How Long Does It Take to Get Cited by ChatGPT, Perplexity, and Google AI Overviews After You Publish?
Published: 2026-08-09 · Author: Alex K · GEO & AI Search
Citation timing runs on three separate clocks: indexing, retrieval, and selection. A page can be indexed within hours and still wait weeks for its first citation, because retrieval and citation are distinct steps. The working ranges measured across 2026 studies are: Perplexity, hours to 7 days; ChatGPT, 7 to 21 days; Claude and Google AI Overviews, 14 to 45 days; and citations that depend on third-party mentions, 30 to 90 days. Those are first-touch numbers, not steady state. The median cited page is 14 months old in Google AI Overviews (Digital Applied, 1,000 AI Overviews, 2026) and roughly 500 days old in ChatGPT (Ahrefs, 1.4 million prompts, 2026). A first citation is early access to a slot that older pages usually hold.
Six reader questions on the gap between publishing and getting cited, answered with 2026 study data, plus the refresh-cadence arithmetic most guidance leaves out.
How fast can a brand-new page earn its first AI citation?
Perplexity is the fastest path. It runs a live web search on every query rather than relying on a stored answer, so a newly indexed URL becomes eligible immediately. Perplexity cited content published within the previous 30 days at an 82% rate in a 2026 analysis by Leapd, and new pages can surface within hours of indexing.
ChatGPT is slower for a structural reason: retrieval does not fire on every prompt. The same 2026 analysis found web search triggers on 53.5% of commercial-intent prompts but only 18.7% of informational ones. Ahrefs, studying 1.4 million ChatGPT prompts in 2026, found the model cites roughly 50% of the URLs it does retrieve.
Multiply those two stages and the wait stops looking mysterious. For an informational prompt, the chance that a single query both triggers retrieval and lands your URL in a cited slot is about 0.187 x 0.50, or roughly 9% — before competing against every other candidate page. A page targeting a question asked 20 times a week has a materially different first-citation date than one targeting a question asked 20 times a day, at identical quality. Low query volume reads like slow indexing but is a separate problem.
Submitting new URLs through IndexNow removes the discovery delay, the only part of this chain you directly control. The llms.txt, FAQPage schema, and IndexNow setup covers that step.
Why does the same page appear in Perplexity weeks before ChatGPT?
The two systems source information differently. Perplexity retrieves and reads candidate pages per query. ChatGPT blends training data with selective retrieval, and its training corpus skews old — 29% of ChatGPT citations reference content published in 2022 or earlier, per Leapd's 2026 analysis. A new page competes against that stored material, not just against other live results.
Ahrefs quantified the recency gap across 17 million citations in 2025: AI-cited content is 25.7% fresher than organic Google results, and ChatGPT cites URLs 393 to 458 days newer for the same queries. Fresher than organic search, still measured in years.
| Platform | Typical lag for a new or changed page | Retrieval model |
|---|---|---|
| Perplexity | Hours to 7 days | Live search on every query |
| ChatGPT Search | 7 to 21 days | Selective retrieval plus training data |
| Google AI Overviews | 14 to 45 days | Ranked index, overlaps organic |
| Claude | 14 to 45 days | Selective retrieval |
| Third-party mention effects | 30 to 90 days | Requires external pages to be crawled first |
The practical consequence: judge a new page on Perplexity at day 7 and on ChatGPT at day 21. Declaring failure at day 10 across all engines discards a page that has not finished its cycle.
How long do citations last once they start?
Citations decay on a measurable curve. Half-life estimates published in 2026 put ChatGPT at roughly 3.4 weeks, Google's AI surfaces at 4.3 to 4.8 weeks, and Perplexity at 5.7 to 5.8 weeks. Citation volume for an unedited page typically peaks between day 30 and day 90 and approaches dormancy near the one-year mark.
That sits oddly beside the 14-month median age of cited AI Overview pages until you account for survivorship. Old pages still being cited are overwhelmingly pages that were updated. Age in these studies is measured from the original publish date, while the freshness signal engines read is the modification date. A page first published in 2024 and revised last month counts as 14 months old and as fresh at once.
Ranking and holding a citation are different jobs. Ranking rewards a page that stops changing; citation rewards one that keeps changing. A plan assuming finished content stays finished watches citations slide by half each month while nothing visibly breaks.
How often do you have to refresh a page to hold its citations?
