Recent reporting has described a further interface change in which certain AI Overviews can expand into a fuller response without a user pressing “Show more.” The available evidence does not establish the scale of that behavior, the countries or devices involved, or a dependable way to trigger it. It also does not conclusively show that an expanded overview automatically transfers someone into Google’s separately branded AI Mode. Those distinctions matter. A selectively expanding summary is not the same thing as a universal, unavoidable conversion of Search into AI Mode.
Even so, the direction of travel raises practical questions for Windows users, website owners, and policymakers. For users, the key issue is control over how results are displayed and where information comes from. For publishers, it is whether answers delivered before the link list reduce visits that once supported reporting, reviews, forums, and specialist documentation. For regulators, it is whether a dominant search platform can use web content to create its own answer layer without unfairly weakening the sources it relies on.
What is established — and what is not
Google introduced AI Overviews as a Search feature that can answer questions with a synthesized response and links. It began rolling the feature out broadly to U.S. users in May 2024. AI Mode, announced separately in March 2025, is a distinct experimental Search experience intended for more complex questions and follow-up exploration.
That product distinction should keep the discussion of expanded overviews grounded. An interface that presents a longer AI-generated answer and an “Ask anything” box may look and behave like AI Mode in important respects. But appearance alone does not prove that the user has entered AI Mode as a separate product. Treating the two as identical would overstate what is currently supported.
Likewise, it would be premature to say every Google query now expands into an AI answer, or that the expansion occurs regardless of what the searcher does. Reporting characterizes the behavior as selective and dynamically determined. The available dossier does not supply a public percentage of affected queries, a rollout timetable, country list, or platform list. It also does not establish a reproducible formula for getting an expanded result.
That uncertainty is more than a technical footnote. Search interfaces are frequently tested, personalized, and changed by query type. A screenshot of a long AI response on one Windows PC or Android phone cannot by itself show what all users will receive.
The control Windows users do and do not have
For people who want conventional search results, Google provides a meaningful but limited escape hatch: the Web filter. Google describes it as a way to show text-based links without features such as AI Overviews. In practice, that can be useful when researching a product, troubleshooting a Windows error, comparing sources, or looking for a forum thread rather than a summary of one.
The limitation is important: this is a filter, not a documented global preference that permanently turns off AI Overviews everywhere. Google also says disabling the relevant Search Labs experiment does not disable all AI Overviews. That makes the user-choice situation more constrained than it may initially seem. Someone can choose a link-focused view for a search session or query, but the standard Search experience can still present AI material by default.
For Windows users, a sensible approach is to treat AI Overviews as a starting point rather than the final authority, particularly for decisions where accuracy, context, or current details matter.
- Use the Web filter when you want source-first research. This is especially useful for driver problems, Windows recovery steps, software compatibility, security advisories, and product purchasing decisions.
- Open and read the underlying pages. A generated answer may compress caveats that are essential to the original publisher’s conclusion.
- Check dates and product versions. Windows features, support policies, hardware requirements, and software interfaces can change faster than a summary reveals.
- Be wary of one-step troubleshooting. Commands that alter the registry, boot configuration, encryption settings, or account permissions deserve verification from official documentation or multiple trustworthy technical sources.
- Search for disagreement, not only confirmation. If a summary presents a simple answer to a disputed question, use ordinary result links to find alternatives and primary material.
This is not an argument that a generated overview is always wrong or useless. For a quick definition, a basic calculation, or a broad orientation to an unfamiliar subject, it may save time. The risk rises when convenience replaces inspection of the source material—especially where the answer concerns security, health, law, money, or a destructive change to a PC.
Why the layout matters to publishers
The concern from publishers is not merely that an AI answer exists. It is that placement and completeness affect the incentive to click. When a response appears before ordinary blue links, provides a direct explanation, and invites a follow-up question within the same page, the user may have less reason to visit the sites that supplied the reporting or expertise behind the answer.
Observed browsing data from Pew Research Center gives this concern a measurable basis, while also showing the limits of what can be concluded. In a March 2025 analysis of browsing by 900 U.S. adults, users clicked traditional search-result links on 8% of visits where an AI summary appeared, compared with 15% of visits without one. Links cited inside an AI summary were clicked on 1% of visits containing a summary.
