Insights — Generative Engine Optimization

How Long It Takes to Get Cited
in AI Search

Every answer to this question is a number, and no number survives being asked where it was measured. What follows is not an estimate. It is our own log — eighteen days of daily measurement on our own domain, from the day of the audit to this morning, with every zero left in.

It is three clocks, not one interval

The wait is not one interval. It is three clocks running at once, and they are not the same length, not under the same control, and not started by the same event.

01

Recrawl — days, and it is yours

The first is recrawl: how long until the page you just published is fetched at all. This one is yours. A correct sitemap, pages that return 200 instead of 404, and a server that does not make a crawler wait are the whole of it, and it is the fastest of the three.

02

Index refresh — not yours, and it varies

The second is index refresh: how long until the index the engine retrieves from contains that page. This one is not yours, it differs between assistants, and it changes without notice. You cannot shorten it. You can only stop assuming it is instant.

03

Corroboration — no upper bound

The third is corroboration: how long until something written by someone else mentions you. This one has no upper bound, because it does not run at all unless somebody starts it. A domain where every statement about the brand was written by the brand can sit at zero indefinitely with all its technical work finished. That is the clock that decides whether the other two ever matter.

This is why the published estimates disagree with each other so violently. An article that says four to eight weeks and an article that says three to six months can both be honest and still be measuring different clocks on different kinds of domain. Neither was measured on yours.

Eighteen days, measured daily, still zero

We sell AI search visibility. On August 11, 2026 our own site had none, and instead of estimating how long that would take to fix, we started recording it daily. Below is what shipped and what the measurement said on the same day, from the audit to this morning.

beaugust.co — what shipped and what the measurement said, August 11 to August 28, 2026
Date What shipped What the measurement said
Aug 11 Audit baseline. Two of the three offices the site declared returned 404 with noindex, and the structured data pointed at another agency's Instagram. 0 of 3 queries, including the exact domain name.
Aug 14–21 Entity work: legal name and five alternate names in the Organization, FAQPage on both home pages, three service pages, privacy policy and terms, a sitemap that matches what exists. 0 of 5 commercial queries. Measured every day.
Aug 21–26 Six buying and mechanism questions put to an answer engine, one per day, phrased the way a buyer phrases them. 0 appearances in 6 answers.
Aug 24–26 Three articles and the section index that is their parent — the first URLs on the domain shaped like articles rather than like sales pages. 0 of 5. No movement in any query.
Aug 27 Three route prefixes stopped returning 404, and the llms.txt stopped naming a URL of our own that did not exist. 0 of 6. A sixth query enters the panel with its article.
Aug 28 Full crawl: 13 of 13 URLs return 200 with a unique title, a meta description, a self-referential canonical, exactly one H1, valid structured data and reciprocal hreflang on the four mirrored pages. 0 of 6. Thirteenth consecutive round.

Eighteen days. Thirteen consecutive rounds on the commercial queries, six answer-engine questions, and a technical baseline that now passes every check the articles on this subject tell you to pass. The count is still zero, and that is the finding, not the failure — it locates the constraint. Two of our three clocks are done. The third has not started, because nothing outside this domain has mentioned it yet.

Method, declared so the numbers can be checked rather than believed. The commercial queries are live web-search verifications read in full, not Search Console positions — they are different measurements of different systems and we never file one as the other. The six answer-engine questions were put to AI Overviews. Every row above has a date and a source in our panel, and the zeros are recorded the same way a citation would be.

Read the full open log, with the baseline and every change →

Three things that move before the citation does

A metric that reads zero for weeks and then reads one tells you nothing on the way there. These three read something every single day, and all three move before the first citation does.

01

The shape of the answer

The shape of the answer. Ask the same question daily and record whether the engine names companies at all, or only lists directories and roundups. An answer that stops listing directories and starts naming firms has moved, even if none of the firms is you — it means the engine now believes the category has identifiable players, which is the state it has to be in before it can name one more.

02

The kind of page cited

The kind of page cited. Record what each cited source actually is: a service page, an article, or a third-party list. This is the single most useful thing to log, because it tells you what to write next. If the answers to your buying questions are assembled entirely out of articles, no amount of rewriting your services page will enter that answer.

03

What the engine does with your name

What the engine does with your name. A brand query is the cheapest diagnostic there is. There is a real difference between an engine that asks you to clarify which company you mean, one that returns other companies with similar names, and one that states plainly that it has nothing on yours. The third is further along than the first, and you only see the progression if you recorded the wording each time.

None of the three is a proxy for revenue and none should be reported as one. They are instruments for deciding whether to keep doing what you are doing, which is the decision a monthly citation count arrives too late to inform.

Read: what the cited passages have in common →

Where this comes from

The log in section 02 is ours, is dated, and is the only empirical claim this page makes. The estimates it disagrees with are linked below so you can read them instead of taking our word for the disagreement.

Frequently asked

How long does it take to get cited in AI search?

There is no interval that holds across sites, and any single number you are given is a number somebody could not have measured on your domain. The wait is the sum of three separate clocks — how often your pages are recrawled, how often the engine you care about refreshes the index it retrieves from, and how long it takes for something outside your domain to mention you — and only the first of the three is under your control. On our own domain, eighteen days of daily measurement from a clean technical baseline have returned zero every time.

If the technical work is finished, why is the count still zero?

Because the technical checklist is the entry requirement, not the competition. Making a page retrievable — one H1, a canonical that points at itself, structured data that says what the visible text says — buys you the right to be considered by a retrieval step that then has to pick you over sources it can corroborate somewhere else. Finishing the checklist removes the reasons to exclude you. It does not create a reason to choose you.

What should be moving while the citation count is still zero?

Three things, and all three move before the citation does. First, the shape of the answer: whether the engine names companies at all or only lists directories. Second, the kind of page it cites: service pages, articles, or third-party roundups. Third, whether the engine still asks you to clarify your brand name or has started treating it as a known entity it simply has nothing on. A run of zeros with all three unchanged means the thing you changed was not the binding constraint.

Does publishing more articles shorten the wait?

It shortens one of the three clocks and does nothing to the other two. Articles give a retrieval step something of the right shape to lift — a passage that answers a question rather than a page that announces a service — so a domain with no articles at all cannot be cited no matter how good its service pages are. But the tenth article does far less than the first, and neither one causes anyone else to mention you, which is the clock that usually binds.

Is a month of identical zero results a wasted month?

No, provided the zeros were recorded with their date and their method. A run of identical results is evidence that the answer is stable, which is what lets you rule out the change you just made as the cause of anything. The month that is actually wasted is the one where nobody wrote the zeros down, because at the end of it the only available conclusion is to try the same fix again.

Why do published estimates disagree so much with each other?

Because most of them are measuring different things and almost none publish the log behind the number. An estimate drawn from established domains adding a page to an already-indexed site is measuring the index-refresh clock alone, and it will look fast. An estimate drawn from a domain nobody has ever mentioned is measuring the corroboration clock, and it will look slow or never resolve. Both can be honest and still be useless to you, because neither was measured on your domain.

Find out which clock is stopped

So the answerable version of the question is not how long. It is: which of the three clocks is stopped on this domain, and what starts it. That is an audit, it takes days rather than months, and the output is a decision rather than an estimate.