How long AI takes to pick up a new page

If the assistant searches the web while answering, a new page can be cited within days of being crawled and indexed. If it is answering from what the model already holds, a new page may take many months to matter and may never appear at all. Those are two different pipelines and the waiting time is different for each.

What is actually happening underneath

For the live path, three things have to happen in order. A crawler has to fetch your page. A search index has to store it. Then a question has to be asked whose search results include it. Any one of those can be the blocker, and the first two are the ones you can check.

For the memory path, your page has to be part of a future training run, and then that model has to be released. You do not control either date, and a single page on a small site is a very small amount of text in that process.

In our own runs across four companies, the first time an assistant cited a newly published page came between 9 and 16 days after publishing. Those are the founder's own runs, operated by hand rather than through the product, so read them as our experience and not as a benchmark. The worked example below is the one of those results that has a URL behind it.

The five stages, and which one you are stuck at

"It has been three weeks and nothing happened" is not a diagnosis. These are the five stages a page passes through, in order, with the check for each.

One, published and reachable. Is the page linked from somewhere a crawler already visits. An orphan page with no internal link waits far longer than it needs to.

Two, crawled. Your server logs will show the bot visits. This is the only stage you can verify directly rather than infer.

Three, indexed. Check in the search console of the engine concerned, or search for a distinctive sentence from the page in quotes. If the page does not come back, it is not indexed, and nothing downstream can happen.

Four, retrieved for a question. Ask a question the page is the natural answer to and see whether it appears among the citations. A page can be indexed for months and never retrieved, because it is not the best answer to anything anybody asks.

Five, quoted in the answer. Retrieved and then actually used. A page can be fetched and ignored because it does not contain a sentence worth repeating.

The fixes are different at every stage. Stages one to three are technical. Stage four is a question and vocabulary problem. Stage five is a writing problem. Knowing which stage you are at saves you from fixing the wrong one.

A worked example

clawlaw.in is our sister company. It published a page answering one narrow buyer question, how to check a company's court cases in India.

Fourteen days later, on 6 August 2026, ChatGPT cited that page as a source for a vendor due diligence question. At baseline the same question set had named the company nowhere in that zone, so on the live path the page went from absent to cited in a fortnight. This is the one result in our own work that a reader outside the company can check, because the answer named the URL.

Keep it in proportion. It is one page, on one assistant, and across our own runs on four companies the first citation came anywhere between 9 and 16 days. It also happened only on the searched path. Nothing published that month entered what the model remembers.

What the result does not tell you is which of the five stages was the slow one, because we did not log the crawl date and the index date for that page. If you want that, write down the publish date, look in the server logs for the first bot visit, and check indexing with a quoted sentence search. Then you know which stage your own wait is sitting at rather than guessing.

What it means for a business

Publishing one page and checking the next morning tells you nothing. Publishing a set of pages and checking every month tells you something by the second or third check.

It also means the slowest part is usually not the assistant. It is your page not being indexed yet, or being blocked, or having nothing on it specific enough to be worth quoting. A page that has been indexed for two months and still never gets cited does not have a timing problem, it has a content problem, and waiting longer will not fix it.

For planning, treat the live path as the one with a schedule and the memory path as something you build towards without a date. Any plan that needs the memory path to move inside a quarter is a plan with a hole in it.

What nobody can tell you

There is no published timeline for any of this, and there is no way to request one. Crawl frequency, index inclusion, retrieval and training schedules are all decided inside these companies and not announced.

So three honest statements. No agency can give you a date for a citation, and the range from our own runs is our experience on four companies, not a promise about yours. A page can be perfectly built and never cited, because retrieval depends on what else exists and on questions you do not control. And an older page can start being cited long after you stopped watching, which is one more reason to keep the monthly reading going rather than declaring the experiment over.

Common questions

Can I submit my page to ChatGPT or Gemini directly?

There is no submission form for an assistant. What you can do is make sure the search index behind it can reach the page: sitemap, internal links, no blocking in robots.txt, and plain HTML. That is the whole of the influence available.

Does updating an old page work faster than publishing a new one?

Often, yes, on the live path, because the page is already crawled and indexed. Updating a page that already gets visits is usually the quickest way to put a new fact into circulation. Change the content properly and show the date, rather than touching the page and hoping.

Why does one assistant cite my page and another does not?

Because they use different indexes and different retrieval steps, and some answers involve no search at all. This is normal and it is why a reading on one app is not a reading of your visibility.

How do I know if a crawler has even visited?

Look at your server access logs and filter by the bot names. This is the one stage with direct evidence, and most teams have never looked. If nothing has visited, stop worrying about your writing and fix reachability.

Should I keep publishing while I wait?

Yes, and publish into a cluster rather than in isolation. One page on a subject is a stray. Several linked pages on one subject make each of them easier to find and easier to trust. Do not publish thin pages just to raise the count.

Is there any point publishing if the model will not retrain for a year?

Yes, because the live path does not wait for a retrain. Most of the movement a smaller company gets in the first year comes from being retrieved, not from being remembered.

What to do first

• Link the new page from a page that already gets traffic. An orphan page waits far longer.
• Make sure it is in your sitemap and that the sitemap is submitted.
• Check the page is readable as plain HTML. If the text only appears after JavaScript runs, many crawlers see an empty page.
• Check your robots.txt does not block the crawlers you want. The bots that feed AI search are named separately from Googlebot, so allowing one does not allow the others.
• Write down the publish date. Then ask, once a month, a question that could only be answered well by that page.

Patience is the honest advice here, and it is not a comfortable thing to sell. The work that shortens the wait is being indexed quickly and being specific enough to quote. Nothing makes an assistant read you tomorrow.