Ramp-up, peak, decay: reading the citation lifecycle of a page in AI search (2026)

Every page AI cites rises, peaks, and fades. Learn how to read first-cited date, ramp-up, peak, and half-life, what 2026 data says about how long citations last, and how to turn the curve into a refresh routine.

Key takeaways

  • An AI citation is not a position you hold. Profound's September 2026 study of 1.19 million page and engine lifecycles found the median page loses half its citation share 11 days after peaking.
  • About 78% of cited pages are single bursts. The other 22% are sustained pages with a median half-life of 27 to 29 days.
  • Engines do not decay in sync (cross-engine correlation is 0.03 to 0.09), so you need curves per page and per engine.
  • Content type changes the curve. In Promptwatch Data, landing pages and docs stay cited longest in ChatGPT, while news loses half its pages within 36 days.
  • Use the curve as a trigger for refresh work, not a calendar. A page falling toward half its peak with no competitor change is the one to update this week.

Why the lifecycle matters

In classic SEO you win a ranking and mostly hold it. Links pile up over months, a domain earns trust over years. A lot of teams carry that mental model straight into AI search, and it quietly misleads them.

The 2026 data says authority in AI answers expires more than it accumulates. A page gets picked up, gets cited heavily for a short stretch, and then the engine rotates to other sources. The answer to the question stays roughly the same. The sources underneath it change.

If you only look at a total citation count, you can't tell these stories apart:

  • A page that collected all twelve of its citations in one week and went quiet.
  • A page still climbing after a month.
  • A page that was cited constantly last quarter and rarely now.

Those need three different decisions. A count can't separate them. A lifecycle view can.

The four numbers that describe a citation curve

Plot citations of one URL on one engine over time and you usually get the same shape. Four measures describe it.

MeasureWhat it meansWhat it tells you
First cited dateThe first day an engine used the page as a sourceWhen the page entered the pool of candidates
Ramp-up (rise)Days from first citation to peakHow fast the page caught on
PeakThe highest point of the smoothed citation shareThe most attention the page has earned
Half-lifeDays from peak until share falls to half and stays thereHow quickly the page is losing its slot

A short ramp-up means the page caught on fast. A long one means it built slowly, which is often a healthier sign for evergreen material. A page that rose for 45 days and then had a 10-day half-life climbed steadily and then shed relevance quickly. That's a specific story you can act on.

Profound's definition is worth knowing because it's strict. A page counts as decayed only when its smoothed share falls to or below half its peak and stays there for 14 straight days, with no bounce back above the line. Profound also only scores pages that have been live at least 28 days, have at least 5 citations in total, and reached a real peak. Pages that are too new or too rarely cited are left out of the averages.

What the 2026 research says

Most citations are single events

Profound measured 883,000 pages across ChatGPT, Perplexity, Claude, Google AI Overviews, Google AI Mode, Gemini, and Copilot over twelve months. Each page counted separately per engine, giving 1.19 million lifecycles. The median page lost half its citation share within 11 days of peaking, and the median gap from first citation to half-peak was 42 days.

The split underneath is the useful part. Of 1.11 million pairings observed for at least two weeks past peak, 78% had a half-life of 14 days or less. Those are burst pages. The remaining 22% are sustained pages, with a median half-life of 27 to 29 days depending on the engine.

Profound's study on the 11-day half-life of an AI citation

Engines decide independently

This one changes how you should monitor. Among pages that last beyond two weeks, a Claude citation was 2.8 times as likely as a ChatGPT citation to still be relevant eight weeks later. And how long a page lasts on one engine tells you almost nothing about another: correlations run 0.03 to 0.09, effectively zero. Only Google's three surfaces move together a little, at 0.12 to 0.17.

So losing a ChatGPT citation is not a reason to panic about Perplexity. It's also not a reason to relax. You need per-engine curves.

