The impact of Google AI Overview on the Internet’s digital ecosystem

AI and SEO is a big topic for me personally, so I felt it was my duty to write this article. I’ll warn you in advance that the article is roughly 47 standard pages long. But I think it’s important, because SGE will soon be live here in the Czech Republic.
For those who don’t want to read the whole piece—although it’s nicely sourced—based on feedback I’m adding a very brief summary, an audio version, and a mind-map here.

In my view the key question is whether to block Google from training AI on the content of my own and my clients’ websites.
Is it enough to add the following to robots.txt?

Update, Google ignores this and this me)
User-agent: Google-Extended
Disallow: /

Update [2025-05-01] Google ignores this setting within SGE, and the restriction applies only to model training (Bard, Vertex AI, Gemini). So in reality you either cut Google off completely, or you cannot prevent the model from learning from your data; Google gives you no choice if you rely on organic Google traffic.

For the other AI crawlers I’m still leaving everything allowed.
The decision is up to you.
Feel free to message me your views in PM or under posts with this article on social media.

Short summary of the article

Google is launching AI Overview (formerly SGE), a feature that uses generative AI to show consolidated answers directly in search results. Instead of clicking links, users get a quick overview of information from multiple sources. The goal is to speed up searching under the slogan “The search engine does the work for you.” The feature went live in the US in May 2024; global rollout is planned. Last week SGE arrived in Europe; it’s not yet in the Czech Republic (Update: it is now 🙂).

What does this mean for users?

Advantages: Faster answers, convenience, ability to handle complex queries at once, conversational mode for follow-up questions.
Disadvantages: Fewer visits to original sites, risk of an information bubble (one AI summary instead of multiple viewpoints), possible AI errors or distortions users may not notice.

Impact on content creators and websites

Threat: A sharp drop in traffic (estimates 20–60 %), threatening ad revenue and content funding—Google effectively uses creators’ content to compete with them.
Monetization: Google is already testing ads inside AI answers, letting it earn even without a click to the creator’s site. That shifts web economics toward the platform.
Visibility & SEO: AI summaries take top position, pushing organic results down. Classic SEO loses some weight. AIO (AI Optimization) and GEO (Generative Experience Optimization) emerge—optimizing for AI with emphasis on quality, authority (E-E-A-T) and structured data. The aim is not just to rank first, but to become a trusted source for the AI.

Do we need to panic? Not quite.

Real-world impact: Early data show smaller effects than feared. Google is rolling the feature out cautiously; AI still has limits (errors, bias), especially on sensitive topics (health, finance) (update: in CZ pharmacies and finance got hit hard). (Update with image: uptake is gradual but still rising steadily—less and less traffic for websites)

Dependence on Google: Google still needs quality web content as AI training data. It can’t afford to destroy the whole ecosystem and is seeking balance, negotiating with some publishers.
Regulation & competition: Google faces regulatory pressure (EU, US) and potential copyright suits. Competition (Bing, Apple?) isn’t yet strong, but it’s there.

Opportunities and the road ahead

Quality over quantity: AI may squeeze out superficial content. Creators must focus on unique, deep, expert content (E-E-A-T) that AI can’t replace.
New revenue models: Diversify—subscriptions, direct support, newsletters, communities, e-commerce, licensing fees for AI use of content.
Adaptation: Leverage AIO/GEO, optimize technically (structured data) and editorially, lean on multimedia (video, audio) and build a direct bond with the audience.

Summary conclusion

AI Overview is a game-changer. It carries risks for the current web model but opportunities for those who adapt. The future hinges on finding a new balance among AI, content creators and users. The keys are quality, adaptation and diversification. The open web isn’t ending—just evolving.

Audio version of the article from Google NotebookLM

(26 min – English)

Mind-map of the content:

Main article full of details

Introduction to SGE

In 2023–2024 Google began a major transformation of search by integrating generative AI. The feature called Google AI Overview (formerly the Search Generative Experience, SGE) experimentally started serving users summary AI answers right on the search results page. Instead of the traditional ten blue links, users now often see an automatically created “overview” of information combining insights from multiple sources. The goal is faster answers—Google promotes the novelty with the slogan “Search will do the work for you with AI Overviews.” In the US, AI Overview went fully live in May 2024, and Google plans to expand it to billions worldwide during the year.
The innovation comes with both great expectations and concern. On the plus side it pledges a more convenient search experience; on the minus side it raises key questions around impact on content creators, web monetization, and the future of SEO.

