The Google search shift
Google has not stopped sending traffic. It has started answering instead.
AI Overviews put a synthesised answer above the results, assembled from sources the user never has to click. For a large share of queries that ends the session. The page that would have earned the visit still exists, still ranks, and never gets opened.
This is the mechanism behind a set of numbers that have been widely reported over the past two years, and it changes what a ranking is worth.
What the published measurements say
How widespread are AI Overviews?
Published industry measurements put AI Overviews in roughly 54.6% of Google searches by query volume — more than half of all searches now return a generated answer above the links.
How much traffic does that remove?
Three figures are commonly cited:
- Zero-click searches — searches where the user gets an answer and never visits a site — are measured at around 58.5% of all Google searches.
- Position-one click-through rate falls from roughly 28% to 19% when an AI Overview is present on the query.
- Ahrefs found that queries carrying an AI Overview saw 34.5% fewer clicks to the underlying websites.
Looking forward, Gartner has forecast that organic search traffic will fall 25% by 2026, attributing the decline to AI agents and AI answer features.
How reliable are these numbers?
They are third-party industry measurements and forecasts, not Norg's own data, and they should be read as such. Methodologies differ, samples differ, and results vary considerably by query type — informational queries lose far more clicks than transactional or navigational ones. The Ahrefs and Gartner figures are attributable to named sources; treat the others as indicative of direction rather than precise.
The direction, though, is consistent across every published study: when a generated answer appears, fewer people click.
What actually changed
Isn't this just featured snippets again?
Featured snippets answered a narrow class of factual queries by quoting one source, with a link attached. An AI Overview synthesises across several sources into new prose, and the brands named inside it are the ones the user considers.
The difference matters commercially. A snippet took a click from one site. A synthesis decides which brands get mentioned at all — and a brand that is not in the generated answer is not in the consideration set, whatever its ranking position underneath.
Why does ranking still matter but matter less?
Ranking determines whether a page is eligible to be drawn on. It no longer determines whether the page is read. Those used to be the same thing.
A page can rank first, be used as a source for the generated answer, and receive no visit. In click terms that looks like failure. In influence terms the brand shaped the answer — but only if the model could parse the page well enough to attribute the claim.
What is the measurement problem?
Most analytics tools classify agent requests as bot traffic and filter them out. A business can therefore lose visibility in AI answers, or gain it, with no movement in any dashboard it currently watches.
This is why the decline tends to be noticed late: the traffic line falls, but nothing explains why, because the channel that replaced it is invisible to the tooling.
What this does not mean
Is SEO dead?
No, and any claim that it is should be treated with suspicion.
Search still drives enormous volume, transactional and navigational queries still produce clicks, and a page has to be crawlable and indexed before it can be drawn on by anything. What has changed is that ranking is now necessary but no longer sufficient. The work that gets a page found is not the same work that gets its facts understood and cited.
Does structured data guarantee citation?
No. Structured data makes a page easier to parse, attribute and trust; it does not compel any model to mention a brand. Anyone offering a guarantee of AI mentions is describing something the platforms do not permit them to control.
What structured data does is remove ambiguity. A model deciding whether to name a brand is weighing whether it can defend the claim. Clear, sourced, machine-readable facts make that defensible. Marketing prose does not.
What follows from it
What should a business measure instead?
Citation frequency rather than click-through: how often the brand appears in generated answers for the queries that matter, how it is described when it does, which competitors appear alongside it, and whether the description is accurate.
None of that is visible in conventional analytics, which is the practical difficulty — the channel has to be instrumented deliberately before it can be managed.
Where does Norg fit?
Norg publishes a business's verified facts in the formats these systems read, and monitors where the brand is cited. The platform overview is at norg.ai/product, how the underlying entity graph is built at norg.ai/platform/data-relationships, and published pricing at norg.ai/pricing.
Client engagements, with named brands, stated metrics and defined measurement windows, are at norg.ai/case-studies.