Key takeaways
- Profound's twelve-month study of 883,000 pages found the median page loses half its citation share within 11 days of peaking. 78% of cited pages fall to half their peak within two weeks.
- Engines do not decay together. Among pages that last past two weeks, a Claude citation is 2.8 times as likely as a ChatGPT citation to still be relevant eight weeks later.
- Pages cited by five or more engines are 5.5 times as likely as single-engine pages to still be above half their peak share at day 56.
- Promptwatch Data tracks a different thing (is a page ever cited again?) and shows landing pages and docs holding 67% to 69% of their citations at 60 days in ChatGPT, while news loses half its pages in 36 days.
- Treat reference-style pages as durable assets, treat news and reporting as bursts, and measure at the page and engine level.
Why citation lifecycles matter
SEO trained most of us to expect compounding returns. A page earns links over months, a domain earns trust over years, and a ranking, once won, is something you hold. Plenty of teams carried that assumption straight into AI search without checking it.
The 2026 data says it does not transfer cleanly. Authority in AI answers behaves more like a perishable good. Pages spike, get cited heavily for a short window, and then drop out of the source pool, often without any change on your side.
This guide pulls together the best public datasets I could find on how long AI citations last, explains where they disagree (they do, and for good reasons), and turns the findings into a content plan.
The headline study: an 11-day half-life
In a post published on September 30, 2026, Profound's Davis McCain reported a measurement of citation durability across 883,000 pages and seven engines: ChatGPT, Perplexity, Claude, Google AI Overviews, Google AI Mode, Gemini, and Microsoft Copilot. The window was twelve months. Each page is measured separately on each engine that cites it, which produced 1.19 million page-and-engine lifecycles.

How half-life is defined
For every page an engine cites, Profound tracks citation share by day. That share is how often the page is cited relative to the other pages in its category. The peak is the day of highest smoothed share. Half-life is the number of days after the peak that the page needs to fall to half its peak level and stay there.
A page qualifies only if it was first cited at least 28 days before the end of the window and had at least 5 citations in the 14-day window at its smoothed peak. The sample was a deterministic 1-in-128 draw of cited URLs.
That definition matters. It measures decay from a peak, not whether a page ever shows up again. Keep that in mind when we compare datasets below.
The numbers
- The median page loses half its citation share within 11 days of peaking.
- The median interval from first citation to the point of losing half its peak share is 42 days.
- 78% of pairings observed for at least two weeks past peak have a half-life of 14 days or less.
- The remaining 22% persist, with a median half-life of 27 to 29 days depending on engine, and a median rise from first citation to peak of 36 to 46 days.
The explanation Profound gives is the breadth of the source pool. When engines can draw from that many pages, the chance that any one page is cited a second time is low, so most pages show up in a single burst and never reach that rate again.
I find this the most useful framing in the whole study. Most of your citations are not a slow-building asset. They are events.
Engines do not move together
The median half-life varies little across engines. The differences show up after about four weeks.
Among the sustained pages, a Claude citation is 2.8 times as likely as a ChatGPT citation to still carry half its peak share two months out. ChatGPT, Google AI Mode, and Gemini tend to cite a page intensively over a few days and then stop.
The shape of the rise differs too. Among sustained pages, 62% of Perplexity's and 63% of Claude's had citations spread across weeks before the peak. On the other five engines that figure was 18% to 26%, and 37% to 46% of pages instead had citations concentrated into a few days.
The most practical finding is the correlation figure. When a page decays on one engine, it tells you almost nothing about when it decays on another. Cross-engine correlations run between 0.03 and 0.09. If you only watch ChatGPT, you cannot infer anything about Claude or Perplexity.