Half-life converts into a workload. To hold a library above half its peak citation volume, each page needs a meaningful update once per half-life:
Weekly refreshes required = library size / platform half-life in weeks
| Library size | ChatGPT (3.4 wk) | Google AI (4.5 wk) | Perplexity (5.7 wk) |
|---|---|---|---|
| 10 pages | 2.9 per week | 2.2 per week | 1.8 per week |
| 25 pages | 7.4 per week | 5.6 per week | 4.4 per week |
| 50 pages | 14.7 per week | 11.1 per week | 8.8 per week |
| 100 pages | 29.4 per week | 22.2 per week | 17.5 per week |
Read the formula backwards and it sets a ceiling: maximum sustainable library = weekly refresh capacity x half-life in weeks. A solo operator who can genuinely refresh five pages a week sustains about 17 pages at ChatGPT's half-life and about 29 at Perplexity's. Publishing the 60th page while the first 40 decay lowers total citations rather than raising them.
Two adjustments raise that ceiling: retire pages that no longer earn citations, and update dateModified whenever the body genuinely changes. The publishing workflow breakdown covers where that fits in a weekly cycle.
Does adding schema or llms.txt shorten the wait?
Evidence splits here, instructively. Ahrefs tracked 1,885 pages that added JSON-LD schema between 2025 and 2026 against roughly 4,000 controls, comparing citations 30 days before and after. Google AI Overviews fell 4.6%, AI Mode rose 2.4%, ChatGPT rose 2.2% — no meaningful uplift.
A separate 2026 study tracking 50,431 citations across 240 pages and six engines over 90 days reported the opposite: deep sameAs schema at +34%, explicit dateModified at +22% overall and +41% on Claude, and llms.txt presence at +11%.
The reconciling detail is the starting population. Ahrefs measured pages already being cited, where markup adds nothing the engine had not resolved. The second study measured a corpus built from a lower base, where markup helps engines identify and date pages they have not yet classified. Schema is a discovery aid, not a ranking multiplier.
For a new site the cheap, high-certainty move is explicit dating — a visible publish date, a real dateModified, and dates that change only when the content does. Both studies agree dating helps; they disagree only about entity markup on established pages.
What is a realistic first 90 days for a new domain?
Concentration is the constraint. Digital Applied's 2026 study of 1,000 AI Overviews found the top 1% of domains capture 47% of all citations, the next 9% take 31%, and the remaining 90% of domains split 22%. A new domain starts in that last bucket regardless of content quality.
Engine choice matters more than effort here. Citation share across six engines in that 2026 tracking study ran ChatGPT Search 36%, Perplexity 26%, Google AI Overviews 17%, Claude 11%, Gemini 7%, and Bing Copilot 3.6%. Perplexity pairs the second-largest share with the shortest lag and strongest recency preference — the only engine where a 30-day-old domain competes on near-even terms. Treat it as the first-90-days scoreboard and AI Overviews as a 6-to-12-month one.
Two page-level factors carried measurable weight in Digital Applied's data: pages over 2,500 words were cited 1.6 times more often than pages under 800 words, and pages carrying named, body-level source citations were cited 2.1 times more often. Depth and sourcing are the levers available before domain authority exists. The eight reasons content gets ignored by AI engines covers the diagnostic side when nothing appears at all, and the GEO Cold Start playbook packages the 90-day sequence with a Claude Code skill.
Frequently Asked Questions
How long should I wait before deciding a page failed?
Twenty-one days for ChatGPT, seven for Perplexity, checked against the prompts you targeted. Before that, there is not enough signal to separate a weak page from an unindexed one.
Does republishing with a new date restart the clock?
Changing a date without changing content produces no durable gain and risks trust signals. Explicit dateModified correlated with a 22% citation lift overall in 2026 testing, but that measured pages whose bodies actually changed.
Why do studies disagree on what share of retrieved pages gets cited?
Ahrefs measured roughly 50% of retrieved URLs cited across 1.4 million prompts in 2026; other 2026 analyses report near 15%. The figures count different denominators — all retrieved URLs versus URLs retrieved per answered query. Both support the same conclusion: retrieval does not guarantee citation.
Is it faster to get cited through Reddit than through my own site?
Often, because those pages are already indexed and heavily crawled, though Ahrefs found Reddit URLs make up 67.8% of ChatGPT's non-cited retrievals — retrieved constantly, cited rarely. The Reddit Marketing Playbook covers placement mechanics.
How large a library should a new site build in its first 90 days?
Smaller than most plans assume. Apply the ceiling formula: weekly refresh capacity multiplied by half-life in weeks. Five refreshes per week supports roughly 17 to 29 pages depending on the target engine.