AI citation decay and content freshness: How often should you update articles?

ChatGPT and other assistants cite newer URLs on average

In July 2025, Ahrefs reported an analysis of approximately 16.97 million cited URLs, including AI citations and organic Google results.

Across ChatGPT, Perplexity, Gemini, and Copilot, cited URLs averaged 1,064 days since publication, compared with 1,432 days for organic results. The AI average was 25.7% younger. Time since the last update averaged 909 days for AI citations and 1,047 days for organic results. Google AI Overviews differed: its cited URLs averaged 16 days older than organic results. (Ahrefs)

The average publication age of about 2.9 years shows that older pages remain part of AI answers. The analysis measured publication and update dates; the accuracy of the information on each page requires a separate check.

Use current factual accuracy to decide whether an article needs revision. Prices, product specifications, market data, regulations, and comparisons can change. Definitions, theory, and historical facts can remain accurate for years. Check the claims and their sources to identify the passages that need updating.

Match update dates to substantive changes

Date information should reflect what changed on the page.

Google's Article structured data uses datePublished for the original publication date and dateModified for the most recent modification. Structured data provides machine-readable information about the article. A sitemap's lastmod field records its last significant update, helping Google identify changes worth recrawling. (Google Search Central article documentation) (Google Search Central sitemap guidance)

For example, replacing statistics with a new study and revising the conclusion warrants corresponding changes to the visible update date, dateModified, and lastmod. Keep datePublished set to the original publication date.

The cited Google guidance supports accurate date information. Its scope covers article metadata and sitemap use; it provides no evidence that changing a date alone increases AI citations.

Set audit frequency around how quickly information changes

The intervals below are operational guidelines based on content type and how often its facts change. Writesonic recommends similar intervals for content refreshes. Use them to schedule checks for stale information. (Writesonic)

Page type	Suggested audit interval	What to check
Pricing, features, comparisons, rankings, and statistics	Every 3–6 months	Figures, specifications, rankings, comparison tables, and sources
Tool tutorials and case studies	Every 6–12 months	Instructions, interfaces, product names, and reported results
Definitions and conceptual guides	About once a year	Examples, sources, links, and factual changes

An audit establishes whether the information still holds. Accurate content can remain as written. A price change or major product update calls for a review when it happens, even if the next scheduled audit is months away.

Record the factual changes found during each audit so the reason for a revision is clear.

Track ChatGPT citations at shorter intervals

ChatGPT had the shortest reported citation half-life in Scrunch's platform comparison, at 3.4 weeks. That finding can inform a monitoring schedule. (Scrunch)

For questions connected to key products or customer acquisition, weekly or fortnightly checks can reveal changes in cited URLs. Choose the interval according to the importance of the question and the amount of variation observed.

When your page disappears from an answer, compare it with the newly cited pages. Check for newer statistics, revised pricing or specifications, additional primary sources, and more specific answers. Those differences identify possible revisions.

Record the changes you make so later citation observations can be compared with the revised content.

Citation persistence differed across Stacker network domains

Stacker and Scrunch compared sources on domains within the Stacker publisher network with sources on domains outside it. The reported half-life was nearly 10 weeks within the network and about 4.5 weeks outside it, an approximately 2.1-fold difference. For ChatGPT, the figures were 7.2 and 3.4 weeks; for Perplexity, 10.4 and 5.7 weeks. (Stacker)

The comparison covered different groups of domains, including Stacker's own publisher network. Establishing the effect of third-party publication on a particular article would require a different comparison. These findings show an association between network membership and citation persistence.

Alongside your own update history, record whether external publications reference the same company, product, or research. That information can help investigate differences between sources that remain cited and those that disappear.

Use factual changes to decide what needs updating

Citation sources can change within weeks while pages published years ago still appear in AI answers. Citation persistence and publication age measure different things.

Decide what to revise by checking current citation URLs, facts on your page, and information gaps relative to other cited pages. Identify outdated claims or missing evidence and update the relevant passages.

Please Cite Me! runs a technical audit and content audit on a public URL to identify issues affecting how AI search can access, understand, verify, and cite the page.

Audit a page whose citations have declined or an important article that has gone a long time without review to check its access, content structure, and supporting evidence.

Audit a page with Please Cite Me! →