Last updated September 2026.
A page that gets cited today is not guaranteed a citation next month. Generative engines re-rank their source pool constantly, and the pages that fall out of it do so on a predictable clock. That clock now has a name and a number.
The decay curve behind citation half-life
Scrunch and Stacker tracked 3.5 million citation events across ChatGPT, Google AI Overviews, Perplexity, Gemini, and other platforms, from September 2025 to March 2026. The finding: the median time for half of a single page’s citations to drop off, measured from that page’s own peak, is 4.5 weeks.
That is the strict definition of citation half-life. It is not how old a page is. It is how long a page keeps the citation rate it already earned.
The same dataset found that content under 13 weeks old accounts for roughly half of all AI citations across the sample. Fresh pages are not just favored at launch. They dominate the pool for as long as they stay fresh.
A separate analysis backs up the shape of that curve at the page-age level. Kevin Indig’s study of 1.2 million ChatGPT citations found citation likelihood falls off fast once content passes its first month:
| Content age | Relative citation likelihood (indexed to under 30 days = 100) |
|---|---|
| Under 30 days | 100 (peak) |
| 1 to 3 months | 68 |
| 3 to 6 months | 31 |
| 6 to 12 months | 12 |
| 12+ months | 5 |
Source: Kevin Indig, analysis of 1.2 million ChatGPT citations, reported via Loamly, February 2026.
Content under 30 days old earns roughly 3.2x more citations than older pages in the same dataset. That gap is the whole reason half-life is worth tracking: it is not a soft trend, it is a curve steep enough to cut your citation rate by two-thirds in the first quarter alone.
How fast each engine forgets a citation
Half-life is not one number. It varies by engine, because each engine retrieves and re-ranks sources on its own schedule.
| Engine | Refresh cycle | Freshness pattern |
|---|---|---|
| ChatGPT | Shortest | Citations skew hard toward content under three months old |
| Google AI Overviews | Middle | Favors content three to six months old on a roughly monthly cycle |
| Perplexity | Longest | Six-week-old and older authoritative content stays active |
Source: Scrunch and Stacker, 3.5 million citation events, September 2025 to March 2026 (reported via Searchbloom, May 2026).
Seer Interactive’s separate study of 47,097 citations across 7,683 dated pages, from March to June 2026, found the same ranking a different way. Gemini cited pages updated within the past year 78% of the time. ChatGPT did so 73% of the time and leaned on blogs and guides for 46% of its citations. Perplexity, the most tolerant of older material, updated within the past year only 65% of the time, pulling evergreen reference pages the other two engines had already dropped.
Google AI Overviews was not part of Seer’s sample, but it draws on Google’s core search index rather than a separate retrieval layer. That lines up with Ahrefs’ broader finding that AI-cited pages run about 25.7% fresher on average than pages ranking in classic organic results (Ahrefs, analysis of 17 million citations, 2025).
(Every figure on this page carries its own source and date, the same bar our methodology holds every dataset on this site to.)
What a full year of neglect costs
The week-to-week curve above explains why a page loses its footing fast. Gander’s analysis of 194,077 URLs across ChatGPT, Google AI Overviews, Gemini, and Perplexity found the longer-horizon version of the same pattern: content has roughly a one-year half-life, and each year of age cuts a page’s AI visibility by 40% to 60% (Gander, April 2026).
Two-year-old, unmaintained content runs at about a third of its original visibility. Three-year-old content falls below a quarter. Neither figure is a cliff. It is the same decay curve you saw above, just stretched across years instead of weeks.
How to measure your own citation half-life
You do not need a data science team to get a working number. You need a weekly cadence and one page to watch.
- Pick a page and a prompt set. Use the queries that page already gets cited for.
- Run the prompt set weekly and log the citation rate. A single run per week is enough once you have a stable prompt set.
- Mark the peak week. That is your baseline.
- Count forward to the week the rate first falls to half the peak. That gap, in weeks, is the page’s measured half-life.
- Repeat across your top pages. Half-life varies by content type, so a single page will not tell you your whole site’s pattern.
Lightweight, weekly trackers are the natural fit for this. Peec AI and Otterly.AI both run prompts on a weekly schedule and keep a rolling citation history per prompt, so the peak-to-half-life gap shows up in the data you already have, without a manual export.
If you want half-life tracking folded into a broader GEO workflow, along with the diagnosis and content work to act on what you find, Temso covers weekly tracking across eight AI engines from $89/mo, with unlimited projects and users on every plan.
For enterprise teams that need years of citation history across a large content library, Profound goes deeper on volume, starting at $99/mo for single-engine coverage and $399/mo for full engine access.
Track the curve before it drops
A citation you cannot see decaying is a citation you will lose without warning. The GEO tool ranking compares every platform that logs citation history on a weekly cadence, so you can pick the one that matches your engine mix and start measuring your own half-life this week.