How Long Does It Take to Get Cited by ChatGPT?

There is no single indexing lag, and that is the honest answer. In the only public controlled test of the question, Google's AI Mode cited 36% of 81 brand-new pages within 24 hours of publication, while ChatGPT Search cited 8% in the same window and did not reach 42% until day 30 (Semrush). Yet the average page an AI assistant cites is roughly 2.9 years old (Ahrefs). Both are true at once, which means the useful question is not "how fast does my page get picked up" but "which clock is my page running against" — and that is set by query type, not by your publishing calendar.

So if you shipped a page last Tuesday and report on Friday, here is what to expect per engine, what to put in the deck, and why a good page often shows a bad number.

The pipeline has four clocks, not one

A page does not move from "published" to "cited" in one step. It passes four gates, each with its own clock, and it can clear all four for a news-shaped query while permanently failing the third for an evergreen one.

Clock one: crawl access. This starts before publish day, not on it. OpenAI's crawler docs are explicit that "it can take ~24 hours from a site's robots.txt update for our systems to adjust" (OpenAI), so a robots.txt fix shipped alongside your page isn't readable the same day — and disallowing OAI-SearchBot means the site "won't appear in ChatGPT search answers" at all. Our walkthrough of llms.txt and AI crawler access in a 20-minute setup covers the retrieval-versus-training bot distinction behind this gate.

Clock two: index or grounding inclusion. Google states the requirement plainly: a page "must be indexed and eligible to be shown in Google Search with a snippet" to appear as a supporting link in AI Overviews or AI Mode — and concedes that "crawling can take anywhere from several days to several months" (Google Search Central). A bot hit is not inclusion either: testing of OpenAI's Web Search API found a cached index kept separate from live fetching, with filtering on what enters it, since "a bot visit does not guarantee inclusion" (LLMrefs). Your log line proving GPTBot came by is evidence of discovery, nothing more.

Clock three: retrieval firing at all. This gate quietly caps everything above it. Semrush's clickstream analysis of over a billion lines of US panel data found ChatGPT enabled web search on 34.5% of queries as of February 2026, down from 46% in late 2024 (Semrush); Profound, across roughly 730,000 conversations, puts it lower still at about 18% triggering at least one search (Profound). If your buyer's prompt is answered from parametric knowledge, publishing speed is irrelevant — there was never a retrieval step to win.

Clock four: selection. Even among retrieved candidates, only a handful get cited: the 1,000-AI-Overview study found an average of 4.2 citations per response (Digital Applied). Four slots, a large eligible pool, a reranker that reshuffles constantly — which is why day-7 absence is rarely a verdict.

Knowing which engine is slow on which clock is what you can actually put in a deck.

Per-engine timelines, with the numbers

The Semrush test is the reference case, and worth stating precisely because most write-ups re-narrate it loosely. Eighty-one FAQ-style pages, published on the Semrush blog (Authority Score 84) in September 2025, tracked daily for 30 days:

Day	Google AI Mode	ChatGPT Search
1	29 pages (36%)	8 pages (10%)
2	36 (44%)	10 (12%)
7	45 (56%)	14 (17%)
14	39 (48%)	28 (35%)
30	21 (26%)	34 (42%)

Two shapes, two reporting strategies. Google surfaces spike early and then rotate — citations peaked at day 7 and fell to 21 pages by day 30, which reads like decay but is better explained as churn. ChatGPT climbs slowly and sticks, roughly doubling between day 7 and day 14 before flattening. Semrush's caveat deserves repeating: these pages had "the best possible conditions to succeed" on an established domain, so a newer site should read the numbers as a ceiling.

Outside that controlled setting the spread is enormous. One practitioner tracking four of his own sites on Bing AI surfaces recorded first citations at 0 days, 3–4 days, 10 days and 64 days — same operator, same engine family (Paul Takisaki). Retrofitting IndexNow to the slowest site moved it from 28.1 citations a day to 30.8, which he declines to call an effect: "four sites, no control group, no p-values." Treat IndexNow as cheap insurance, not a measured lever.

The engines otherwise differ mostly in how much real-time material they lean on. Across 31 million tracked AI citations, the share going to news domains ran from Grok at 23% down to Claude at 5.1%, with ChatGPT at 11.9% (Goodie). Grok is the fast outlier because it draws on two pools: "the live web and the X platform's own post stream" (Contently). Claude sits at the other end, so news-shaped work is the last thing to expect movement on there — our engine-by-engine breakdown of what earns AI citations in 2026 covers how differently the four source answers.

