Back to overview

MOZOM-analyse

Google provides the answer: who decides what remains visible?

Type: Analyse Author: the editorial team Published: 6 augustus 2026 om 10:27 Report a correction
AI caricature of a huge search gateway ranking news while a large AI answer screen separates the audience from the original sources.
Source
Google, Pew Research Center, CMA, European Commission and scientific research
MOZOM headline
Google provides the answer: who decides what remains visible?
Original headline
From search engine to answer engine: the private rules behind news, sources and traffic
Author
the editorial team
Date
6 augustus 2026 om 10:27
Subject
MOZOM investigates how Google uses ranking, Preferred Sources, AI Overviews and AI Mode to increasingly determine which sources remain visible, which answers readers see first and how much traffic the original creators still receive.

Summary of the original report

Google recognizes that Google News is algorithmically influenced by relevance, prominence, authoritativeness, currency, usability, location and language. Personalization partly arises from interests that Google derives from activity on Google products and YouTube. Google says it will not accept payment for higher rankings and wants to promote original journalism and diverse perspectives. That's the official explanation. However, the scale makes these design choices publicly relevant. The British CMA found that more than 90 percent of British searches are done via Google. Google itself reported more than 2.5 billion monthly users of AI Overviews in May 2026. A search product with such reach does not need to ban a source to noticeably shift information traffic. In an American Pew measurement of 68,879 searches, users clicked on a regular result in 8 percent of visits with an AI summary, compared to 15 percent without a summary. Only 1 percent clicked on a source in the AI ​​summary itself. A May 2026 preprint examined 55,393 current queries. AI Overviews appeared on 13.7 percent of all queries, but on 64.7 percent of query queries. Nearly 30 percent of the domains cited were not among the simultaneously displayed results on the first page. Of 98,020 separately checked claims, 11 percent were not supported by the cited pages. That doesn't mean 11 percent was demonstrably false; often the necessary coverage or context was lacking. It does mean that a neat list of sources is no guarantee that the answer follows exactly from those sources.

What has been measured, and what does it prove or not?

The figures come from different countries, years and methods. They should not be added up as if they were one experiment. Together they show scale, selection and possible effects; they do not prove a secret party political mission.

Layer or studyMeasured resultWhat this supportsWhat this does not prove
Market power in the UK According to the CMA, more than 90 percent of UK general searches are done via Google. AI Overviews and AI Mode fall within the designated search activity. Google is not an ordinary competitor for citizens and publishers, but an almost inevitable gateway. According to the CMA, strategic market status is not in itself a finding of abuse or political direction.
Google Top Stories Audit 6,302 unique links from 2017 came from 678 sources; the largest 20 percent of sources received 86 percent of impressions and the top three received 23 percent. Attention and the associated traffic were highly concentrated among a limited group of major media outlets. The audit is old and could not determine whether SEO, publication volume, source quality or algorithmic bias was the cause.
Preferred Sources Google reports more than 345,000 unique sources chosen and an average of twice as many click-throughs to a user-chosen preferred source. Active source selection can give users and small publishers more countervailing power. It has not yet been determined whether well-known brands are chosen much more often due to their existing audience reach and thus receive a new flywheel.
Pew click measurement Across 68,879 searches, clicks on a regular result were 8 percent with an AI summary and 15 percent without; an AI source received a click in 1 percent of the cases. An answer above the links is associated with significantly less visits to the original sources. The American observational measurement from March 2025 is not a random experiment and does not measure each publisher separately.
AI claims and sources A 2026 preprint analyzed 55,393 queries and 98,020 verifiable statements; 11 percent were not supported by the pages cited. Citing sources alone does not make a composite AI response verifiable or complete. Unsupported is not the same as proven false; the study is a preprint and was still under review.
Wikipedia traffic A second preprint compared 161,382 article-language pairs and estimated approximately 15 percent less daily traffic to English Wikipedia articles after AI Overview exposure. There is early causal evidence that AI responses can take attention away from an original source of information. Wikipedia is not your average news publisher; the effect varied by subject and cannot be applied one-to-one to all media.
Reddit as evidence to the contrary A preprint found 12.0 percent more comments and 12.4 percent more responding users in AI Overviews-enabled Reddit groups; after the introduction of AI Mode, a large part of that profit disappeared. Mention by AI can also initially generate additional participation for experiential content. The effect strongly depends on content and interface; So 'AI always reduces traffic' is too absolute.
UK Publishers Rule 2026 Google must provide publishers with effective AI controls, clear explanations, engagement metrics, proper attribution, and a clear route to the source. A regulator recognizes that resource use, visibility and measurability with a dominant search engine can pose a competitive problem. The measure does not automatically prevent source concentration and is not a judicial decision that Google manipulated politics.

