How long does ChatGPT keep citing a page? Half-life by content type

Pages ChatGPT cites do not stay cited forever, and the rate of decline depends heavily on the type of page. In Promptwatch's ChatGPT content half-life report, published October 5, 2026, 78% of landing pages and 76% of documentation pages cited on day zero were still being cited 30 days later. News pages fared worst: almost a quarter were never cited again after day zero, and half were gone after 36 days.

The method matters, so here it is. Promptwatch took every classified page ChatGPT cited on July 6, 2026 and tracked each one for 90 days, through October 4. That gives a clean cohort, and it also gives a hard edge, which affects how some of the numbers should be read.

The ranking by staying power

Here is the order, built from the figures in the report.

Documentation is the stickiest. 76% was still cited at 30 days and 67% at 60 days, and 93% of documentation pages were cited again at some later point, against 77% of news pages.

Landing pages are close behind: 78% at 30 days and 69% at 60 days.

Comparisons keep 72% at 30 days and 63% at 60 days.

Listicles keep 60% at 60 days, so they are a bit below the comparison pages but above the pages that follow.

Product pages, reviews and how-to guides end up together. About 55% of each was still cited after 60 days, and all three reach half-life after 74 to 77 days.

News decays fastest. Half the pages were gone after 36 days, and 41% were still cited at day 60.

The report notes that pricing pages are classified as landing or product pages, so they sit inside the landing and product groups instead of having a line of their own.

What the half-life number can and cannot say

There is one trap. For landing pages, documentation, comparisons and listicles, the 50% line was crossed only in the last week, and only because the tracking window ended. The report says their real half-life is likely close to or beyond 90 days. We have no figure for it. Anyone who writes "landing pages last 90 days" has misread the chart.

Other limits apply as well. This is one cohort, from one day in July, tracked by one vendor, on one engine. A different start date, with different prompts, might give different curves. A page that stops being cited has not necessarily decayed in value: ChatGPT may have shifted to a different source for that question, and the data does not tell us why.

Also, the report counts a page as "still cited" at a given day. It does not say how often it was cited or in what position.

What to do with the curve

The report's own advice is that landing, documentation and pricing pages are long-term assets, that product pages should not be assumed to stay cited on their own, that comparisons and listicles deserve investment because both keep more than 60% after two months, and that news is for short bursts. Here is how we would put that to work.

1. Set an update cadence by type. Short-lived formats get faster reviews. News and time-bound pages are the first to drop, so treat any citation they earn as a bonus, not an income.
2. Keep product page facts current. Product pages sit near the bottom of the stable group, at about 55% after 60 days. Specs, availability and prices are the things that go stale, and the report suggests you do not assume the page stays cited on its own.
3. Protect the long-lived pages from careless edits. Landing and documentation pages seem to hold their citations for a long time. Before a redesign or a URL change, check which of those pages are cited, and keep the information that was being cited.
4. Publish comparisons and listicles you intend to maintain. Our guide on how often to update a listicle has a practical schedule.
5. Mark a date to look back. After 30 and 60 days, check whether each page you care about is still cited. That is the same checkpoint structure the report uses.

How to measure decay on your own pages

You can do a version of this by hand with a spreadsheet: list your important URLs, run the prompts they should answer, record whether each is cited, and repeat at 30 and 60 days. It works, but it is slow and the sample is thin, because answers vary from run to run.

This is where Promptwatch is the platform we rank first, since the report comes from the same data. Its Citation Trends view gives each page a first cited date, ramp-up days, a peak citation rate and a decay half-life, so you can compare your own pages with the pooled curves above instead of guessing. The Page Tracker lets you follow specific URLs, yours or a competitor's, which is how you test a single page after an edit; our walkthrough on using the Page Tracker to prove a page gets cited shows the setup.

When a page starts to slip, Content Refresh suggests diff-style updates to the existing page. Our post on content refresh for pages AI stopped citing explains the workflow, and Citation Trends and decay half-life covers how to read the curve. Plan-wise, brand plans begin at Essential ($95 a month, 7-day trial on monthly billing), and crawler logs, which help separate a decay problem from an access problem, start on Professional at $245 a month. Essential has none.

The takeaway

Landing pages and documentation held on best in this cohort, comparisons and listicles stayed above 60% after two months, product pages and reviews and how-tos sat around 55%, and news faded fastest. Treat long-lived formats as assets to protect, keep product facts fresh, and check your own pages at 30 and 60 days rather than trusting a pooled curve.