Profound reported a median 6.81 days to first citation and a 90th-percentile time of 37 days in its study of roughly 900 marketing pages. Those figures describe that study’s observed citation timing, not a guarantee that 90% of every new page cohort will be cited within 37 days. First citation and continued citation are separate outcomes.
What does the AI visibility timeline actually measure?
AI visibility changes at different speeds across tools. A new page can appear in search before it appears in an AI answer, so track each platform separately.
Three timelines stack on top of each other. The crawl timeline is when user agents such as OAI-SearchBot, GPTBot, PerplexityBot, and Googlebot can fetch your page. The index timeline is when that crawl turns into a queryable record. The citation timeline is when an actual user prompt retrieves and surfaces your page in an answer. Most teams ask "how long does AI visibility take" expecting one number, and instead need to plan for three. Parse tracks AI visibility across ChatGPT and Google AI Mode, so the timeline data in this post reflects citation behavior across all three.
The benchmark: 6.81 days to first citation, 37 days to 90%
The cleanest data point comes from Josh Blyskal at Profound, who analyzed roughly 900 newly published marketing pages tracked across billions of LLM response logs. Half of those pages picked up their first AI citation within 7 days. Ninety percent were cited by day 37. That timing describes the observed study sample. Passing day 37 without a citation does not by itself diagnose a technical or structural problem.
Median time to first AI citation across ~900 marketing pages (Profound)
Pages picking up their first citation within 7 days (Profound)
Pages cited within 37 days (Profound)
Semrush test pages tracked for 30 days across Google AI Mode and ChatGPT search
The day-37 threshold matters for budget conversations. It tells you when to escalate from "wait" to "diagnose." Common technical reasons content fails to cite past that window include robots.txt blocks for OAI-SearchBot or PerplexityBot, missing or stale Bing indexing, weak entity coherence between the page and the brand, or a structural answer pattern that AI retrievers cannot extract cleanly.
Why the platforms move at different speeds
Each AI surface has a different retrieval pipeline. Google AI Mode and AI Overviews ride on Google's search systems and use query fan-out to issue related searches across subtopics and data sources. ChatGPT search depends on OpenAI's OAI-SearchBot and broader search infrastructure, while Perplexity describes PerplexityBot as the crawler that controls whether pages can be indexed for its search surface. Treat those as separate clocks instead of one "AI search" clock.
Semrush ran a 30-day citation experiment tracking 81 newly published FAQ-style pages. Within 24 hours, 29 pages were cited in Google AI Mode, while ChatGPT search cited eight pages in the same window. The pattern reverses on the back end: ChatGPT keeps a steady upward curve through day 30, while Google AI Mode's citations swing hard.
| Platform | First citation typical window | 24-hour citation rate | Day-30 retention |
|---|---|---|---|
| Google AI Mode | Hours to 1 day on a strong domain | 29 of 81 pages (Semrush) | Volatile; 21 pages cited on day 30 |
| ChatGPT search | Days to weeks | 8 of 81 pages (Semrush) | Steady; 34 pages cited on day 30 |
| Google AI Overviews | Depends on Google Search retrieval and fan-out | Not isolated in the Semrush test | Track separately from AI Mode |
| Perplexity | Depends on PerplexityBot indexing and real-time search retrieval | Not isolated in the Semrush test | Track separately from Google and ChatGPT |
| Training-data answers | Model release cycles | Not citation-based | Separate from web retrieval |
Use this table to set per-platform expectations with your CEO instead of one rolled-up answer. If your goal is fast presence in Google AI Mode, you optimize for Google's existing ranking signals. If your goal is ChatGPT citation, you accept a slower onset and plan for compounding.
If you want to know when AI changes its answer about your brand, start with a free brand check. It takes a minute.
The training-data line: the slow lane no one talks about
Training data citations are a separate timeline and a separate strategy. When ChatGPT answers a prompt with no web search, it leans on the corpus baked into the model at training time. That corpus was frozen months before the model shipped. New content cannot enter until the next training cut, which means anything you publish today might not influence parametric memory for six to eighteen months.
This matters because most teams confuse the two. A page can be cited in ChatGPT search within a week and still be invisible when the user has web search turned off. The fix is to stop treating "ChatGPT visibility" as one channel. Track retrieval-grounded answers and memory-grounded answers separately. The retrieval timeline rewards velocity, technical hygiene, and freshness. The memory timeline rewards entity coherence across the open web and time spent at the top of high-authority corpora like Wikipedia, Wikidata, and major news outlets.
The cold-start timeline for a brand starting from zero
The 6.81 day median assumes a publisher that already has crawler access, an indexed domain, and some entity authority. New brands and new domains face a longer first leg. The pattern we see in Parse's brand index is roughly this: weeks one and two are crawl and index, weeks three through six are first inconsistent citations on long-tail prompts, and weeks eight through twelve are when category and competitor-set prompts start surfacing the brand at all.
Do not apply the Profound benchmark to a brand that has no entity footprint yet. The study measured newly published pages, not the full job of making a company recognizable to retrieval systems. A new brand needs crawler access, index coverage, entity consistency, and repeated prompt sampling before the citation clock is meaningful. For the operating cadence behind that ramp, see the first 90 days of AI visibility.
A CEO who reads "6.81 days" and expects board-deck results in a week is going to misallocate budget. The right framing for stakeholders is two horizons. The crawl-and-citation horizon is days to weeks. The competitive-position horizon, where you actually move share of model on the prompts that matter, is one to two quarters. Promising the first horizon when finance is asking about the second is how AI visibility programs lose funding.
Getting cited is the easy part: staying cited is the budget question
Semrush's same 30-day experiment contains the line every CMO needs to hear before approving annual spend. Citations are not static. Pages cited on day one were not the same set of pages cited on day 30. Google AI Mode citations rose to 48 pages by day six, then fell to 21 pages by day 30. Microsoft built citation trends, cited pages, and grounding queries into Bing Webmaster Tools for the same reason: AI visibility is a moving measurement surface, not a one-time ranking report.
That decay is the real operating problem. The first citation is a one-time win. The maintenance budget is what keeps the curve from collapsing. Parse's data on how long an AI citation lasts puts numbers on that churn. Plan for an update cadence on cited pages, a refresh trigger when citations drop on a watched prompt, and a content velocity floor that keeps the most-cited corner of your site young. We laid out the cadence numbers in content freshness and AI citations and the volume side in content velocity for AI visibility.
A timeline you can give to a stakeholder
Give leadership a four-stage plan, with evidence to review at each stage.
Weeks 1 to 2 are technical baseline. Confirm crawler access for GPTBot, OAI-SearchBot, PerplexityBot, and ClaudeBot. Confirm Bing indexing. Lock the prompt set, the competitor set, and the platform mix. Nothing meaningful happens publicly yet, but skipping this step is the single most common reason later phases stall.
Weeks 3 to 8 are first-citation evidence. You should be seeing inconsistent citations on long-tail prompts, share-of-voice moving on low-competition queries, and the first signals of which content patterns AI retrievers are actually pulling from. SparkToro's repeated-prompt research is the right reminder here. Single-shot checks are not signal yet.
Weeks 9 to 16 are competitive position. You compare against the named competitor set, watch Share of Model on the prompts that map to revenue, and use citation gap analysis to decide where to push content next. This is the horizon where ROI conversations start landing, and the bridge into the framework in how to measure the ROI of AI visibility.
Quarter two onward is maintenance and compounding. Refresh cadence, new content velocity, earned media work for retrieval-grounded citations, and entity work for the slow training-data lane. The graph that lands in board decks one year in is shaped by what you do in this phase, not by anything that happened in week one.