Breadth predicts lasting

Pages cited by five or more of the seven engines were 5.5 times as likely as single-engine pages to still sit above half their peak share after eight weeks. Kollavo's summary of the same data puts it at 6.8% versus 1.2%, with 2.0% for two engines and 3.5% for three to four. Those are low numbers in absolute terms, which is a sobering reminder of how rare durability is.

Date-tied content fades first

In Profound's data, report-style pages (news, announcements, research reports) had the shortest life: a 27-day median half-life among sustained pages and 7.0% still above half-peak at day 56. Comparison and listicle pages held longest, at 29 days and 11.2%.

Recency is doing some of the work

Quattr's framing is that retrieval systems re-rank on every query and treat recency as a cheap proxy for accuracy, so a fresher page tends to displace an otherwise equal older one. Discovered Labs found that Claude shows the steepest decay curve, with 60% of its citations on content under six months old. Treat vendor-specific timelines like Quattr's 30 to 90 day peak window as rough guides rather than constants. They come from different methods than Profound's.

Two different half-life metrics, don't mix them

This trips people up. Profound measures decay from each page's peak share, per engine. The median there is 11 days. Promptwatch Data measures something else: the share of a day-0 cohort that ChatGPT still cites at all. Those are survival rates, and they run much longer.

Neither is wrong. They answer different questions. Never put an 11-day peak-share half-life next to a 74-day survival half-life and call it a contradiction.

Half-life by content type in ChatGPT

Promptwatch tracked every classified page ChatGPT cited on July 6, 2026 for 90 days. Here's what Promptwatch Data's content half-life report found about the share still cited later:

Content typeStill cited at 30 daysStill cited at 60 daysHalf-life notes
Landing pages78%69%Crosses 50% only in the final week; true half-life likely 90+ days
Docs76%67%93% were cited again after day 0
Comparisons72%63%Crosses 50% only in the final week
Listiclesn/a60%Better than product pages
Product pages, reviews, how-tosn/aabout 55%Half-life 74 to 77 days
Newsn/a41%Half gone by day 36; almost a quarter never cited again after day 0

There's a window artifact to keep in mind. Every line drops in the last week because fewer days remain for a page to be cited again, so the longest-lived types are probably more durable than the chart shows.

What to do with this:

  • Treat landing pages, docs, and pricing pages as long-term citation assets. Keep prices, features, and positioning accurate.
  • Don't assume product pages stay cited on their own. Specs, availability, and prices go stale.
  • Put lasting effort into comparison pages and listicles.
  • Use news and press for short bursts of visibility, and plan on them fading.

For context on what ChatGPT cites in the first place, Promptwatch's citation share by content type data shows product pages at 22.4% of citations (Sep 28 to Oct 4, 2026), followed by landing pages and listicles.

How to read each stage

Ramp-up

A page is discovered, crawled, and tested against more queries. During ramp-up the question is access and fit. If a page never gets going, check whether AI crawlers can reach it and whether it answers an actual prompt. Crawler logs help here, since a page that bots never read can't climb. Promptwatch's AI crawler logs and the free Agentic Readiness Check are two ways to separate an access problem from a content problem.

Peak

The peak is the most share the page will ever have. Note it, then note what the page looked like then. Which prompts did it win? Which competitors were absent? That snapshot is your protect list. Don't tinker with a page at peak unless it's wrong, but do record what made it work.

Decay

Decay is where the decisions are. Three patterns need different responses:

  1. Fast decay on one engine only. Probably that engine rotating sources. Watch, don't rewrite.
  2. Decay across engines with a competitor now cited in your slot. Compare their page to yours, then refresh.
  3. Decay with nothing changing around you and nothing changed on your page. This is the signal to act. Update the facts, dates, and examples, and add what has changed in the topic.

Dormancy

By six months to a year without updates, content can effectively vanish from AI answers, according to Quattr. Can it come back? Yes, pages can regain citations after a real update. A cosmetic date change is not a real update.