How AI Overview reshapes the digital ecosystem – Impact on user experience

Example of Google AI Overview on mobile: the generative answer to a practical query (how to remove grass stains) contains concise steps and cited sources (“How to Get Grass Stains Out of Jeans” by Duer, “How to Get Grass Stains Out” by Maytag) followed by a Sponsored section with relevant products (cleaning agents). A note at the bottom warns that this is an experimental generative-AI feature.

For ordinary users, AI Overview delivers a more convenient path to information. Instead of opening several pages, they often get a direct answer right on the results page. Google essentially “does the googling for you”, searching and processing information in the background. A summarized AI extract appears above search results, merging data from multiple sources. For a complex query with several criteria (e.g. “best restaurant for a family with kids and a dog near the Empire State Building with no long wait”), SGE produces a unique consolidated snapshot sourcing several sites at once—giving the user a single comprehensive overview rather than a link list they must piece together.

Besides one-off snapshots, Google AI Overview offers a conversational mode; users can ask follow-up questions and the AI responds in context, turning search into an interactive dialogue. The result is a smoother, more personalized experience. Sundar Pichai, Google’s CEO, says generative AI lets Google meet “deeper and broader information needs,” enabling even nuanced complex questions to be answered “all in one.” Google’s internal data show early users like receiving a quick overview plus links for deeper study, and that with AI Overviews people actually search more and are happier with results.

The big change for users is speed and convenience. AI Overview replies in seconds (Gemini upgrades halved response times since the May launch), saving the clicks normally spent on multiple pages. It shines on multi-step queries; AI handles the reasoning and presents a full answer. Google frames it as “taking the work out of searching,” serving users a quick essence of knowledge.

But there’s a downside: users no longer need to visit the target sites for answers. Reading the AI summary may suffice, letting them move on without seeing the creators’ pages. They miss context, depth, and varied perspectives. In the extreme, AI Overview can become the sole voice summarizing internet consensus; users may not sense diversity of views. And as this article notes, answers can be wrong or biased and users might not notice because, unlike classic search, they aren’t comparing multiple results. It’s a trade-off: dramatic convenience and speed at the cost of direct open-web interaction mediated by an AI layer.

Impact on content monetization and the web economy

The new search model has deep consequences for online-content economics. Digital content is largely funded indirectly via traffic and ads. Users consume articles, tutorials, or reviews free while sites monetize visits with ad impressions or affiliate fees. If AI Overview gives the answer without the click, the chain breaks—creators lose visitors and revenue. Adtech firm Raptive says Google’s AI overview “uses your work to compete with you.” Content is taken and shown directly, pushing source links off-screen. That can be an existential blow.

Early analyses forecast severe impacts. Raptive data shown at Google I/O 2024 suggest SGE could cut search traffic up to 66 %. Other estimates range 20–60 %. eMarketer says such a fall could cost publishers up to \$2 billion a year in ad revenue. Category breakdowns show, for example, ~20 % hits to food sites, ~29 % to travel/family niches. A dramatic audience loss slashes ad income and can threaten survival; Raptive’s CEO warned of “massive damage to the internet as we know it.”

Meanwhile Google itself likely benefits. The company is already monetizing AI answers with sponsored elements. In October 2024 Google announced ads are part of AI Overview and started testing paid links inside it. Publishers may lose visits, yet Google still earns. With ~80 % of Alphabet revenue from search ads, Google will protect ad success in SGE.

In short, AI-generated answers shift web economics toward the platform. Users get faster answers, but the value created by quality content remains locked within Google. Creators provide data yet gain no reward, prompting fears that AI Overview is a “devastating” tool sideling trustworthy human creators.

Impact on site visibility and the role of SEO

AI Overview also upends site visibility and the nature of SEO. Even before generative AI, Google pushed direct answers; now the AI summary takes prime SERP real estate, often above traditional links. SGE is dubbed “snippets on steroids.” Click-through rates drop as users see the answer first; on mobile the snapshot can fill the screen.

Traditional SEO—aiming to rank high—now offers less payoff if Google answers the query itself. Visibility shrinks; users may not need the website.
SEO pros thus ask how to optimize for AI answers. New terms arise: AIO (AI Optimization) and GEO (Generative Experience Optimization). Google seems to weigh authority, relevance, and quality even more heavily, relying on E-E-A-T. Big established sites often appear in overviews; authoritative content is preferred. Building reputation, expertise, and trust is key.