Why the median cited page is still years old

Ahrefs analysed 16.975 million cited URLs across seven platforms and found the average cited page was about 1,064 days old — roughly 2.9 years. AI-cited content was 25.7% fresher than organic Google results, yet AI Overviews cited content 16 days older than organic. A separate Ahrefs analysis of 1.4 million ChatGPT prompts put the median cited page around 500 days (Ahrefs), and the 1,000-AI-Overview study landed independently in the same place: median age 14 months, a blunt "AIOs are NOT recency-biased," and recency mattering only for explicit news intent.

Now the apparent contradiction. Seer Interactive analysed 7,683 pages carrying 47,097 citations across ChatGPT, Gemini and Perplexity from March to June 2026 and found 75% of cited pages had been updated within the last year, 88% within two — Gemini 78%, ChatGPT 73%, Perplexity 65% (Seer Interactive).

These studies do not disagree; they read different dates. Ahrefs measured days since publication, Seer measured last modified. A page first published in 2022 and updated in March 2026 is ~1,400 days old to Ahrefs and "fresh" to Seer. Seer caught this directly: where both dates were readable, 72% of pages looked fresh by last update but only 42% were genuinely published in the past year. Their conclusion is worth pinning above the editorial calendar — "the freshness LLMs reward is being manufactured by updates, not by new publishing."

The clock the engines respond to, then, is the update clock. If you need visibility to move this quarter, refreshing a two-year-old page that already has authority beats shipping something new and waiting out ChatGPT's 30-day curve. Don't panic-refresh everything, though: Seer found pages cited consistently across all four months had a median update age of 0.47 years — older than pages that produced one-month spikes at 0.16 years. Durable citation and freshness spikes are different games.

Set your cadence by query type, not by the calendar

If freshness is decisive on some queries and nearly irrelevant on others, one site-wide cadence is guaranteed to be wrong somewhere. Seer's freshness rates by content type map onto how often each asset needs touching:

Page type	Cited pages updated ≤1 year	Practical cadence
Marketplace / listing	78%	Continuous — staleness is the product
Comparison / review	77%	Quarterly, plus on any competitor pricing change
Reference / documentation	74%	Semi-annual, but only with real changes
Brand / corporate	72%	Semi-annual
Blogs & guides	67%	Annual refresh of the ones already earning citations
News / editorial	45%	Lowest bar — these live or die on day one

Read the news row carefully, because it inverts the intuition. Only 45% of cited news pages were updated within a year — news gets cited in its first days and is then never refreshed. That is not permission to publish news slowly; it means the freshness advantage there is spent immediately. Ahrefs found the same split from the other side: within news categories, younger content won when relevance tied, while in the general search results channel — 88.46% of ChatGPT's citations — established older pages outperformed fresher ones.

And the freshness boost itself is not a constant. A controlled experiment injecting artificial publication dates into TREC passages found all seven tested LLMs promoted the apparently-newer text, shifting the mean publication year of the top 10 forward by up to 4.78 years (arXiv). Much of that preference lives in the reranker, not the index — so a frontier model swap can change what a fresh date is worth. Your page's timeline is not a property of your page alone.

What belongs in the deck at day 7, 14, 30 and 90

• Day −1: confirm access, not content. OAI-SearchBot, GPTBot, ClaudeBot, PerplexityBot and Googlebot all allowed. Since OpenAI takes ~24 hours to re-read robots.txt, this cannot be a publish-day fix.
• Day 0: ping IndexNow, submit in Search Console and Bing Webmaster Tools — and take a baseline reading, because you cannot detect a lift you never measured a "before" for.
• Days 1–7: measure crawl, not citation. Server logs are the only honest early signal. In one 48-day log study, ChatGPT-User referrals rose fivefold before GPTBot ever appeared — the answer bots move first (Wislr).
• Day 7: Google surfaces are the only realistic early read. If nothing has happened there, debug indexation, not the copy.
• Day 14: the earliest defensible ChatGPT read. ChatGPT sat at 17% on day 7 and 35% on day 14, so "no ChatGPT movement" before day 14 is reporting noise.
• Day 30: the honest checkpoint, with an asterisk. Fair for ChatGPT, where the curve flattens. On Google surfaces use a rolling average, or a point-in-time count will understate what you earned — as it did when AI Mode fell from 45 pages to 21.
• Day 90: the update-clock decision. By now you know whether the page joined the stable core or the rotating tail, and whether it deserves a refresh or a rewrite.