Turn the curve into a monthly routine

Here's a routine small teams can keep up:

  1. Pull the pages that have passed peak and are heading toward half. Sort by how much they were worth, not by publish date.
  2. Pick one refresh action: the page with the sharpest fall that you still own the best answer for.
  3. Pick one protect action: a long-lived page (docs, landing, pricing) and verify every fact on it.
  4. Check the next month whether the refreshed page's curve bent. If it didn't, the page may need a different angle, not more words.
  5. Once a quarter, look at breadth. Pages cited by several engines last longer, so ask why your best ones are cited by one.

The bigger point from Profound's team is that refresh cadence usually runs on a calendar or a publish date, so the page that actually needed work sits for another quarter. Let decay trigger the work instead.

Tools that expose the lifecycle

Not every AI visibility tool shows timing. Plenty only count mentions. These are the ones that surface the curve itself.

ToolLifecycle viewNotes
PromptwatchCitation Trends: first cited date, ramp-up, peak, decay half-life per cited page; 22 content types and 20 source typesAlso covers crawler logs, page tracking, and a Content Agent for refreshes
ProfoundCitation Decay: first cited, rise, peak, half-life per URL, filterable by model and owned domainsStrict scoring bars; enterprise focus
QuattrCross-platform GEO dashboard with decay framingEmphasizes recency and per-platform half-life

Promptwatch's Citation Trends classifies every citation and follows each cited page through ramp-up, peak, and decay, then connects that to the rest of the loop: page tracking, crawler logs, the Action Board, and the Content Agent, which can draft and publish refreshes to WordPress, Webflow, or Framer. That matters because seeing decay is only half the job. Teams that only monitor end up with a dashboard full of falling lines and no time to fix them.

Favicon of Promptwatch

Promptwatch

Track and optimize your brand visibility in AI search engines
View more
Screenshot of Promptwatch website

Profound's version is a solid fit for enterprise teams that already live in its platform.

Favicon of Profound

Profound

Enterprise AI visibility platform tracking brand mentions across ChatGPT, Perplexity, and 9+ AI search engines
View more
Screenshot of Profound website

Quattr covers the same ground from an enterprise SEO angle.

Favicon of Quattr

Quattr

AI search presence optimization platform
View more
Screenshot of Quattr website

If you want to compare more options, the GEO software directory at bestgeosoftware.com lists dozens of trackers, and Promptwatch has a side-by-side comparison of GEO platforms. For a head-to-head, see Promptwatch vs Profound.

Pitfalls when reading curves

  • Comparing vendors' half-lives as if they were one metric. Check how each defines the peak, the smoothing, and what counts as decayed.
  • Judging pages that are too young. Profound requires 28 days live and at least 5 citations. Do the same, or you'll chase noise.
  • Reading a single engine as the whole market. With correlations near zero, one engine's decline says little about the others.
  • Ignoring prompt changes. A falling curve can mean the prompt set moved, not that your page got worse. ChatGPT's own behavior shifts too: Promptwatch Data recorded a jump in site: operator queries on August 8, 2026, and a drop in Reddit citations on August 14.
  • Chasing every burst. If 78% of pairings are bursts, most decay is normal. Save your effort for pages that should have been sustained.

A simple way to start this week

Start with ten pages that matter commercially. Write down the first cited date, the peak date, and today's share on each engine you care about. Mark which are landing pages, docs, comparisons, or news. The news pages you can leave alone. The landing and docs pages that are sliding are your first refresh queue. The free AI visibility report is a quick way to see where you stand before you set up anything heavier.

The lifecycle idea sounds abstract until you see your own best page fall to half in two weeks. After that, you stop treating citations like rankings and start treating them like a rhythm you have to keep up with.

Share:

© 2026 Surferstack · Find the best Marketing tools for your GTM motion · RSS

Surferstack is an affiliate review site. When you click links to vendors or buy through links on our site, we may earn an affiliate commission at no extra cost to you.

Surferstack is a review website based on user reviews on Reddit and G2, and on publicly available information. We keep everything as up to date as possible, but pricing and features can change. Always confirm the details with the vendor before purchasing.