Some old tactics fade. Chasing long-tail keywords or creating pages for every variant may be useless if AI merges similar queries into one answer. Keyword stuffing offers no help; the model looks at meaning, not literal matches. Even backlinks play a subtler role—useful for trust signals but invisible to users if AI silently leverages them.

SEO therefore evolves into optimizing content for AI. AI-first optimization means structuring content for machine understanding: structured data (schema.org), clear headings, summaries. Some creators now add key takeaways up front for both readers and AI. GEO goes further, focusing on overall digital experience—interactive, personalized content suited for multimodal, conversational search.

For visibility, those deemed trustworthy and included in AI answers keep at least indirect exposure. Sites with shallow content may vanish. SEO becomes interdisciplinary, blending classic optimization, content strategy, data analytics, and even PR. Experts advise thinking beyond “Search Engine” toward the “Search Experience”: the goal is not merely ranking first but being part of the best answer, whether delivered by a web page or an AI generator.

For clarity, let’s summarise the key changes in the impact of AI Overview on the main players of the digital ecosystem:

Stakeholder	Positive changes with AI Overview	Negative changes with AI Overview
Users	Faster answers without having to click through multiple pages; the ability to refine a query conversationally (greater comfort). AI can take more factors into account (personalisation), so the search better matches specific needs.	Less exposure to various opinions and sources (narrower information bubble); risk of AI error or outdated data – the user may mistakenly trust a single answer. Less contact with the open web (the user stays within Google’s environment).
Content creators	High-quality content can gain source status in AI answers (indirect recognition of authority). For complex queries, AI points to several different websites, giving users a broader palette of pages than if they clicked just one link.	Significant drop in organic search traffic (by tens of percent) = decline in ad revenue and a threat to the financial sustainability of content. Lower brand visibility – the reader often does not know where the information comes from unless they explicitly click the source.
Digital marketing (advertisers, SEO)	Potential for new ad formats directly in the generative SERP (less competition in classic positions, but the chance to be part of the AI answer). Users who do click through from AI are probably more decided and informed = more qualified leads with higher conversion rates.	Fewer organic impressions = less room for SEO results, the need to rethink optimisation strategies (new costs for AIO / GEO). Measurement uncertainty – it is difficult to determine how much the AI answer contributes to drops or changes in user behaviour, complicating attribution models.
Google (platform)	Keeping the user in its own ecosystem longer (higher engagement); the ability to offer innovative functionality and fend off search competitors (retain market share). New monetisation opportunities (ads in AI answers, shopping recommendations) without mediating via third-party websites.	Criticism from the industry and content creators for disrupting the open web; risk of regulatory intervention or lawsuits (copyright issues around AI training, etc.). Without quality web content, the AI would have nothing to draw on – Google must be careful not to “saw off the branch it’s sitting on”.

Potential exaggeration of concerns

(Update: In reality, the fears mainly materialised with the arrival of AI Mode.)

Despite alarming predictions, it is worth asking whether some concerns about AI Overviews are somewhat exaggerated – at least in the short term. The first months of SGE operation did not fulfil the darkest scenarios. For example, in Q2 2024 major publishers such as The New York Times and the media house IAC (owner of Dotdash Meredith) reported that the impact of AI Overviews on their traffic was so far negligible. Dotdash found that AI answers appeared in only 15 % of their largest search-query categories and the traffic drop was virtually zero. Similarly, Ziff Davis (owner of tech sites like PCMag or IGN) stated that AI Overviews showed up for only 8 % of the queries important to them and so far they do not perceive it as a significant change in the search experience. The New York Times did not even mention SGE’s impact in its quarterly report and instead reported a 41 % increase in operating profit and gained 300,000 new digital subscribers in the same quarter – suggesting it did not feel an immediate revenue threat from this new feature.

These reports indicate that during SGE’s pilot phase its reach was limited. After a shaky start (see the next section on technology limits), Google deployed AI Overviews cautiously – according to data from BrightEdge, generative answers appeared for about 15 % of queries in May 2024. In addition, it initially skipped certain sensitive areas (e.g. health, finance) and query types where AI could err or where it already had specialised boxes (such as recipes or local businesses). The scale of impact was therefore not blanket-wide and many users did not encounter AI Overviews at all if they were not in an included location or did not search in a specific way. That explains why some publishers reported “no significant change”.