One free input for that stack: Bing Webmaster Tools' AI Performance report (public preview since February 2026) shows total citations, cited pages per day and the "grounding queries" used to retrieve your content (Bing Webmaster Blog).

When "no movement" isn't your page's fault

Two things routinely make a good page look like a failure.

The engine changes who it credits, constantly. Across 43,000+ keywords, AI Overviews had a 70% chance of changing between consecutive observations, with 45.5% of cited URLs swapping out while the answer stayed nearly identical at 0.95 cosine similarity (Ahrefs). Engines change their sources far more often than what they say. SISTRIX, tracking 82,619 prompts over 17 weeks, measured weekly domain churn-in of 5% on AI Overviews, 56% on AI Mode and 74% on ChatGPT Search, with "no stabilisation, no levelling off" (SISTRIX).

Sometimes the whole engine moves. seoClarity tracked ChatGPT citation volumes across five markets and recorded a fall of over 90% at the March–April 2026 trough, with the US zero-citation rate doubling from 28% to 48% in March before rebounding in May (seoClarity). A page shipped that March would have looked like a failure for reasons that had nothing to do with it. One engine read in isolation can't tell those apart; four read side by side can — the same logic behind our guide to separating a model-version artefact from a real visibility drop.

The measurement discipline that makes any of this reportable

All of the above collapses if your evidence is a single scan. Repeated-sampling research across Perplexity, SearchGPT and Gemini found run-to-run citation-set overlap (Jaccard) medians of 0.29–0.31 for Gemini, 0.33–0.40 for SearchGPT and 0.50 for Perplexity, concluding that "single-run measurements of domain visibility in generative search cannot be interpreted as reliable estimates of an underlying true citation rate" (arXiv). The same work sizes a defensible sample: for a citation-share confidence interval 0.05 wide, roughly 30 queries on Gemini, 90–100 on Perplexity, 150+ on SearchGPT. A parallel paper reaches the same conclusion via bootstrap variance (arXiv).

The arithmetic is what makes it land. Say your page's true citation probability on a prompt is 20%. One check misses it 80% of the time. Five checks still miss it about a third of the time (0.8⁵ ≈ 0.33). Ten miss it roughly 11%. "We shipped it, we checked, nothing" at day 7 is not evidence — it is one draw from a noisy distribution.

That is why Geoptimizer reports its headline score as a 7-day rolling window with a confidence band and labels single on-demand scans as snapshots: a number with an interval survives contact with a nondeterministic engine, and a screenshot doesn't. Running one prompt set across ChatGPT, Gemini, Claude and Grok on a fixed cadence is also what lets you say "ChatGPT moved, Gemini didn't, so it isn't the page." The prompt set matters as much as the cadence — start with how to choose the 25 prompts worth tracking.

FAQ

How long does it take for ChatGPT to cite a new page? In the only public controlled test, ChatGPT Search cited 10% of new pages within 24 hours, 17% by day 7, 35% by day 14 and 42% by day 30. Day 14 is the earliest defensible read, and the curve flattens near day 30. Those figures come from a domain with Authority Score 84, so treat them as a best case.

Why isn't my page cited even though GPTBot crawled it? A crawl is discovery, not inclusion. OpenAI keeps a cached index separate from live fetching and filters what enters it, so a bot visit doesn't guarantee your page is retrievable. Google's requirement is stricter and stated outright: a page must be indexed and snippet-eligible in Search to appear as a supporting link in AI Overviews or AI Mode.

Does updating an old page work better than publishing a new one? Often, yes. 75% of AI-cited pages were updated within the last year, but only 42% were actually first published within it — the freshness engines reward comes largely from updates to established pages. Refreshes with no substantive change are opportunity cost rather than a lever.

Why does my citation count change between scans? Because engines rotate sources far more than they rotate answers. 45.5% of the URLs an AI Overview cites change between consecutive responses while the answer stays 0.95 semantically identical, and weekly domain churn on ChatGPT Search runs about 74%. Report a rolling window with a confidence band, not a point-in-time count.

The short version

There is no universal indexing lag — only four clocks running at different speeds per engine, and a selection step that reshuffles constantly. Google's AI surfaces can move within a day, ChatGPT realistically needs two to four weeks, and Grok can be near-real-time on anything the X stream touches. Underneath all of it, the median cited page is over a year old, because engines respond to the update clock more than the publish clock.

The one thing you can't fix retroactively is a missing baseline. Take a reading across all four engines before your next page ships — Geoptimizer's free plan covers ChatGPT, Gemini, Claude and Grok — so that day 30 has something to be compared against.