It is the question almost every client asks us right after they publish: I just put this page live, when will the AI engines know it exists? The honest answer is that there is no single number, because "AI" is not one system. ChatGPT, Perplexity, Google AI Overviews, Gemini, and Copilot each find and use new content on their own clocks, and those clocks run from a few hours to several months apart.
The good news is that the pattern is predictable once you understand the two very different ways AI uses your content. One path is fast and you can influence it this week. The other is slow and largely out of your hands. Below is what we see in our own testing and across the audits we run, plus the concrete moves that consistently shorten the wait.
The two ways AI "knows" your page
Before you can answer how fast AI indexes a new page, you have to separate two mechanisms that often get lumped together:
- Live retrieval (fast). The engine fetches current web pages at the moment you ask a question, then writes its answer from what it just read. This is how Perplexity works by default, how ChatGPT works when it browses, and how Google AI Overviews pull from the live index. A brand-new page can show up here in hours.
- Training memory (slow). The model also carries knowledge baked in during training. That snapshot is frozen until the next training run, which can be months old. If an engine answers from memory and does not browse, it simply will not know about anything you published recently.
This split explains the single most common frustration we hear: "ChatGPT found my page yesterday and today it acts like it never existed." Same prompt, different mode. When it browses, it sees you; when it answers from memory, it does not. For a deeper look at the retrieval side, our explainer on what answer engine optimization is and how it works walks through how these systems choose what to read and cite.
How fast does each engine pick up a new page?
Here is the rough timeline we observe across new pages we publish and monitor. Treat these as typical ranges, not guarantees, because crawl frequency depends heavily on how established and well-linked your site already is.
| Engine | How it finds pages | Typical time to pick up a new page |
|---|---|---|
| Perplexity | Live web retrieval on most queries | Hours to a few days |
| ChatGPT (browsing on) | Live search when it decides to browse | Hours to days |
| Google AI Overviews | Tied to the standard Google index | Days to a few weeks |
| Gemini | Google index plus live grounding | Days to a few weeks |
| Copilot | Bing index plus live retrieval | Days to a couple of weeks |
| Any model's training memory | Periodic retraining only | Months, if ever |
The headline takeaway: the live-retrieval surfaces are the ones worth optimizing for speed, because they are where you can realistically be present within the first week. Chasing the training-memory layer is a losing game for fresh content, since you cannot influence when the next model version ships.
What slows a new page down
When a page does not get picked up, the cause is almost never that AI "decided" to ignore it. In the audits we run, the delays trace back to a short list of fixable problems.
Blocked crawlers
This is the one that quietly sinks more sites than any other. If your robots.txt disallows the AI crawlers, the engine cannot read the page no matter how good it is. Check that you are not blocking GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, or Google-Extended. We routinely find businesses that accidentally blocked these while trying to "keep AI off our site," then wonder why they are invisible.
Orphaned pages
Crawlers follow links. A page with no internal links pointing to it is an island, and islands get visited late or never. Every new page should be linked from at least one page that already gets crawled often, such as your homepage, a hub, or a recent article.
Thin or buried answers
Even when a page is read, it may not be used. If the answer to the user's question is buried six paragraphs down or wrapped in marketing language, the engine may skip it in favor of a competitor who states the answer cleanly up top. Live-retrieval engines are quoting machines: they reward content they can lift in a sentence or two, and they pass over pages that make them work for it.
A brand-new domain with no history
Crawl frequency is partly a function of trust. An established site that publishes regularly gets revisited often, so a new page on it is discovered quickly. A brand-new domain with no inbound links and no track record sits in a much slower queue. This is not something you can shortcut overnight, but it is a reason to keep publishing and earning links steadily rather than expecting a cold site to get instant pickup.
What speeds a new page up
These are the levers that move the needle, ordered roughly by impact in our experience:
- Make it crawlable. Allow the AI bots in robots.txt, return a clean 200 status, and keep the page free of render-blocking issues. This is table stakes.
- Link to it immediately. Add internal links from high-traffic, frequently-crawled pages the day you publish. This is the fastest way to get a crawler to discover the URL.
- Submit your sitemap and update lastmod. For Google-based surfaces, an accurate sitemap with a fresh lastmod date and a Search Console indexing request prompts a quicker recrawl.
- Lead with the answer. Put a direct, two-to-three sentence answer near the top, then support it. Answer-first structure makes a page easy to quote, which is what live-retrieval engines reward.
- Add valid schema. FAQPage, Article, and Organization markup help engines parse what the page is and lift the right snippet. Our guide to structured data for AI search covers exactly which types matter.
- Earn a trusted mention. A link or citation from a source the engine already trusts, a directory, a respected publication, a review platform, is the single biggest accelerator we see. It both surfaces the URL and signals that the page is worth using.
Indexed is not the same as recommended
Here is the distinction that saves clients a lot of disappointment. Getting a page read in days is very different from getting your business recommended when someone asks an engine for help. Recommendation is a slower, cumulative outcome built on consistent signals: a coherent site, real reviews, accurate listings across the directories AI trusts most, and content that repeatedly answers the questions your buyers actually ask.
A useful reference point comes from a public result we like to cite: Keith Akada, a Seattle mortgage broker, went from effectively invisible in AI search to the number-one AI-recommended broker in his market, with roughly 30 leads and four closed deals, in about six weeks. The individual pages were crawlable within days. The recommendation, the part that produced leads, took the full six weeks of stacked signals to materialize. That gap between "indexed" and "named" is exactly what most timelines miss.
A realistic timeline to expect
Putting it together, here is what we tell clients to plan for after they publish a strong, crawlable page:
- Day 0 to 3: Perplexity and a browsing ChatGPT can reference the page once it is linked and crawlable.
- Week 1 to 3: Google AI Overviews, Gemini, and Copilot begin pulling it as normal indexing catches up.
- Week 4 to 8: If the rest of your signals are in place, the engines start recommending you, not just citing the page.
- Months out: The page may eventually fold into a model's training memory, but that is a bonus, not something to wait on.
One more practical note: do not judge pickup speed by a single prompt. Because engines flip between live retrieval and training memory, you should test the same question several times, across a few days, and in more than one tool before concluding a page is missing. We routinely watch a page that "wasn't there" yesterday surface cleanly the next morning once a browsing pass picks it up. Patterns over a week tell you far more than any one screenshot.
If you are weeks past publishing and the live-retrieval engines still cannot find a page, the problem is almost always mechanical, blocked crawlers, no internal links, or a sitemap that never updated, rather than a content quality issue. Fix the plumbing first, then the speed follows. The businesses that show up fastest in AI are simply the ones who made themselves the easiest to read and the most worth quoting.