Expect weeks for the mechanical half of AI visibility work and months for the half that decides whether an assistant says your name. Google publishes both halves. Its crawling guidance states that "Crawling can take anywhere from a few days to a few weeks." Its guidance on diagnosing search traffic drops goes further: "some changes can take effect in a few days, while others could take several months."
Think about changing your address. The mail redirect starts inside a week. Your dentist, your accountant, the friend who still sends a card every December: they update whenever they next hear from you, and that can take a year. Both things are true at the same time, and only one of them runs on a schedule.
One clock runs on crawling. The other runs on corroboration. Most owners get told about the first clock and then get judged on the second, which is why a serious first quarter so often feels like nothing is happening. Usually something is. You are reading the half of the scoreboard that moves last.
This is a plain map of the timing: what Google actually documents, why two clocks exist, what genuinely moves in the first ninety days, what only appears later, and how to tell whether the work you are already paying for is on track. No day counts, because nobody honest has one.
What Google Puts in Writing About Timing
Start with the one participant that documents its own behavior. Google's page on asking it to recrawl your URLs says: "Crawling can take anywhere from a few days to a few weeks." That is the first leg of any timeline, and it runs on somebody else's schedule. A page an engine has not fetched is not available for that engine to quote. Nothing downstream starts until it does.
Recrawling is not the same as reassessment, and Google separates them. Its troubleshooting guidance says "some changes can take effect in a few days, while others could take several months," and adds a specific reason for the slow end: "it may take months before our systems determine that a site is now producing helpful content in the long term." On the narrower question of changed URLs during a site move, it offers a rough marker: "a medium-sized website can take a few weeks for Google to notice the change; larger sites can take longer."
Then comes the sentence every timeline promise has to survive. Google writes: "there's no guarantee that changes you make to your website will result in noticeable impact in search results." That is written about Search rather than about AI answers, and it is still the company with the most documentation and the most control over its own platform, declining to promise an outcome on it. Read that twice.
The reason it carries into AI answers is that Google says its generative features run on the same fundamentals. Its guidance on optimizing for them, last updated on July 10, 2026, says you do not need new machine readable files, AI text files, markup or Markdown to appear in Google Search including its generative capabilities. Google's guidance names no AI-specific submission route. There is no fast lane on offer.
- Crawling: a few days to a few weeks, per Google's own recrawl documentation.
- Noticing changed URLs after a site move: a few weeks for a medium-sized site, longer for large ones.
- Quality reassessment: potentially months, and explicitly not guaranteed.
Every range above is Google's, about Google. No model company publishes an equivalent for its own assistant. That absence is the honest reason the rest of this answer is a framework rather than a number.
Why Two Clocks Run at Different Speeds
The first clock is retrieval, and it is the quicker one. Google describes AI Mode using a "query fan-out" technique, issuing multiple related searches concurrently across subtopics, and tapping "fresh, real-time sources like the Knowledge Graph, info about the real world, and shopping data for billions of products." When a product looks things up as it answers, a corrected page becomes eligible to shape what it says once that system can crawl, index and retrieve it. Eligible is not the same as chosen.
The second clock is memory, and it has no timetable you can plan around. A model's training has a cutoff, which is why an assistant can describe the practice you ran three years ago with total confidence. Providers publish cutoffs for individual models, and products increasingly layer live search on top, so this is less binary than it sounds. What none of them publish is a date by which a given fact about your business will change an unaided answer. That mechanism is covered properly in the training data echo, and the question of which engine leans on memory versus live results is answered in where ChatGPT gets information about your business.
So engine differences do change this answer. A product that searches the live web while it answers can potentially reflect a corrected page sooner than one relying on model knowledge alone, and the gap between them is not something anyone outside those companies can quantify. That is the argument for reporting engines separately alongside any aggregate. A single blended number averages two clocks that do not tick together. It hides the one thing you wanted to know.
The rest of this is our operating framework rather than anybody's published finding, and it is worth saying so plainly. In our experience, sorting your own problem into one of the two clocks makes the wait legible. Wrong or missing details, unreachable pages, and questions you have simply never answered in public tend to move on the faster clock. Nothing outside your own website backing up what you claim tends to move on the slower one, because that repair depends on other people publishing, and on engines finding it later.