Hence, at least in the early phase certain fears may have been overstated. Speculation about a 60 % traffic drop haunted publishers even before launch, but the first quarter of operation confirmed nothing of the sort. It is important to add, however: for now. In August 2024 eMarketer analysts warned that the initial calm does not mean the long-term threat has vanished – AI Overviews are only at the beginning and Google has clearly stated that it “believes this feature is the future of search and will gradually expand it”. In other words, we may be witnessing the calm before the storm. The fact that the first few months did not bring large declines may simply be due to SGE’s limited availability. Once Google gains confidence and switches on AI Overviews for a larger share of queries or more countries, the effects may become apparent.

For now, however, some strong sites have managed to thrive despite SGE. The New York Times, which has diversified revenue (not only advertising, but also subscriptions, events, etc.), is to some extent insulated from the direct impacts of search-traffic declines. A strong brand and loyal audience provide a cushion. This suggests that the effects may not distribute evenly – larger, established players can remain relatively fine, while smaller publishers, more dependent on organic search, will feel the changes more painfully. In any case, it is necessary to keep a critical distance: AI in search is still experimental and, as Q2 2024 data show, the worst-case predictions do not always come true. The concerns therefore may not be immediately realised, even though their underlying structural shift is real.

Limits of the technology and AI bias

Another important counter-argument is the fact that generative AI technology has its limits and is not infallible. Right at the launch of SGE in May 2023, it became clear that AI can hallucinate incorrect or even dangerous answers. A well-known example from early tests was advice to “eat rocks and stare at the sun” in the context of some query – an obviously undesirable and potentially harmful output. These missteps sparked criticism and forced Google to slow the rollout and monitor answers more closely. All AI Overviews also feature a prominent warning such as “Generative AI is experimental”, indicating users should take the answers with caution. All of this illustrates that AI is not error-free.

AI bias is another issue. Large language models generate answers based on the probability of word continuation learned from massive data. They can therefore inherit and amplify systemic errors or prejudices contained in the training data. For example, if a minority perspective is under-represented on the internet, AI may omit it in a summary – inadvertently reinforcing a single viewpoint. Or, if factual errors exist in the data, the model can repeat them. Although Google combines the LLM with its current search index, SGE answers have still been observed containing factual inaccuracies or outdated information. Technical limitations are real: the model does not know what it does not know and cannot always reliably judge the truthfulness of its statements.

These shortcomings limit SGE’s usability in certain areas. Google purposely dampened generative answers for queries where errors could have serious consequences – for example medical advice, legal information, financial counselling, etc. For such queries AI Overviews either do not trigger at all, or they produce very cautious answers with a referral to consult a professional. This means that in many important segments (YMYL – Your Money, Your Life) classic results still dominate and the role of websites remains indispensable. The technology simply is not ready to take over all content.

Besides bias, there is also the question of recency and context. AI models (such as GPT-4, on which SGE is apparently based, supplemented with Google data) do not have direct access to the real world at query time – they draw from trained knowledge and current web indexation. But if a completely new event occurs (breaking news) or a truly up-to-date figure is needed (e.g. today’s stock price, yesterday’s match result), generative AI can fail or only take data from the index without deeper context. In practice, for such cases Google rather shows a traditional news card or a live result than an AI summary, because the model is not designed for real-time information. AI limitations therefore enforce that the good old open web must still be at hand.

Another limit is understanding language nuances and tone. An AI snapshot may sometimes come across as too confident, even when it should present a range of opinions or uncertainty. For controversial questions (vaccines, political events) AI might – if it summarises the majority view – suppress minority voices, even though they may be relevant. Google therefore claims to be working on an option to “adjust your AI Overview”, where the user gets a choice to simplify or expand the answer. This is an interesting improvement, but again it shows that one generic answer may not fit everyone or everything.

In terms of user trust, AI mistakes are crucial: if users repeatedly encounter obvious errors or bias, they may turn away from generative results and return to the traditional style of searching (multiple sources, their own judgement). Google must balance innovation and reliability. That is why it is still testing SGE in stages and collecting feedback. It is fair to say that AI Overviews are, for Google, still an “early days” technology. Improvements can be expected (Pichai signalled the arrival of the multimodal Gemini model, which could further enhance AI capabilities), but it is also possible that some limitations (especially bias) will remain a perpetual struggle.

