There is no fixed answer. With web search on, a new page can be cited once it is crawled, indexed and judged relevant, which can take days or weeks. From memory, it waits for a model trained after the page existed. The only reliable number is the one you measure: publish, log each step, and re-ask on a schedule.
Most of the confusion comes from treating ChatGPT as one system. It is two. When it answers from memory, it uses what the model learned in training, and a page published last week is simply not in there. When it searches, it reads the web as it is today, and a new page can appear as soon as it can be found.
So the useful question is not how long ChatGPT takes. It is how long each step takes for your site, and where your page gets stuck. This note gives you a test that answers that, with a log you can copy.
Why is there no single answer?
Memory and the live web are two different systems with two different clocks.
When ChatGPT answers without searching, it draws on its training data. That data has a cut-off, and new content only enters it when a new model is trained and released. You cannot speed that up, and nobody outside the model maker knows exactly what went into a given training run. If you are waiting for a new page to show up in memory, you are waiting for a model release.
When ChatGPT searches, it reads pages that exist now. OpenAI says ChatGPT search draws on third-party search providers and on its own crawler, OAI-SearchBot, and that public websites can appear in its results. So a new page can reach a search answer as soon as those systems have found it and judged it relevant to the question. That depends on how quickly your site gets crawled, which varies a lot by site.
Our sibling note, Remembered or retrieved: the two ways an AI assistant can name your company, explains how to tell which system produced an answer. For this test, keep them apart from the start: every question gets asked once with search off and once with search on, and the two are logged separately.
What has to happen before a page can be cited?
Four steps, and a page can stall at any of them.
Between publishing and being cited, a page has to clear four steps.
- Reachable. Your robots.txt, firewall and bot protection allow the crawlers in. OpenAI documents OAI-SearchBot as the crawler for ChatGPT search and says sites that block it will not appear in those answers. It also notes that robots.txt changes can take around 24 hours to take effect on its side. That figure is about robots.txt, not about how fast new content appears.
- Found. A crawler or search engine learns the URL exists. Internal links from pages that are already crawled, an up-to-date sitemap, and direct submission all help.
- Indexed. The page is stored and can be returned for relevant searches. You can check this in Google Search Console with the URL Inspection tool and in Bing Webmaster Tools, both of which also let you request crawling of a URL.
- Retrieved and cited. For a specific question, the assistant searches, the page is among the results it reads, and it is useful enough to cite. This step depends on the question and on what else competes for it.
Bing supports IndexNow, an open protocol that lets a site notify participating search engines when a URL is added or changed, and Bing Webmaster Tools lets you submit URLs directly. OpenAI does not name its third-party search providers in its publisher FAQ, so treat submitting to the major search engines as a way of shortening the “found” step generally, not as a switch that turns on ChatGPT.
How do you run the test?
One new page, one question it should answer, and a log you fill in on fixed days.
Set it up in under an hour.
- Pick a question your buyers ask that no page on your site currently answers well. Write it down word for word. This is the test question.
- Before publishing, ask the test question in ChatGPT with search on and off, in a fresh chat with memory and personalisation off. Record who is named and which domains are cited. This is your baseline.
- Publish one page that answers the question directly in its first paragraph, in text, linked from at least one page that is already indexed.
- Confirm the page is reachable: check robots.txt and any bot protection for OAI-SearchBot, Googlebot and Bingbot.
- Submit the URL in Google Search Console and Bing Webmaster Tools, or ping IndexNow if your site supports it. Note the time.
- Watch your server logs for visits from each crawler, identified by user agent. OpenAI publishes IP ranges for its crawlers, so you can confirm a visit is genuine rather than someone borrowing the name.
- Ask the test question on a fixed schedule: day 1, 3, 7, 14, 30 and 60. Same wording, same modes, fresh chat each time. Log the result.
Ask a second, related question on the same days as a control: one that your page should not affect. If both answers change at once, something other than your page probably moved.
What should the test log look like?