Breadth predicts durability
Profound's table of how many engines cite a page is the clearest result in the study:
| Engines citing the page | Pages | Median rise | Still above half-peak at day 56 |
|---|---|---|---|
| 1 | 679,077 | 19 days | 1.2% |
| 2 | 275,530 | 31 days | 2.0% |
| 3-4 | 190,439 | 33 days | 3.5% |
| 5 or more | 43,770 | 46 days | 6.8% |
A page cited by five or more of the seven engines is 5.5 times as likely as a one-engine page to still hold half its peak after eight weeks, and it takes more than twice as long to peak.
Two caveats. First, even the best group is only 6.8% at day 56, so "durable" here is relative. Second, this is correlation. Pages that every engine likes are probably pages that are authoritative, specific, and well structured, and those traits likely drive both the breadth and the lifespan.
What fades first: pages tied to a date
Profound classified 133,000 sustained pages by purpose and grouped them into four families. Reporting pages came last on all seven engines. The other three families held at 1.4 to 1.6 times the reporting rate.
Reporting pages also peak faster, with a median rise of 34 days against 41 to 43 for the rest. They arrive fast and leave fast. That is intuitive, and it is nice to see it confirmed at this scale.
A second lens: does ChatGPT cite the page again?
Promptwatch published a different kind of lifecycle study on October 5, 2026. The Promptwatch Data half-life report takes every classified page ChatGPT cited on July 6, 2026 and follows that cohort for 90 days through October 4. Each line shows the share of those day-0 pages that ChatGPT still cited on that day or later.

Why the two datasets look so different
They measure different things, and it is easy to misread them as contradicting each other.
| Profound | Promptwatch Data | |
|---|---|---|
| Engines | Seven | ChatGPT only |
| Sample | 1-in-128 sample of cited URLs over twelve months | Every classified page cited on July 6, 2026 |
| Metric | Time for citation share to fall to half its peak | Share of the cohort still cited again later |
| Question answered | How quickly does a page's citation intensity fade? | Does a page stay in the source pool at all? |
A page can lose most of its citation intensity after a burst and still be cited occasionally for months. Profound's metric catches the first. Promptwatch's survival measure catches the second. Both are true at once, and I would use both.
What Promptwatch Data found, by content type
From the ChatGPT cohort tracked for 90 days:
- Landing pages and docs lasted longest. 30 days out, ChatGPT still cited 78% of landing pages and 76% of docs. At 60 days it was still citing 69% and 67%.
- Comparisons held 72% at 30 days and 63% at 60 days. Listicles held 60% at 60 days.
- Product pages decayed faster than you might expect. 55% were still cited at 60 days, level with reviews and how-to guides, and all three reached half-life after 74 to 77 days.
- News was the fastest to fade. Almost a quarter of news pages were never cited again after day 0, half were gone after 36 days, and 41% were still cited at day 60.
One technical note from the report: lines drop faster in the last week because fewer days remain for a page to be cited again. That is a window-end artifact. Landing pages, docs, comparisons, and listicles only cross 50% in the final week because of it, so their true half-life is likely close to or beyond 90 days.
Read next to Profound's finding that reporting fades first, the picture is consistent. Date-bound content decays fast on every measure. Reference content, such as docs and landing pages, sticks.
The environment itself is unstable
Page-level decay is only part of it. The citation pool also shifts under your feet.
BrightEdge looked at week-to-week movement and found that among the roughly 3% of domain-prompt combinations that changed, 87% were declines. Over 51% of citation volume was associated with declining domains. Most changes were binary: domains did not lose a few points of share, they went from cited to not cited on a given prompt. Only about 0.4% of tracked domains gained new citations that week, so losses were not flowing neatly to new winners. BrightEdge's analysis calls it a tightening citation radius.
Platform changes add step-change shocks on top. Promptwatch Data recorded reddit.com's share of ChatGPT citations falling from about 4% to about 0.5% on August 14, 2026. Earlier, average citations per ChatGPT response dropped around the GPT-5.3 rollout on March 4, 2026. Neither event had anything to do with the quality of the pages involved.
The lesson: some of what looks like decay is the engine changing its behavior. Annotate your charts with known platform events before you blame your content.
