Google's own figure for how long a structured data change takes to register is hours to one or two weeks, with weeks or never at the slow end. Gary Illyes, an analyst at Google, put that row on screen on 2 October at Search Central Live Deep Dive Europe in Barcelona, inside a set of tables covering how long crawling, indexing and serving actually take.
The numbers came off a slide, not a help page
The tables did not appear in a Google blog post or a help page. They reached anyone outside the room through an attendee's write-up: John Campbell, Head of Innovation and AI at the agency ROAST, recapped each of the three days, and PPC Land relayed the serving-and-ranking day on 3 October. No methodology, sample or period was attached to any of the figures. They are a senior Google engineer's sketch of the system rather than a published commitment, and nothing comparable exists in the documentation.
The sketch is detailed. Discovery of a new URL runs at about 20 hours typically, weeks to never at worst. A URL Google already knows gets refreshed in about 30 days. Sitemap processing takes about 24 hours, or up to 14 days, or never. End-to-end indexing is about an hour and a half. Recovery from a core update is three to six months, stretching to between six months and a year when the wait is for the next update to come round.
A restaurant menu moves faster than one to two weeks
Prices move, suppliers fall through and the special changes with them. Against that cycle, hours to one or two weeks is not a comfortable range, and the comfortable-sounding end of it describes the typical case rather than the guarantee.
Illyes attached a caveat that matters more than any single row: the processes are linked. A page cannot be indexed before it has been crawled, so the delays stack. Read the structured data row next to the 30-day refresh figure for a known URL and the two sit awkwardly together — markup cannot be re-read ahead of the page that carries it, and the slower number is the one doing the gating.
"Never" is a quality verdict rather than a queue
Campbell's commentary on the tables noted how often "never" appears, and that it is usually tied to quality: "Fast technical fixes don't help if Google doesn't think the page is worth it." Three of the rows carry quality as the stated reason for the slowest case, including the structured data one.
That reframes the problem for a restaurant site. A menu page Google has judged thin does not get a slow update. It gets no update, and the markup on it keeps describing a menu the kitchen has already moved on from.
Google's engineers disagree about whether AI reads the markup
The day before the timings, Ryan Levering, who works on structured data at Google, described what happens to Schema.org data once it is collected: cleaned, filtered, then either shown in ordinary search results or used as context for AI Overviews and AI Mode. The text in the schema travels with the text on the page, so in his phrasing "it does feed Google's AI features" — not as code, but added to context.
The next day's graphic said the opposite, that AI Overviews ignore structured data and work from normal indexed text, and Illyes said AI results do not need structured data because they take text and build the result from it.
Both accounts fit if the markup is read as writing rather than executed as instructions, and that reading has an edge to it. The strings inside a MenuItem are editorial copy, read much as a sentence in the page body is read. A dish name entered as "Soup" earns nothing from sitting in a well-formed JSON-LD block.
What the markup still buys is precision — a price in a price field, a dietary flag that means one thing — and a shape Google can clean and filter without inferring anything. Neither of those survives the timing row. Precision that is a fortnight out of date is worse than vagueness, because nothing downstream marks it as old.