One row per event, with a date on every row.
| Event | Where to see it | Date and time | Note |
|---|---|---|---|
| Baseline asked (search on and off) | ChatGPT, fresh chat | Who was named, which domains cited | |
| Page published | Your CMS | URL and exact test question | |
| URL submitted | Google Search Console, Bing Webmaster Tools, IndexNow | Which of the three | |
| First Googlebot visit | Server logs | User agent and status code | |
| First Bingbot visit | Server logs | User agent and status code | |
| First OAI-SearchBot visit | Server logs, checked against OpenAI’s published IP ranges | User agent and status code | |
| Indexed in Google | Search Console, URL Inspection | ||
| Indexed in Bing | Bing Webmaster Tools | ||
| First cited in ChatGPT with search | Scheduled re-ask | Cited URL, named or not | |
| First visit from ChatGPT | Web analytics, utm_source=chatgpt.com | Landing page | |
| Named from memory | Scheduled re-ask, search off | Usually not within the test window |
Leave empty rows empty. An empty row on day 30 is a finding: it tells you which step the page is stuck on. No crawler visit points to reachability or discovery. A crawl with no index points to the page itself, such as thin content or a duplicate. Indexed but never cited points to the page not being the best answer for the question, which is a content problem, not a technical one.
What should you expect, and what should you not promise?
Expect a range, measured on your own site. Promise nothing else.
Sites that are crawled often, with strong internal linking, tend to see new pages found quickly. Sites that are rarely crawled can wait much longer for the same step. That is why a timeline from someone else’s site tells you little about yours, and why the log is worth more than any figure you read online, including any figure you might read here.
Being indexed does not mean being cited. A page can be found, indexed and still never chosen for the question, because another page answers it better or because the assistant searches with different words than you expected. Look at the domains that are cited instead of you. They show what the assistant considers a good answer.
Being cited once does not mean being cited every time. Answers vary between runs. A page that appears on day 7 and not on day 14 has not necessarily lost anything. Look for a pattern across several dates before drawing a conclusion.
If you work with an agency or a tool that promises a timeline, ask three things.
- Which step does the timeline cover: crawled, indexed, cited from search, or named from memory?
- On which sites was it measured, and how many?
- Can they show the log?
Do not promise anyone a timeline. Promise a log.
How often should you run it?
Once properly, then every time you change how you publish.
Run the full test once, on one page, to learn how your site behaves. After that you know roughly where your pages stall and how long each step tends to take for you. Run it again when something changes: a new site, a new CMS, a change to your bot protection, or a new section of the site.
Keep the monthly buyer-question check going alongside it, as described in Does ChatGPT recommend my company? A one-hour check you can run today. The test tells you how fast one page moves. The monthly check tells you whether the whole effort is changing what buyers are told.
How long it takes is not a number you can look up. It is a number you can measure, one dated row at a time.
Questions people ask next
How long does it take for ChatGPT to find a new page?
There is no published figure. With web search on, a page can be cited once it has been crawled, indexed and judged relevant, which varies by site from days to weeks. From memory, it has to wait for a model trained after the page existed.
What is OAI-SearchBot?
The crawler OpenAI uses to surface websites in ChatGPT search. OpenAI says sites that block it will not appear in those answers. It is separate from GPTBot, which relates to training.
Does submitting to Bing help with ChatGPT?
It helps your page get found by Bing, and OpenAI says ChatGPT search uses third-party search providers alongside its own crawler. OpenAI does not name those providers in its publisher FAQ, so treat submission as general good practice rather than a direct route into ChatGPT.
How can I tell if ChatGPT sent me traffic?
OpenAI says ChatGPT adds utm_source=chatgpt.com to referral links from its search results, so those visits can be separated in your web analytics.
Can Throughline tell me when a new page starts showing up in AI answers?
It asks your buyer questions of the major assistants and Google’s AI Overviews on a schedule and dates whether you were named and which pages were cited, so a first appearance shows up as a dated entry. Crawling and indexing stay with the search engines.