TL;DR
- Median time-to-first-citation in ChatGPT for an optimized page sits around 7 days; the 90th percentile is roughly 37 days.
- ChatGPT Search is meaningfully slower than Google AI Mode at picking up brand-new content — about 8% of pages cited in the first 24 hours vs. 36% for AI Mode in one controlled test.
- Once ChatGPT does cite a page, the citation tends to be sticky rather than oscillate day to day.
- New domains are the hard case: most of the timeline is governed by domain authority and chunk-level citability, not by crawl frequency.
- If you cross day 37 without a citation, treat it as a diagnostic signal — usually a structural or trust problem, not a patience problem.
ChatGPT's citation behavior is driven by two different systems: the static base model (trained on a fixed snapshot) and the RAG-based browsing layer that retrieves live results at query time. For practical GEO work, the question "how long until ChatGPT cites me?" almost always means the second system. The answer depends on your domain's existing reputation, how citable each page is at the paragraph level, and whether the engine's retrieval index already trusts sources in your topic cluster.
What the benchmarks actually say
The cleanest published numbers come from Josh Blyskal at Profound, summarized by RedPinGeek. Across roughly 900 newly published marketing pages tracked against billions of LLM response logs, the median time to first citation was 6.81 days and the 90th percentile was 37.10 days. That means half of well-optimized pages earn their first ChatGPT citation inside a week, and nine out of ten get there inside about five weeks.
Semrush's controlled experiment gives a less flattering view for fresh content. Publishing 81 net-new pages and tracking them with their Enterprise AIO tool, only 8% were cited by ChatGPT Search in the first 24 hours, compared to 36% for Google AI Mode. By day 30, ChatGPT Search had cited about 42% of test pages — and a large share were never cited at all in the observation window. The encouraging finding: when ChatGPT did cite a page, the citation tended to persist rather than flicker.
The gap between the two datasets is mostly explained by selection. Profound's median measures pages that eventually get cited and were already optimized for AI retrieval. Semrush's 42% measures the base rate across an arbitrary batch. Use Profound's number for planning when you've done the work, and Semrush's for setting expectations when you haven't.
RAG vs. training-cycle timelines
AuthorityTech draws the distinction that matters most for new domains: RAG-based responses vs. training-based responses. For ChatGPT with browsing enabled — which is what most users hit by default in 2025 — a Tier-1 earned-media placement typically shows up in citations within a 3–14 day window after it goes live. For the base model without web access, you're effectively waiting for the next training cut, which historically runs 3–18 months out.
In practice this means a brand-new domain has two parallel clocks running. The RAG clock is short and responsive: get mentioned on a source ChatGPT already trusts, and you can appear in retrieved citations within days. The training clock is long and structural: only sustained presence across the web pushes you into the model's parametric knowledge, which is what powers offline answers and brand recognition in zero-search responses.
Why new domains are the hard case
The timelines above assume an existing domain publishing new pages. A brand-new domain inverts the difficulty profile. Winston Digital makes the point bluntly: AI engines cite chunks, not pages, and the two dominant inputs are source domain authority and chunk-level citability. A new domain has zero of the first variable, so even excellent chunks struggle to clear the trust threshold for retrieval.
The piece argues that an old domain with weak content is a stronger starting position than a new domain with great content, because "old domain + new high-quality content" compounds faster. For a literally new domain, the realistic playbook is:
- Earn placements on domains ChatGPT already trusts in your topic cluster. Reverse-engineer those by inspecting current citations for your seed queries.
- Treat the first 60–90 days as a trust-building window, not a citation window. Your own pages are unlikely to clear the bar; the goal is to be the entity ChatGPT cites through a trusted source.
- Ship citable chunks anyway — self-contained paragraphs under explicit headings, with the answer in the first sentence — so that when domain trust does kick in, the retrieval layer has something to grab.
What accelerates the timeline
A few levers measurably compress the wait:
- Tier-1 earned media within the first 30 days. A single placement on a source ChatGPT already retrieves can pull you into citations within the 3–14 day RAG window.
- Topic concentration. Publishing 10 pages on one narrow cluster beats 10 pages across unrelated topics — retrieval rewards depth on a specific query intent.
- Chunk structure. H2 question, one-sentence direct answer, supporting evidence, source link. This is the shape retrieval prefers.
- Schema and
llms.txt. Neither is magic, but both reduce the parsing tax on the retrieval layer. - Sticky URLs. Don't change slugs during the first 90 days. Re-indexing penalties are real.
What does not work: publishing volume for its own sake, keyword stuffing for AI ("optimized for ChatGPT" copy), or waiting passively for a new domain to age into trust without earned mentions.
How to set expectations internally
For an established domain with reasonable authority: plan for the first cited page within 1–2 weeks, with a fair share of pages never cited at all. Use day 37 as the diagnostic line — past that, audit the page rather than wait.
For a new domain: assume 60–120 days before your own URLs appear meaningfully in ChatGPT citations, and front-load the work on earned mentions during that window. Track brand-name retrieval separately from URL citations; brand mentions in answers usually precede URL citations by several weeks.
FAQ
Does ChatGPT cite pages faster if I submit them somewhere?
No. There's no submission endpoint for ChatGPT Search. Citations happen through whatever retrieval index OpenAI is using at query time, which is updated by their crawlers and partner sources on their own schedule. Earned media on already-indexed sites is the closest thing to a "submission."
Will a new domain ever appear in ChatGPT's base model answers?
Eventually, yes, but only after sustained web presence across the next training cut — typically a 3–18 month horizon per AuthorityTech. For new domains, optimize for the RAG/browsing layer first; the training layer is a downstream effect of being talked about.
Is 30 days enough time to judge whether GEO is working?
For individual pages on an established domain, yes — 30 days covers most of the citation distribution. For a new domain, no. Judge new-domain GEO on a 90-day window using brand mentions and earned placements as leading indicators, not own-URL citations.
