Guide
How Fast Does ChatGPT Start Recommending You?
Citation speed follows a curve, not a switch. In the one dataset that has actually tracked it, ChatGPT cited just 8 of the tracked pages on day one, climbing to 42% of the set by day 30, still rising when the tracking stopped. Almost nothing happens in the first two days.
Say you rewrote your homepage and FAQ two weeks ago, checked ChatGPT this morning, and found nothing. That is not a signal. It is the middle of the normal wait, not the end of it.
Citation speed is the gap between a page becoming crawlable and an AI assistant actually citing it in an answer, and that gap is measured in weeks, not hours.
What does the citation curve actually look like?
Semrush tracked how quickly ChatGPT started citing a set of newly published pages over a 30-day window. The pattern was slow at first, then steady.
| Time since publication | Pages cited (of the tracked set) |
|---|---|
| Day 1 | 8 |
| Day 7 | 14 |
| Day 14 | 28 |
| Day 30 | 34 (42% of the set) |
This is the exact question the check answers: not whether you have fixed the site, but whether an assistant has actually noticed yet.
Semrush’s own tracking stops at day 30, and the count is still climbing there, not levelling off: 8 pages cited on day one, 34 pages (42% of the set) by day 30. There is no published figure past that point, so day 30 is the last confirmed reading on a curve that was still rising, not a finish line.
Two days proves nothing. Neither does day seven, with only 14 pages in. The count kept climbing all the way to day 30, and it was still moving upward when Semrush’s tracking stopped.
Why does it take weeks instead of hours?
An assistant cannot cite a page it has not fetched, and it cannot fetch a page on a crawl schedule it does not control. A new or edited page has to be crawled again, folded into whatever index or retrieval layer the assistant draws from, and then actually selected as the best answer to a real question. Each step runs on its own clock. None of them fires the moment you hit publish.
We check this directly rather than guessing at it: an agent runs real buyer-intent questions through the assistants and records whether and how a business is mentioned, instead of one model being asked to simulate another’s answer.
What should you actually do while you wait?
Use day 30 as your first checkpoint. Not the day after a change.
Before day 30, keep collecting evidence rather than re-editing. Save the exact questions you are testing, run them again on a schedule, and note which pages you changed and when. That log is what tells you whether nothing happened yet, or whether nothing is going to happen.
Past day 30, give it more room before concluding anything. Semrush’s own count was still rising when its tracking stopped, so a page sitting at zero on day 30 has not necessarily missed its window. A sensible second checkpoint is day 60, double the stretch Semrush actually measured. If you are still at zero there, stop waiting and start checking the setup instead: is the page actually crawlable, does it directly answer the question a buyer would ask, and is the fact pattern on it consistent with what else exists about the business online. A slow citation is normal. A structural block is not, and no amount of extra waiting fixes it.
Common questions
Does editing an existing page restart the clock? Not from zero. An assistant that has already crawled and cited a page tends to pick up a meaningful edit faster than it found the page the first time, because the crawl and retrieval path already exists. The 30-day window above applies most cleanly to genuinely new pages.
Does the timeline differ between ChatGPT, Perplexity and Claude? Probably, but there is no published curve for Perplexity or Claude the way there is for ChatGPT. Semrush’s tracking covered ChatGPT search and Google AI Mode only. The mechanism holds regardless of platform: a page has to be crawled, folded into whatever the assistant retrieves from, and then selected as an answer, and each step runs on its own schedule. Until you have your own evidence for a given platform, assume a similarly slow start rather than a fast one.
How do you know if you have actually been cited, rather than guessing from a handful of manual prompts? Repeatedly typing questions into ChatGPT yourself only samples what you happen to ask. An agent set up to run the buyer-intent questions systematically, on a schedule, reports the real mention rate, which is a different thing from a spot check that happened to come back empty.
Sources
- Semrush, “How Fast Do AI Search Platforms Cite New Content?” (2025): semrush.com/blog/how-fast-do-ai-search-platforms-cite-new-content
Get a free report on your own business
Related: How do I check what ChatGPT says about my business? · Does robots.txt block ChatGPT? Not the bot that actually cites you
Courage Horizon sets up AI agents for UK small businesses: software that does one job, such as chasing unpaid invoices, and asks you before it sends anything. Being found when customers ask ChatGPT is part of that work. Get a free report on your own business.