--- language: en title: "How Long Does an AI Citation Last? A Look at the Profound Study" description: "The study tracks the declining share of pages cited by seven engines. Learn about the 11 days, the methodology, and how to plan reviews without chasing fluctuations." canonical: "https://tropk.ai/en/resources/research/how-long-does-an-ai-citation-last-a-look-at-the-profound-study" last_modified: "2026-10-12T16:14:23.004Z" --- # How Long Does an AI Citation Last? A Look at the Profound Study Reviewed on 2026-10-06. ## The finding and the question it actually answers This is Tropk’s editorial analysis of research published by Profound on September 30, 2026, and revised on October 6. The data is from Profound. The useful conclusion is that a page’s presence as a source can change quickly and differently across engines. The 11-day figure describes the median half-life of [citation share](/en/resources/glossary/citation-share) after its peak; it does not describe an article’s shelf life, the duration of a visit, or an obligation to update the entire site every eleven days. For marketing, the productive question is: which pages support important answers, for how long, and in what context? A fleeting citation on a peripheral topic may have little value. A persistent reference in response to a decisive question deserves attention, but you still need to check what the answer says about the brand. Frequency alone does not demonstrate trust, preference, or revenue. ## Key points - The 11 days in Profound’s research describe the median half-life of citation share after the peak. - The analysis tracks 883,000 pages across seven engines from September 2025 to September 2026. - The measure uses smoothed share and a sustained decline; a single answer does not define the end of a citation. - Share loss can result from changes in the denominator, questions, sources, or the engine itself. - Review pages for errors, gaps, and commercial relevance, without turning 11 days into a mandatory update schedule. ## How the research defined duration The [original methodology](https://www.tryprofound.com/blog/the-half-life-of-an-ai-citation-is-11-days) tracks 883,000 pages from September 8, 2025, to September 8, 2026, across seven engines. Each page-engine pair is a separate unit: 1,188,816 qualifying combinations. The measure uses daily share smoothed with a 14-day average. The decline must remain at half the peak or below for 14 consecutive days. Cases with no observed decline by the end of the study are treated as censored in the survival analysis. This definition avoids treating a single answer as the end of a cycle. It also means the page’s publication date does not start the clock. You need to know the first observation, the peak, and the subsequent tracking period. Without that history, applying the term half-life to a monthly report would simply mean using the name of a different measure. ## What the numbers can distinguish The median interval between the first citation and half the peak was 42 days. Among combinations observed for at least two weeks after the peak, 78% lost half their share within 14 days. The study also finds little synchronization between engines: for most engine pairs, lifecycle correlations range from 0.03 to 0.09; among Google’s three surfaces ([AI Mode](/en/resources/glossary/ai-mode), [AI Overviews](/en/resources/glossary/ai-overviews), and [Gemini](/en/resources/glossary/gemini)), they range from 0.12 to 0.17. These findings are in the [Profound report](https://www.tryprofound.com/blog/the-half-life-of-an-ai-citation-is-11-days); they are not measurements made by Tropk. A peak and a recurring baseline call for different decisions. A page about a new development may attract concentrated references while the topic is being discussed. Integration documentation may support questions for longer. This distinction should come from reviewing your own business’s pages and questions, without assuming that any editorial format has a guaranteed lifespan. ## Why a decline alone does not identify the cause Share depends on the denominator. A page may maintain the number of references yet lose relative share because other sources start appearing more often. It may also disappear from some questions, continue appearing in others, or receive fewer citations after a product change. That is why analysis should show absolute counts, share, the set of questions, and the engine. Adding up all platforms into a single curve can hide opposing movements. Updating a page after a decline and seeing a recovery does not prove that the update caused the return. Engine changes, news, competitors, and answer fluctuations are still possible factors. The research describes observed cycles; it does not randomly assign pages to editorial interventions. A responsible decision should document these alternatives before attributing an outcome to content work. ## How to apply this to a Brazilian company’s content library Start with pages related to real commercial questions: selection criteria, features, integrations, availability, and contract terms. Match each page to the questions it should help answer. Separate