From ranked links to a platform-generated answer

The decisive change is functional. The classic search engine placed sources in an order and ultimately let the reader read elsewhere. The AI answer engine reads, selects, combines and presents the outcome within the same commercial system. This creates three gates. Gate one is ranking: prominence, authority, topicality, location and language determine who becomes visible. Gate two is preference: the user can highlight resources, but must already know about them and actively choose them. Gate three is synthesis: Google determines which statements end up in the short answer, which context is omitted and which references are sufficiently visible. A source can even be cited at gate three without appearing on the first results page. According to the 2026 study, this indicates a separate selection mechanism in addition to the normal ranking. The social impact does not therefore have to come from a political assignment. Product goals such as speed, ease of use, safety, authority and answering the question automatically determine which material is given priority. Because Google can continue to show advertisements on the same search page while outgoing visitors decrease, there is also a commercial interest in this. That's not evidence that Google is deliberately taking traffic from publishers. It is a structure of interests that justifies independent control.

The agenda may be in the rules of the game

Google doesn't need to say which political opinion is the right one to influence the public information diet. Whoever determines which sources are authoritative, which publications are highly cited, which safety limits apply and which answer appears first, also determines which issues appear normal, questionable or marginal. In communication science, this is agenda setting: not necessarily prescribing what people should think, but taking into account what they think about and from what sources. The legally safe conclusion is therefore not that Google is demonstrably carrying out a secret political agenda. The sharper conclusion is that Google's product choices themselves can have agenda-setting consequences and that, due to its scale, the company exercises almost institutional power without the publicity obligations of a public newsroom.

Counter-evidence that belongs in an honest analysis

Not every study points to a closed filter bubble. The peer-reviewed study Google News and Machine Gatekeepers found little evidence for personalization in the United Kingdom, putting the classic filter bubble claim into perspective. Google says it will not accept payment for higher Google News rankings, wants to show original journalism and different perspectives and is building new source carousels, Highly Cited labels and Preferred Sources. The Reddit study also shows that visibility in an AI response can increase participation under some circumstances. Also, source concentration is not exclusively a Google product. Large media publish more, are cited more often, have stronger brands, better technology and more resources to follow current events. Yet the power problem remains. The fact that such differences begin outside Google does not absolve the dominant ranker from the question of whether its systems reduce or further amplify them.

Possible message behind the news

Google organizes the web for your convenience. The critical counter-reading is that convenience here can also mean obedience to one information architecture.

Neutral conclusion

Google is not a neutral conduit. This already follows from its own ranking documentation and becomes more important now that the same company also writes the answer. The figures show source concentration, fewer clicks on AI summaries, a measurable traffic effect on Wikipedia and an error-prone step between source and statement. At the same time, Preferred Sources and the Reddit study show that user choice and positive effects are possible. A secret political agenda has not been proven. But Google does have agenda-setting power: the company determines the private rules under which sources are found, weighed, summarized and sometimes skipped. Google doesn't have to remove an article to reduce its influence. It may be sufficient that the answer comes before the source.

Source:

Related analyses