How long do AI assistants keep citing an article, and when should you update or replace older content?

The short answer: there is no fixed lifespan

Nobody can honestly tell you that ChatGPT keeps citing an article for six weeks or Perplexity drops it after a quarter. Citation behavior depends on the model, the prompt, the country, the competing pages and whether the assistant fetches live results or answers from memory. Any single number you read in a blog post without a method behind it is a guess dressed up as a statistic. We are not going to give you one either.

What you can do is measure your own pages. Citations are not a permanent badge. They behave more like rankings: a page earns them, holds them for a while, and then loses them to something newer, clearer or better sourced. The useful question is not "how long do citations last" but "how fast does this page's citation rate decay, and what should I do when it does".

What a citation lifecycle looks like

When you track a page over time, most citation histories have four visible stages. The labels below are ours, not an industry standard.

First cited. The date an assistant first named the page as a source for a tracked prompt.
Ramp-up. The days between publishing (or first citation) and the page reaching its normal citation rate.
Peak. The highest share of tracked responses that cite the page.
Decay. The slow or fast slide after the peak, usually summarized as a half-life: how long until the rate halves.

Once you see those four numbers per page, the update-or-replace decision stops being a hunch. A page with a long half-life and a stable rate is doing its job. A page that peaked fast and decayed in weeks is telling you the topic is time-sensitive, or that a competitor published something stronger.

Why citations fade

Assistants that search the web pick sources again on every answer. So a citation fades for ordinary reasons:

A competing page covers the same question more directly, with fresher facts.
Your page contains claims that are now out of date, and the assistant's answer has moved on.
The prompt itself changed. Users and the model's query fan-outs drift toward different wording.
The crawler can no longer fetch the page cleanly, because of a robots rule, a CDN block or errors.
Third-party pages that mention your brand changed, so the assistant has other sources to lean on.

Only the first three are content problems. The fourth is a technical one, and it is the one teams most often miss. If GPTBot, OAI-SearchBot or PerplexityBot are hitting errors on a page, refreshing the copy will change nothing.

When to update, merge or replace

Here is the decision rule we would apply to any older article, based on what its own citation data shows.

Signal on the page	Likely cause	Action
Still cited, rate sliding slowly	Facts aging, competitors catching up	Update: refresh data, add current examples, tighten the answer
Never ramped up after publishing	Wrong topic, wrong wording, or crawl problems	Check crawler logs first, then rewrite around the real prompt
Peaked fast, decayed fast	Time-sensitive topic	Schedule recurring updates, or accept a short life
Cited, but the answer it supports is wrong now	Outdated claims	Update immediately, even if the rate looks fine
Two of your pages split the same prompts	Cannibalization	Merge into the stronger URL and redirect
Topic no longer matches how buyers ask	Intent shift	Replace with a new page, keep the old URL redirected

Updating is almost always cheaper than replacing, because an existing URL already carries links and crawl history. Replace only when the page's job has changed.

How to measure it in practice

You need three kinds of data joined together: whether the page is cited, whether the bots are fetching it, and whether any traffic arrives. Most tools in this category only cover the first. A tracker answers "was the brand mentioned this week", which tells you very little about decay.

Promptwatch is the platform we would start with for this job because the pieces are in one workspace:

Citation Trends classifies every citation by content type and source type and reports citation rate over time, plus first cited, ramp-up days, peak rate and decay half-life for each page. That is the lifecycle above, as columns.
Page Tracker lets you follow specific URLs, yours or a competitor's, so you can see when a rival page takes over a prompt.
Agent Analytics shows IP-verified AI crawler logs, a crawl-to-citation path with a citation rate per page, and error tracking with failure explanations. If a page stopped being fetched, this is where you see it.
Content Refresh produces diff-style update suggestions for existing pages, and content gap analysis scores coverage against the prompts you track.
Action Board turns the findings into a prioritized to-do list, with a Discovery agent that investigates weekly and suggests up to five evidence-backed actions per run.
A/B period comparison in the dashboards lets you check whether an update actually moved the citation rate.

Monitors run prompt panels daily per model, country and language, so the time series is dense enough to show a trend rather than a snapshot. Paid plans start at Essential ($95/month); crawler logs start on Professional ($245/month), and that matters here, because the crawl side of the diagnosis needs them.

How other tools handle this

Fairness matters, so here is what the alternatives offer for this specific question, using the data in our catalog.

AthenaHQ tracks how a brand appears in AI answers and turns gaps into actions through an Action Center and a content optimization agent. Starter is $295 a month on credits, where one credit is one AI response. It is a reasonable pick if the action queue matters more to you than crawler data, but its catalog entry lists no AI crawler-log product, so you cannot see the fetch side of a decaying page.

Scrunch AI includes AI agent and bot traffic reporting and sells an Agent Experience Platform that serves AI-optimized pages to crawlers. Core is $250 a month. It is stronger on the crawler side than a pure tracker, though its page-level citation history is not documented in our listing at the depth described above.

Profound sells enterprise marketing agents with Answer Engine Insights and Agent Analytics alongside. Its Enterprise plan is custom and, per our existing entry, starts around $40,000 a year, which is a different buying decision from a self-serve platform.

Pure monitoring trackers can tell you a mention happened. They rarely tell you when a page started to lose, why, and what to change.

A practical refresh routine

Pick the 20 pages that matter most to revenue, not all of them.
Record first cited, peak and current citation rate for each.
Check crawler logs for those URLs before touching any copy.
For pages with a sliding rate, update the facts, the opening answer and the sources. Note the date you published the change.
Re-measure after a couple of weeks with a period comparison, and only then decide whether to go further.
Retire or merge pages that never ramped up after two honest attempts.

Who should pick which

You want the full lifecycle view (citations, crawler fetches, traffic, refresh suggestions) in one place: Promptwatch.
You want an action queue and can live without crawler logs: AthenaHQ.
You care about crawler-side serving and bot reporting: Scrunch AI.
You have an enterprise marketing budget and want agents beyond AI search: Profound.

One caveat applies to every option. Tracking cannot guarantee that an updated page will be cited. It tells you what changed and gives you something concrete to fix.