What this means for a content plan
Split your content into two portfolios
A durable portfolio: landing pages, docs, pricing, comparisons, and category listicles. These stay in ChatGPT's source pool the longest according to Promptwatch Data, and they are the pages most worth keeping accurate. Stale prices or outdated specs on a page that gets cited for months is a bad trade.
A burst portfolio: news, press, reports, and event content. Expect a fast rise and fast fade. Use it for short spikes of visibility and for feeding durable pages with fresh data, not as a base for steady citations.
Plan refreshes around the clock, not the calendar
If the median burst is over in about two weeks, a quarterly refresh cycle misses most of the action. Product pages are worth watching specifically: Promptwatch Data shows them reaching half-life around 74 to 77 days in ChatGPT, so pricing, availability, and spec changes should not wait on a content calendar.
Aim for breadth, not just ChatGPT
Because cross-engine correlations are near zero, single-engine monitoring will mislead you. A page that is fading on ChatGPT may be rising on Claude or Perplexity. Pages cited by five or more engines were the most durable in Profound's data, so formats and structures that travel well across engines deserve priority. Clear factual claims, strong structure, and fast, accessible pages are a reasonable starting point.
Do not treat a spike as a win
A week of heavy citations is an event. Judge a page by whether it is still above half its peak after eight weeks, not by its best day.
Measuring lifecycles yourself
You need three things: citation data per page and per engine, a way to separate access problems from content problems, and a place to log platform events.
- For ChatGPT, Perplexity, Gemini, Copilot, and AI Overviews at once, a platform that tracks real-interface citations by page is the baseline. Promptwatch, Profound, Peec AI, and Otterly.AI all sit in this space, with different depth. Promptwatch also adds IP-verified AI crawler logs and a Page Tracker for watching specific URLs, which helps tell whether a fading page is being crawled and not cited, or not crawled at all.
- Google Search Console is still useful for seeing whether a page that fades in AI answers keeps its organic footing.
- For a wider look at what is available, the GEO software directory at bestgeosoftware.com lists many of the options.


A few definitions to settle before you build a dashboard:
- Pick a decay metric and stay with it. Share-of-peak and ever-cited-again are different questions.
- Track per page and per engine. Do not average across engines.
- Require a minimum citation count at peak (Profound used 5 in 14 days) so tiny samples do not produce noisy half-lives.
- Note the window-end effect when your tracking period is short.
Quick reference: what to expect by content type
| Content type | Signal from the data | Suggested approach |
|---|---|---|
| Landing pages and docs | 67% to 69% still cited by ChatGPT at 60 days | Keep accurate; treat as long-term assets |
| Comparisons | 63% at 60 days | Maintain and expand |
| Listicles | 60% at 60 days | Refresh regularly |
| Product pages | 55% at 60 days; half-life 74 to 77 days | Keep specs, availability, and prices current |
| Reviews and how-tos | 55% at 60 days | Update when products change |
| News and reporting | Half gone by day 36; fades first on all seven engines in Profound's data | Use for bursts, not baseline |
Limits of the evidence
Both studies come from vendors with a product to sell, and both have clear scope limits. Profound's data is a 1-in-128 sample and filters to pages with enough citations to measure. Promptwatch's cohort is a single day's cited pages on one engine. Neither tells you causation, and neither says what to change on a specific page to extend its life.
It is also a fast-moving target. Engines change their behavior several times a year, as the Reddit and GPT-5.3 events show, so re-check any benchmark every quarter instead of treating it as a constant.
For more first-party citation research across engines, the Promptwatch Data hub publishes regular reports, including which content types ChatGPT cites the most.
Bottom line
Citation lifecycles are short for most pages and long for a minority. The pages that last tend to be reference-style, broadly cited across engines, and not tied to a date. Build around those, use news and reports for bursts, and measure each engine on its own. If your reporting still shows one blended citation number, split it by page and engine this week.
