A September 30, 2026 study from Profound reports a median AI citation half-life of 11 days after a page's peak. The measure describes declining citation share in the company's sample, rather than an expiration date for the information on a page.
A citation is a source link included in an AI answer. For a publisher, the useful question is how often an article continues to be selected as a source, and how that pattern changes across different answer services.
What AI citation half-life measures
Profound analyzed 883,000 pages across seven answer engines over 12 months, producing approximately 1.19 million page-and-engine records. It measured citation share relative to other pages in a category, using a trailing 14-day average.The clock starts at the peak. The measured decline must reach half that smoothed share and persist there for 14 days. In the study, the median interval from a page's first citation to half its peak share was 42 days.
The starting point changes the meaning of the result. "Eleven days after peak share" and "eleven days after publication" describe different timelines. A page can remain accurate and accessible after its share of citations declines.
Why each answer engine needs its own record
The study found different citation patterns across engines and weak correlations between their lifecycles. It also reported an association between being cited across more engines and longer persistence.For a team maintaining AI documentation, this suggests a useful measurement design: record the same question separately for each service. A combined total can conceal that one service still cites a page while another selects a different source. Peec AI's analysis of ChatGPT retrieval provides related context on how source selection works.
These are observational findings from Profound's measurement system. They help frame what to track; they do not establish a publishing schedule that guarantees future citations.
Updating an article when the answer needs to change
Take a documentation page explaining an API parameter. If the accepted values change, revise the explanation, examples and relevant links, then record what changed. If the parameter is unchanged but the page appears less often in answers, first inspect the questions and competing sources before rewriting it.A useful editorial log includes the query, answer engine, observation date and cited page. Add the reason for any content revision. That connects maintenance to the quality of the answer and gives the team evidence to compare after the update.