content that depends on recent facts from relatively stable explanations. This classification guides reviews based on information needs, without turning the publishing calendar into a goal. For a pricing page, a review makes sense when terms change or are poorly explained. For a guide, it may make sense when context is missing or a reference no longer supports the claim. Changing the date without changing the information solves neither problem. Before producing a new version, check accessibility, available content, and whether the page fits the question. The article does not provide a specific breakdown for Portuguese or Brazil. Therefore, the 11 days should not become a Brazilian benchmark. The local value lies in tracking by engine and intent and discovering the trajectory of your own pages. Comparisons with foreign companies should take language, category, and questions into account, as well as domain size. ## An editorial test that does not confuse movement with proof Choose a small group of pages with demonstrably incomplete information and a similar group without that gap. Document the problem before editing. Keep questions, location, and engines consistent during tracking; record the date and content of each change. The comparison group helps show whether movement occurred across the entire topic, but it does not eliminate all differences between pages. As an operational proposal, track for four weeks before expanding the intervention. This period is a starting point for the work, not a timeframe derived from the study or a promise of recovery. Analyze citations shown, factual accuracy, and coverage of important questions. If the answer cites the site for an incorrect claim, an increase in the count is not an improvement. ## What to record and how to decide what to do next - Save the exact question, engine, date, cited URL, and the passage it supports. Record collection errors separately from answers with no references. - Compare counts and share using the same set of questions. Explain changes in the baseline instead of hiding their effect in an average. - Prioritize reviews when there is an error, missing information, or recurring loss on a relevant question. A one-off fluctuation calls for investigation, not an automatic rewrite. ## Resources for turning analysis into action See the difference between [AI citation](/en/resources/glossary/ai-citation) and [citation rate](/en/resources/glossary/citation-rate), and how to evaluate [content updates](/en/resources/glossary/content-updates). The [trend and update prompt](/en/resources/prompts/prompt-for-writing-about-a-trend-or-update) helps structure a brief with evidence and dates. To organize recurring observations and action owners, learn about [Monitoring](/en/product/monitoring) and [Strategy](/en/product/strategy). The analysis should maintain a verifiable history; the tool does not turn association into causation. ## Sources and references - [The half-life of an AI citation is 11 days]() — Profound — Davis McCain ## Frequently asked questions ### What does the 11-day half-life of AI citations mean in Profound’s study? According to Profound’s definition, 11 days is the median half-life of a page’s citation share after its peak. The measure uses a 14-day average and requires a sustained decline to half the peak or less for 14 consecutive days. It does not represent an article’s shelf life or the duration of a visit. ### Does Profound’s study recommend updating every article every 11 days? No. Profound’s research describes observed citation-share trajectories; it does not establish a mandatory update schedule. Pages should be reviewed when information is incorrect, missing, outdated, or there is recurring loss on relevant questions. Changing only the publication date does not fix these problems. ### Does a decline in AI citation share prove that a page’s content has gotten worse? No. Share depends on the denominator: a page may maintain its number of references yet lose relative share when other sources appear more often. Changes in questions, products, models, and sources can also affect the trajectory. Compare absolute counts, share, and context by engine before attributing the decline to content quality. ### How can you track the duration of AI citations for a Brazilian brand’s pages? Match pages to real commercial questions and record the question, engine, date, cited URL, and supported fact. Keep the sample comparable, distinguish collection errors from answers with no references, and document content changes. Profound’s study does not publish a Brazil-specific breakdown, so local trajectories need to be tracked without applying 11 days as a benchmark for Brazil. ## Continue a leitura - [Content updates]() - [AI citation]() - [Prompt for writing about a trend or update]() - [Citation rate]() - [Response variability]() - [AI response monitoring]() - [AI discovery strategy]() ## About the author [Luiz Gardelin]() Luiz Gardelin is the founder of Tropk.ai and works in GEO/AEO, product vision, strategy, and growth, helping brands understand and improve their presence in AI responses. --- [All pages in Markdown]()