How fast can a new page get cited by AI?

A new page can be cited within a day of publishing, and on one engine most of them are. Semrush published 81 test pages and tracked their citations for 30 days in research released in December 2025, running the associated queries daily and recording when each page appeared as a source.

Google AI Mode moved first. That study found 29 of the 81 pages, 36%, cited the day after publication, rising to 48 pages by day six. ChatGPT search started slower and climbed more steadily: 8 pages on day one, 14 after a week, 28 after two weeks and 34 pages, or 42%, by day 30. Two engines, the same pages, completely different curves.

Speed here is a property of the engine rather than of the page. An engine reading a live search index can consider a page as soon as it is crawled. An engine that leans more on its own retrieval pipeline and reranking takes longer to admit a new document to the candidate set. Neither timescale has anything to do with how good the page is, which is why "when will we see results" cannot be answered without naming the engine.

The same study also showed the early peak falling away, with AI Mode citing 21 pages by day 30 after peaking at 48. Fast in is not the same as staying in, and a report celebrating week-one citations is reporting the easy half. Our note on measuring AI citation decay covers the second half in detail.

Why doesn't the first AI citation stick?

The first citation rarely sticks because answer engines rotate their sources instead of ranking them once and leaving the list alone. Trakkr Research's citation decay study logged 108,650 citation URLs across 10,991 brands and 8 tracked models, running identical prompts daily, and found that 73.5% of citations appeared exactly once and never returned, with brand mentions halving roughly every 31 days.

Consistency across immediate repeat runs is just as low. AirOps, in its research on citations and mentions, reported that only 30% of brands stayed visible from one AI answer to the next, and 20% across five consecutive runs. It also found brands earning both a citation and a mention were 40% more likely to resurface than brands earning a citation alone.

Read together, those numbers redefine what finishing looks like. The goal is not a citation, it is a citation rate: the share of runs across a fixed prompt set in which you are named. A programme that stops at the first screenshot has measured a sample of a rotating system and called it an outcome, which is how teams end up reporting a win in month two and a mystery in month four.

The practical consequence for timelines is that you cannot know whether something worked for at least several weeks. One good answer proves the page can be retrieved. A rising citation rate across eight or twelve weekly runs proves the change held. Anyone promising verified results inside a month is promising a measurement that cannot exist yet.

Why does displacing a competitor take longer than winning a new question?

Displacing a competitor takes longer because incumbency at the domain level is remarkably stable. BrightEdge's 2026 analysis of week-to-week citation changes found that 96.8% of cited domains and 97.2% of mentioned brands did not change at all week over week, and that the top two citation positions held steady 99.4% of the time.

Set that against the 73.5% one-and-done rate the Trakkr study found for individual URLs and the asymmetry becomes clear. Which page gets used churns constantly. Which domains are eligible barely moves. Breaking into an answer where two established sources have held the top slots for months is a different project, on a different timescale, from being the first decent answer to a question nobody has covered properly. Telling the two apart takes repeated runs, which is what the SIGNALS methodology for measuring citation decay between runs is for.

That is the argument for starting narrow. A specific, qualified question with a weak incumbent answer can be won in weeks, and each win adds the cross-domain and on-site presence that makes the harder questions reachable later. Going straight at the category head term means competing with the sources the engines already trust for it, with no measurement to show progress in the meantime.

BrightEdge also found that the movement which does happen is mostly downward: among the roughly 3% of domains that shifted week over week, 87% were declines. Positions are lost more often than they are taken, which means the fastest available gains often come from other people's decay, not from your own force. Our guide to how AI engines choose which sources to cite covers what decides the substitution.

What makes AEO faster or slower for your site?

Four things make AEO faster or slower, and only one of them is the writing. Whether the engines can already crawl and index you decides whether anything you publish is even eligible. Whether you already have index presence decides how quickly a new page is considered. How contested the question is decides whether a good answer is enough. And how much third-party presence you have decides whether you survive the buying questions, which retrieve comparison pages instead of vendor sites.

What you change	How fast it can move	What the evidence says
Crawler access and rendering	Days, once recrawled	A page outside the index cannot be cited at all, so this gates everything else
A new page on an uncontested question	1 to 30 days depending on engine	36% of new pages cited within a day in AI Mode, 42% by day 30 in ChatGPT search (Semrush, 2025)
Rewriting a page that already ranks	Weeks to a readable trend	Citation rate has to be sampled repeatedly, since 73.5% of citations never recur (Trakkr Research)
Taking a slot from an established rival	Months	96.8% of cited domains unchanged week over week; top two positions 99.4% stable (BrightEdge, 2026)
Third-party presence and mentions	Slowest, on other people's schedules	Brands with both citations and mentions are 40% more likely to resurface (AirOps)

Sequence the work by that table, not by what is most interesting to write. Technical access first, because it gates everything. Then the adjacent questions where a good answer can win on merit. Then the contested ones, once you have the presence to compete. Our AI citation checklist orders the page-level items by measured impact.

What should you measure at 30, 60 and 90 days?

Measure something different at each checkpoint, because different things are knowable at each. At 30 days, measure eligibility: are the pages indexed, are the crawlers fetching them, and has any engine cited any of them even once. Those are binary facts you can establish quickly, and a failure here means the work has not started, not that it is failing.

At 60 days, measure citation rate rather than citation. Run the fixed prompt set weekly, count the share of runs in which you are named, and compare that share against the baseline you took before any changes. A rate rising from one run in ten to three in ten is real progress even though seven runs still ignore you, and that is exactly the kind of movement a screenshot cannot show.

At 90 days, measure position against the competitors named alongside you. Citation share tells you whether you are gaining ground or the whole category is rotating, which is the difference between a programme that is working and a market that is churning. This is also the first point at which a 31-day half-life has run its course twice, so a trend that survives to here is more likely to be a trend.

Keep the prompt wording fixed across all three checkpoints, because a rewritten prompt resets the comparison. Our guides to tracking AI visibility across engines and automating the reporting cover the mechanics, and they matter more than usual here: without a stable baseline, every timeline claim afterwards is an anecdote.

Why do the AEO timelines you read online disagree?

The timelines you read online disagree because almost none of them are measurements. Search the question and you will find 30 to 60 days, 90 days, three to six months and six to twelve months asserted with equal confidence, usually with no sample size, no method and no dataset named. They are pricing and expectation-setting devices written for proposals, and they contradict each other because nothing anchors them.

What has actually been measured is the parts, not the whole. Time to first citation has been measured. Citation half-life has been measured. Run-to-run consistency has been measured. The commercial path from a set of page fixes to a sustained flow of enquiries has not been measured in any published study, because it would require tracking many companies through many engines over many months with clean attribution, and nobody has published that work.

Treating the gap honestly is more useful than filling it. The defensible statement is that retrieval responds in days, citation patterns become readable over a quarter, and displacing entrenched sources takes longer than either. What varies most between businesses is the starting point, not the mechanism, so a timeline quoted before anyone has looked at your index presence and your category's incumbents is a number chosen for the conversation rather than for you.

The question to ask any provider, including us, is what they will measure and when they will show you. A programme that names its prompt set, takes a baseline before changing anything, and reports citation rate on a fixed cadence can be held to account at 30, 60 and 90 days. One that quotes a confident timeline and reports screenshots cannot. Our note on how to choose an AEO agency covers the questions that separate the two.

What else do people ask about how long AEO takes?