What Actually Moves in the First Ninety Days
Take a baseline before anything ships, while the number is still whatever it is. The version worth having asks the five engines your buyers actually use, ChatGPT, Gemini, Perplexity, Claude and Grok, the questions they actually ask. You can run a rough version yourself in an afternoon. Checking whether ChatGPT recommends your business walks through it, free, in about half an hour.
Then judge the quarter on the indicators that are capable of moving inside it. How much of your buyer question set is now covered by something you have published. Whether the engines can reach and read those pages. Whether your name, people, credentials and locations agree with each other everywhere a machine can look. How you get described when you do come up. Which competitors get named instead, and on what basis.
Those are leading indicators. They are unglamorous by design, and they are what we report at day ninety, because they are the inputs and they are capable of moving. The discipline for turning them into a defensible number, an appearance rate with a stated denominator rather than a rank, belongs to measuring whether AI visibility work is actually working, and it is worth reading before you accept anyone's dashboard.
Here is what should not carry the verdict at quarter one: AI attributed lead volume, revenue, or a screenshot of one good answer on one engine on one afternoon. Report the first two if you have them, with the attribution limits and the small sample said out loud. Never let a screenshot stand in for a trend. If the category itself is new to you, start with the plain definition of AI visibility and come back.
What Only Shows Up Later
The lagging indicators are the ones you actually want: your name arriving unprompted in a shortlist, inquiries from people who say an assistant sent them, first calls that start further along than they used to. They lag because they depend on evidence accumulating that you do not publish yourself, and Google has already told you what that feels like from the inside. Months, in its words, before its systems determine that a site is producing helpful content in the long term.
Size plays into the wait, though not the way it feels from the inside. A firm with twenty years of work and four pages describing it is not lagging because it is small. It is lagging because most of its record is still private, and the slow clock cannot start until something exists off your own website for an engine to find and check. The larger your existing public trail, the more of that clock has already quietly run without you noticing.
That is also why the second quarter is where the shape of the work becomes visible. Corroboration arrives on other people's schedules. Never on yours. As a working heuristic, something has to be reachable before it can be retrieved, retrieved before it can be quoted, and quoted repeatedly before it reads as settled, and each of those steps carries its own lag. A recommendation is what the earlier rungs produce once they hold, which means it is the last thing to appear and the first thing everyone asks about.
So read lagging indicators across repeated, comparable periods rather than off a single one. Our own practice is two quarters against the same question set before anyone calls it a result. One quarter is a weather report. Two is a trend you can act on.
How to Tell Whether Yours Is On Track
If you already have an agency and have been told the AI side is handled, you do not need to fire anyone to find out whether that is true. You need to know which layer is covered. The coexistence map for owners with an incumbent SEO partner lays that out; the timing version is a shorter set of questions.
- Ask what the baseline was. Which questions, which engines, what date. No baseline means no timeline can be assessed, only asserted.
- Ask what is expected to move by day ninety, in writing, before day one. Leading indicators only. If the answer includes lead volume, the expectation is being set to fail.
- Ask what has actually been fetched. Published is not the same as crawled, and crawled is not the same as quoted.
- Ask what exists off your own website. If the answer is nothing yet, the slow clock has not started.
On track at day forty five looks modest and specific: the technical layer checks out, new answers to real buyer questions are live and fetched, your identity agrees with itself across profiles, and the panel is being re-run on schedule. Stalled looks like activity with no measurement attached, or measurement with no dates on it. Both are common. Neither is malice.
Then the limit that makes this question hard in the first place. Model and product behavior changes when the companies change it, sometimes announced, never negotiated with your content calendar. No outside marketer can guarantee that an assistant will name you, on any date. Readiness is what you can build and prove. A guaranteed date is not a timeline. It is a sales technique, and the honest version of this answer loses deals to it.
If you want a dated starting point before you judge anyone's work, the free Recommendation Check asks your buyers' questions across five engines and scores your record against 109 checkpoints in eight zones. Scores run 0 to 109. Ninety and up means AI treats you as a primary source. Under 35 means you are effectively invisible. There is no discovery call attached to it.
Want to Learn More?
Probably Genius builds and operates the recommendation layer around expert-led businesses, then measures it in the open. That is the job inside The Answer: twelve done-for-you thought-leadership articles a month, your expertise structured so machines can read it, corroboration beyond your own pages, and a monthly five-engine probe with every answer saved word for word. The published methodologies show how it runs, including the 109-Point Diagnostic behind the free check.
You're probably a genius at what you do. We make sure AI gets the memo.