process. It says the sitemap lastmod value is used “if it’s consistently and verifiably (for example by comparing to the last modification of the page) accurate.” A significant update changes the main content, structured data, or links; a copyright-date change does not qualify (Build and Submit a Sitemap). Google also advises publishers to label a prominent visible date, mark it up with datePublished or dateModified, keep the visible and structured values consistent, and avoid future dates or the date of the event being described (Add a Byline Date).

Google’s AI-features documentation does not describe freshness as an eligibility requirement. It says that “there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary,” and that no special schema.org markup is needed (AI features and your website; Optimizing for generative AI features). Date markup helps an engine distinguish among several dates on a page; it does not create freshness by itself.

6. When to review and update content

6.1 The five decay triggers

A page needs revision when one of these triggers occurs, not simply because a calendar interval has elapsed.

You cannot detect the second and third triggers by reading your own page. They require the measurement methods described in GEO Metrics.

6.2 What counts as an update

6.3 How to set the review cadence

Use the table in §4 to set a review interval by query class, then revise only when one of the triggers in §6.1 occurs.

The interval prompts a review; it does not require an edit. When a quarterly review is treated as a quarterly update, a page with no factual drift may receive only a new date. A review that finds nothing has changed is still successful and should end without an edit.

GEO Wiki uses an internal nextReviewDue field to schedule the next review. Most reviews leave lastUpdated unchanged.

7. Date-bumping and other freshness anti-patterns

The experiment in §3.2 showed that injected dates can move rerankers. Injected dates can therefore manipulate reranking results, but the resulting costs make the practice inadvisable.

Google itself identifies the first practice as an example of search-engine-first content. Its self-assessment guidance asks publishers: “Are you changing the date of pages to make them seem fresh when the content has not substantially changed?” (Creating Helpful, Reliable, People-First Content).

A page should become more recent because its claims were brought up to date. Changing the date alone reverses that relationship.

Date-bumping can produce an effect; it demonstrably moved lab rerankers. The problem is that any gain is small and temporary, while the loss of trust is neither small nor temporary. The Fang et al. result describes a bias that engines have an incentive to mitigate, not a lasting advantage. Deliberate manipulation of date signals is also a recognized spam pattern; see GEO Spam and Manipulation. The benefit to one publisher also erodes once competitors optimize the same factor (C-SEO Bench, Puerto et al., NeurIPS ‘25 D&B).

8. Why freshness matters for GEO and what to do

Among the signals used for grounding, freshness is inexpensive to fake but costly to maintain. That cost is what gives genuine currency its value. Freshness is also the only major signal that degrades passively. Citability and E-E-A-T decay slowly, if at all; a page’s currency can deteriorate when an external fact changes.

References

Academic:

• Fang, H., Tao, S., Chen, N., Chang, K.-X. & Sakai, T. (2025). Do Large Language Models Favor Recent Content? A Study on Recency Bias in LLM-Based Reranking. SIGIR-AP ‘25. arXiv:2509.11353 · ACM DL · paper summary
• Wan, A., Wallace, E. & Klein, D. (2024). What Evidence Do Language Models Find Convincing? ACL 2024 Main. arXiv:2402.11782 · paper summary
• Puerto, H., Gubri, M., Green, C., Oh, S. J. & Yun, S. (2025). C-SEO Bench: Does Conversational SEO Work? NeurIPS ‘25 Datasets & Benchmarks. arXiv:2506.11097

Original-data studies:

• Law, R. & Guan, X. (2025). New Study: AI Assistants Prefer to Cite ‘Fresher’ Content (17 Million Citations Analyzed). Ahrefs

Official platform documentation (as of 2026-07):

• Google Search Central: Add a Byline Date to Google Search Results · Help Google Search know the best date for your web page
• Google Search Central: Build and Submit a Sitemap
• Google Search Central: AI features and your website · Optimizing your website for generative AI features on Google Search
• Google Search Central: A Guide to Google Search Ranking Systems (the “Freshness systems” entry) · Creating Helpful, Reliable, People-First Content