Those figures do not prove that AI Overviews alone caused every difference. Queries that receive AI summaries may differ systematically from those that do not; a straightforward fact lookup may naturally lead to fewer clicks than a complex research task. Nor do the figures quantify the effect of the later reported auto-expansion behavior specifically. But they are strong evidence against assuming that citations within a generated answer automatically deliver traffic comparable to a standard ranking.
A separate preregistered field-experiment preprint involving 1,100 participants reached a related conclusion: removing AI Overviews and AI Mode increased publisher click-through, while an AI-Mode-only condition reduced it. Because it is a preprint, the work should not be treated as a final regulatory finding or settled scientific consensus. Still, its experimental design makes it relevant evidence in a debate that too often relies on anecdotes from publishers or assurances from platforms.
Google has disputed the broader claim that its AI search features harm discovery, saying that people find the dynamic experience more helpful and that AI Overview links receive more clicks than comparable traditional listings. That claim and Pew’s results are not necessarily directly incompatible: they use different comparison groups and measures. A link could perform better than a particular “comparable” conventional placement while the presence of a summary coincides with fewer external clicks across the overall results page. What the public evidence does show is that the question cannot be settled with a single click statistic.
From traffic debate to competition scrutiny
The conflict has moved beyond product criticism. In September 2025, Penske Media and affiliated publishers sued Google and Alphabet in federal court in Washington, D.C., alleging antitrust and related harms connected to AI Overviews and Google’s use of publisher content.
The lawsuit is not a judicial finding that Google broke the law. Penske alleged that reduced Google referrals caused its organic affiliate revenue to fall by more than one-third by the end of 2024. That is a litigant’s allegation, not an established measurement of industry-wide impact or a proven causal result. Google sought dismissal, and no final ruling on that request had been reported by the end of August 2026.
Europe has also become a major venue for the dispute. A coalition including the Independent Publishers Alliance filed a complaint with the European Commission concerning AI Overviews in mid-2025. The Commission then opened a formal investigation in December 2025 into whether Google’s use of web-publisher and YouTube content for AI purposes breached EU competition rules. The investigation includes issues affecting AI Overviews and AI Mode.
A formal investigation is not a verdict. It does, however, mean the issue is no longer only about whether an individual reader likes generated summaries. Competition authorities are examining the structure of a system in which the company operating the primary discovery gateway can also summarize, retain, and monetize attention around material created by others.
The counterargument deserves serious consideration. Search has always evolved: snippets, knowledge panels, maps, calculators, product modules, and direct answers have all reduced clicks for some query types while making other searches quicker. Google can reasonably argue that a better answer page helps users and that people may still click when they want depth. The difficult policy question is where an improvement in convenience becomes an unfair extraction of value from sites that cannot realistically opt out of the answer layer without also sacrificing discoverability.
What automatic expansion could change
If selective automatic expansion becomes widespread, its significance would be less about one missing button press than about the default path it creates. A user who sees a compact overview must actively request more AI text. A user whose overview opens itself may be encouraged to remain in a conversational, on-page experience before encountering the ordinary web.
That could be helpful for multi-part questions and iterative research. It could also make the boundaries between Search, AI Overviews, and AI Mode less obvious to ordinary users. Clear labeling matters here. Users should be able to tell whether they are reading a generated summary, browsing independently ranked results, or using a conversational AI environment—and they should be able to move among those modes without friction.
The available evidence does not support a quantified claim that this reported expansion has already reduced publisher revenue or referrals by a specific amount. Any such figure would need direct measurement that separates the expansion from AI Overviews generally, AI Mode, changing search demand, rankings, and other traffic factors.
A practical standard for the next version of Search
The most useful test is not whether AI belongs in Search at all. It is whether users retain a real choice to search the web, whether sources remain visible and meaningfully reachable, and whether publishers can continue to fund the work that makes high-quality information available.
For users, the immediate answer is practical: use the Web filter when you want links first, inspect original sources for consequential tasks, and do not mistake a concise generated response for complete research. For site owners, the evidence suggests that measuring referrals and engagement by query type will be more informative than treating all AI appearances as identical. For regulators, the open questions are becoming more concrete: choice architecture, source attribution, data use, referral effects, and whether publishers have a viable way to participate in search without supplying material for competing AI answers.
Google’s expanding AI search presentation may improve convenience for some tasks. But convenience should not erase transparency or meaningful access to the open web. As the interface changes, the central question is whether Search still sends people outward when the web—not a summary of it—is where the value resides.