Google is testing another way to remove friction between a compact AI Overview and the much larger answer hidden behind it. In the observed interface, the familiar “Show more” control is replaced by a temporary “Loading...” state. Within seconds, that state disappears and the AI Overview expands, without requiring the user to press a conventional expansion button.
The experiment is limited and should not be mistaken for a universal interface change. Search Engine Watch documented the behavior in late September, but the format is not consistently reproducible across users and queries. Google has not announced a broad rollout or published metrics showing whether the new treatment changes engagement.
For publishers, however, the test raises a useful measurement question. If Google changes expansion from an explicit user choice into a loading state that automatically reveals more of the answer, the number of people exposed to deeper AI Overview content—and potentially to additional source citations—could change even when the underlying ranking and citation systems remain identical.
The important change is not the word “Loading”
At first glance, replacing “Show more” with “Loading...” looks like microcopy. The observed behavior makes it more consequential. Search Engine Watch reports that the loading control appears briefly and then disappears as the full AI Overview becomes visible, leaving no “Show more” button.
A “Show more” control communicates an optional action. The user decides whether the visible summary is sufficient and explicitly requests the rest. A loading indicator communicates something different: more content is already on its way.
That changes the interaction model from user-initiated expansion toward system-initiated expansion. The distinction could affect how long the AI answer becomes before a searcher reaches conventional organic results, and how many elements inside the expanded answer receive an opportunity to be seen.
The test follows Google’s move toward dynamically expanded AI Overviews
The experiment does not appear in isolation. In August, Google confirmed that for some queries AI Overviews may dynamically expand when its systems determine that a larger answer would be useful. In those cases, the full response can appear without the traditional “Show more” click, accompanied by an open “Ask anything” follow-up interface.
Google said its research indicated that the dynamic experience made Search more helpful and encouraged deeper follow-up exploration. It did not disclose what percentage of queries receive the expanded treatment.
The “Loading...” treatment can therefore be read as another interface variation around the same broader question: how much of the generative answer should Google expose automatically before asking the user to do anything?
Expansion depth can become a citation-visibility variable
For SEO teams, the most interesting consequence is not whether the interface contains one label or another. It is whether the interface changes which parts of the generated response enter the user’s viewport.
An AI Overview can contain source links at different positions throughout the response. If only the compact opening section is visible, citations appearing deeper in the expanded answer have less opportunity to be seen. If Google automatically exposes the complete response, those same citations can become visible without an explicit expansion click.
That does not mean a citation will receive a click. It means the interface can alter the exposure opportunity around a citation independently of whether the underlying source selection changed.
Citation selection and citation exposure should be measured separately
This is an increasingly important distinction in AI-search analytics. A page can be selected as a source but appear in a part of the answer that few users reach. Conversely, a citation in the opening portion of an AI Overview can receive substantial visual exposure even when users never expand the response.
NetContentSEO has previously examined this broader measurement problem in its analysis of how being cited by an AI system is not necessarily the same as shaping the answer. The “Loading...” experiment adds another layer: even a correctly selected citation has an interface position and an exposure probability.
A useful measurement model therefore needs at least three distinct events: whether a source influenced or supported the generated answer, whether it received a visible citation, and whether the user was actually exposed to the part of the interface containing that citation.
The test could also push conventional results farther down the page
Automatically expanded AI Overviews consume more vertical space than compact ones. Reporting on Google’s August dynamic-expansion behavior noted that the fuller AI Mode-style response can push standard organic results farther below the fold.
The loading experiment potentially produces the same spatial effect. If the compact answer expands before the searcher begins scrolling, the first traditional organic result may move farther from the initial viewport even though its ranking position has not changed.
This creates another reason to avoid interpreting rank alone as visibility. Two queries can place a page at the same conventional organic position while presenting dramatically different amounts of AI-generated content above it.
The correct experiment would measure behavior, not screenshots
A screenshot can prove that the interface exists, but it cannot tell publishers whether the test is beneficial or harmful. The useful question is behavioral: does replacing “Show more” with “Loading...” increase the percentage of sessions in which users see the expanded answer?
A robust measurement framework would compare queries and sessions receiving the traditional control with those receiving the loading treatment, then record expansion completion, scroll depth, time before scrolling, citation visibility and outbound clicks. It would also need to control for answer length, query intent, device and the position of citations inside the response.
Without that data, claims that the new interface increases or reduces publisher traffic would be premature.
Publishers should monitor where citations appear inside the answer
Most AI-visibility monitoring focuses on whether a domain is cited. The interface experiments make citation position increasingly relevant. A citation that appears in the opening block, a source card near the top and a citation near the end of a long expanded answer are not equivalent exposure events.
Teams tracking important queries should therefore capture the full rendered AI Overview, not merely record the cited domains. For each citation, they can log whether it is visible before expansion, how far into the expanded answer it appears and whether the query receives compact, dynamically expanded or loading-state treatment.
Over time, this can reveal whether apparent citation gains are genuinely improvements in user-visible presence or simply appearances deeper inside increasingly long AI responses.
Search Console cannot yet answer the UI question directly
Google’s generative-AI reporting has improved, but publishers still do not receive the session-level interface data required to evaluate a test like this directly. The “Loading...” experiment concerns how an AI Overview is rendered and expanded, while the most useful publisher metric would be whether a particular citation became visible and was subsequently clicked.
That gap means third-party SERP capture and controlled testing remain necessary for detailed interface analysis. A publisher can observe changes in impressions and traffic, but attributing those changes specifically to a loading-state experiment requires much more granular evidence.
The limited rollout is itself part of the measurement challenge
The behavior is not universally reproducible. That makes it risky to build conclusions from a single account, location or query set. Google frequently runs interface experiments on subsets of users, and the same query can render differently depending on test assignment and other context.
Any monitoring project should therefore preserve screenshots or recordings, timestamps, device type, query, locale and whether the behavior was repeatable. Researchers should also distinguish a persistent treatment from a transient loading anomaly.
The current evidence supports describing “Loading...” as a limited UI test. It does not support saying Google has replaced “Show more” across AI Overviews.
A small interface experiment can change the economics of AI visibility
AI-search visibility is often discussed as if the decisive event were simply whether a model cites a page. Google’s experiment shows why that model is incomplete. The interface decides how much of the generated answer a user sees, where citations sit within it and how far conventional results are displaced.
If “Loading...” leads more users into the complete answer than “Show more,” deeper citations gain exposure while organic results below the AI Overview may lose above-the-fold visibility. If users interpret the loading state negatively or scroll away before expansion completes, the effect could move in the opposite direction.
Google has not supplied the data needed to choose between those outcomes. That uncertainty is precisely why the test is worth measuring.
The next AI Overview KPI may be citation viewability
The September test is small, limited and unconfirmed as a permanent product direction. Yet it points toward a metric publishers increasingly need: not just citation presence, but citation viewability.
As Google experiments with compact answers, dynamic expansion, AI Mode-style follow-ups and loading states, the same cited source can receive very different levels of real-world exposure depending on the interface around it.
For publishers and SEO teams, the practical response is to preserve query-level evidence and separate source selection from source exposure. The question is no longer only, “Did Google cite us?” It is also, “Did the user ever reach the part of the AI answer where that citation appeared?”