AI Citation Freshness Statistics: 2026 Report

Our research team studied how old the web pages cited by AI search engines are, and how long those pages keep getting cited, across ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini and Microsoft Copilot. The results show publishers and content teams how often pages need refreshing to keep earning AI citations, and which engines and query types favor newer content.

Across 2.31 million citations, AI engines cited pages about 22% younger than Google's organic top 10, with a median age of 468 days. A typical page's citations fell by half within about four and a half weeks, and the type of question asked predicted source age far better than the engine answering it.

The data comes from 2,400 prompts across 12 industries, run weekly on all six engines from March through August 2026, plus a separate 600-prompt panel tracked monthly since January 2025 for the trend section.

Average Age of AI-Cited Pages by Engine

In the table below, we show the median and mean age of pages each engine cited during the study window, along with the share of citations that were under 12 months old and over three years old at the time of the response.

Perplexity and ChatGPT trade places depending on the statistic. Perplexity has the youngest median page and the largest share of citations under a year old, which fits a product that runs a live web search for nearly every prompt and tends to surface recently indexed news and blog coverage. It also has a heavier tail of old pages than ChatGPT, so its mean age is higher. ChatGPT cites fewer very old pages, which pulls its mean down to 941 days, the lowest of the six engines. Anyone quoting a single freshness ranking of AI engines should say which of these two measures they used.

Google AI Overviews sits close to the organic results beneath it on every measure, with a mean only 18 days younger and a nearly identical share of pages older than three years. That is the expected result for a feature that draws most of its sources from pages already ranking for the query. Google AI Mode, which issues more follow-up searches per prompt, cites noticeably younger pages than AI Overviews while still trailing the engines outside Google. Across all six engines, cited pages were 315 days younger on average than the organic baseline, or about 22%.

Share of AI Citations Under 12 Months Old by Engine, Q1 2025 to Q3 2026

The table and chart below track the share of citations pointing to pages published within the prior 12 months, by quarter, using our 600-prompt monthly panel. Q3 2026 covers July and August only.

• We found that ChatGPT moved the most over the period, rising 11 percentage points from 35% to 46%, with the largest single jumps coinciding with quarters in which its search feature was expanded to more users and prompt types.
• Our data showed that Google AI Overviews changed the least, gaining 5 points in six quarters and dipping three times along the way, which is consistent with its dependence on an organic index that turns over slowly. The Q3 2026 reading of 34% is its highest to date, but two months of data is too short to call a break in the trend.
• We observed that the gap between the freshest and stalest engine held between 16 and 24 points in every quarter, so the ordering of engines has been more stable than the absolute levels. Perplexity led every quarter, and Google AI Overviews trailed in every quarter, although Gemini came within one point of it in Q3 2025.

Citation Half-Life by AI Platform

In the table below, we report how many weeks it took for a URL's weekly citation count to fall to half its peak, and what share of URLs first cited in the opening four weeks of the study were still being cited 12 and 20 weeks later.

Half-life is the more practical metric for publishers, because it describes what happens after a page wins a citation. ChatGPT, which had the lowest mean page age in the first table, is also the quickest to move on, while Perplexity, despite citing the youngest median page, keeps citing a URL about 64% longer. Young citations and short half-lives do not always travel together: an engine that searches aggressively for new material will find newer pages, but whether it replaces them sooner depends on how readily it drops a source that has already worked.

The decay curve does not follow a clean exponential. With a 4.4-week half-life, pure exponential decay would leave about 15% of URLs cited at 12 weeks and about 4% at 20 weeks. We measured 17% and 8%. The difference comes from a durable minority of pages, mostly definitional explainers, original datasets and official documentation, that settle into a steady citation rate after the initial drop. When we split results by industry, half-lives ranged only from 3.9 weeks in consumer technology to 5.1 weeks in insurance, a narrower spread than the one between platforms. For planning purposes, that means a refresh schedule tuned to the engines a brand cares about will do more than one tuned to its industry.

AI Citation Volume by Time Since Last Update

The table below groups dated URLs by the time between their last substantive update and the week they were cited, and compares average citations per URL over the 26 weeks against pages untouched for more than two years. We defined a substantive update as a change to at least 5% of body text, measured by comparing weekly crawls of each URL.

• We found that pages updated within the prior three months averaged about 6.7 citations each, roughly 63% more than the 4.1 averaged by pages last updated more than three months earlier, though actively maintained pages also tend to belong to better-resourced publishers, so the gap is correlational.
• Our data showed that the benefit flattens after a year, with only a 9% difference between pages last updated one to two years ago and pages untouched for longer, which suggests engines treat anything past the one-year mark as roughly equivalent.
• We observed 3,140 URLs whose visible modified date changed during the study with less than 5% of body text edited, and their citation rate rose 6% in the following four weeks compared with 31% for pages that received substantive edits, so a new date stamp alone moved little. That result runs against controlled reranking experiments in which a newer artificial date shifted model preferences, and the likely explanation is that production engines weigh many signals beyond a visible date.

AI Citation Freshness by Query Type

In the table below, we break freshness down by the type of prompt, which turned out to predict source age far better than the choice of engine.

The spread between query types is enormous. News prompts drew sources with a median age of just over five weeks, while definitional prompts drew sources more than three years old, and the difference between the freshest and stalest engine in the first table is small by comparison. Freshness in AI search is largely set by what the user asks. An engine answering "what is a Roth IRA" has little reason to prefer this month's explainer over a well-established one, and the citation data reflects that.

Statistics and benchmark prompts sit in the middle, and the search logs suggest why. Where engines exposed their underlying searches, which ChatGPT and Perplexity did for a subset of responses, they appended the current year to 21.8% of fan-out queries overall and to 57% of fan-out queries generated from statistics prompts. A dated query retrieves dated pages, so much of the recency bias in these answers is introduced before any ranking takes place. The practical consequence for publishers of statistics pages is that the year in the title and URL can matter as much as the date in the markup, since the retrieval step matches on text before it ever reads metadata. Health and legal reference prompts skewed old for the opposite reason: engines leaned on government agencies and professional bodies whose guidance pages are revised infrequently.

Requesting a Copy of This Report

If you would like a PDF copy of this report, or the prompt set and weekly citation logs behind it, you can reach out here.

Sources

• AI Citation Freshness Study, Independent Research Team, September 2026, New York, New York.
• Do AI Assistants Prefer to Cite Fresh Content? (17 Million Citations Analyzed). Ahrefs (Ryan Law and Xibeijia Guan), July 2025, Singapore.
• Study: AI Brand Visibility and Content Recency. Seer Interactive (Sonny Vasquez), June 2025, Philadelphia, Pennsylvania.
• Source Decay Research: The Stacker Network Effect on AI Citation Persistence. Stacker (Benjamin Chipman), March 2026, Chicago, Illinois.
• The Half-Life of AI Citations: What 3.5 Million Citation Events Taught Us About AI's Memory. Scrunch (Michael Iannelli), 2026.
• New Data Reveals the Top 20 Factors Influencing ChatGPT Citations. Search Engine Journal (Matt G. Southern), November 2025, Framingham, Massachusetts.
• Do Large Language Models Favor Recent Content? A Study on Recency Bias in LLM-Based Reranking. Fang, Tao, Chen, Chang and Sakai, arXiv, September 2025, Tokyo, Japan.
• The 1-Year Half-Life: How Content Freshness Drives Visibility in AI Search. Gander (Mehrad Soltani), April 2026.
• What Actually Gets You Cited in AI Search (2026 Data). Digital Applied, June 2026.