Overall, the technological imperfections of AI represent a brake on the full deployment of SGE. Google cannot afford to disappoint users with utterly nonsensical answers – it would endanger the reputation built over 25 years that Google provides reliable results. Until it is sure of quality, it will deploy AI Overviews cautiously and with restrictions. This gives content publishers some time and space to adapt and the open web a chance to breathe before any total “take-over” by an AI system (if that ever happens). AI is powerful, but it is not omnipotent – and that is an important counterargument to fatalistic visions of traffic extinction.

Google’s dependence on quality content

Although it may sound paradoxical, Google – even with all its AI autonomy – still critically needs quality content from creators. The search engine has always operated as an intermediary, indexing third-party content and helping users find it. Now, while the role of intermediary is shifting to “I’ll answer myself”, AI answers have no other source than web content itself. Google therefore faces a dilemma: if its AI outputs were to seriously endanger the functioning of the open web, it would deprive itself of fuel. This cannibalism has its limits. Google’s leadership publicly admits this – Sundar Pichai stated: “We will continue to prioritise approaches that send valuable traffic and support a healthy, open web.” It presents SGE as an “on-ramp to explore the web”, not a closed goal in itself. These statements, some critics say, are “just words without concrete metrics”, but they signal that Google feels the need to reassure the ecosystem that it doesn’t want to destroy it.

After all, Google faced a similar situation when introducing featured snippets. There were concerns then that direct answers (definitions, short facts) would reduce clicks. Google argued that snippets serve as an invitation to deeper exploration, not a replacement for entire content. To some extent that held – many snippets answered simple queries quickly, but for detail the user still had to click. Now the situation is more serious because AI can deliver more extensive answers. Nevertheless, Google claims that in its experiments people visit a more diverse range of websites thanks to AI Overviews and that “links included in AI Overviews get more clicks than if the page appeared as a traditional result for that query.” This is an interesting insight – it suggests that, for complex queries where a user might not even know which link to click, AI offers specific source recommendations. That can put some websites in the spotlight and even bring them more users than they would get through classic ranking. In other words, Google is trying to prove that AI Overviews are not a black hole for traffic, but potentially a curator that can help websites find their target audience. Whether this is true on a large scale is not yet definitively proven – the data come directly from Google and lack independent verification – but at least it shows that Google is actively looking for ways to maintain traffic flow to the web.

Further argument can be seen in the steps towards cooperation with publishers. If Google really planned to ignore creators’ interests, it would not make conciliatory gestures. Yet we see Google and other AI players starting to strike deals with content providers. For example, OpenAI (the maker of ChatGPT) has partnered with the Associated Press to license content for AI training. Likewise, Google faces pressure from large media houses – a group of U.S. publishers (including the NYT) is reportedly preparing a billion-dollar lawsuit over using their content to train AI without a licence. Google is also directly sued by groups of authors (e.g. writers) for copying their works into training data. These legal threats pressure Google to change its approach or negotiate. Publishers openly say: “We need Google to take creators into account.” And Google has, since SGE’s pilot phase, made certain adjustments in favour of creators – Raptive observed that Google shortened generated answers (to leave room for links), began displaying more prominent links to sources, and was more cautious in some categories (e.g. recipes). This shows that feedback from publishers has at least partly been heard and Google is fine-tuning settings for AI Overviews to mitigate negative impacts.

Ultimately there is a broader argument: Google long-term needs quality web content to fulfil its mission “to organise the world’s information.” If SGE dramatically reduced the creation of freely available content (because it became unsustainable), Google would soon hit the ceiling of what its AI can do. AI is no magic fountain of wisdom – it is only as good as the data it is trained on and indexes. Therefore, Google is to some extent bound to a symbiosis with the content ecosystem. This realisation gives creators some leverage: they can push for better terms (e.g. a share of AI ad revenue, better source visibility), because Google knows it needs them. Sure, Google has immense power and resources – perhaps it could survive with a limited open web and rely more on its Knowledge Graph or licensed data – but that would be a step backwards. So far everything suggests that Google wants to “have its cake and eat it”: benefit from AI innovation while also keeping creators alive. For content creators this means they are not powerless. They can join forces (like the mentioned group of publishers) and demand respect for their rights and interests. And Google will have to listen to some extent if it wants the long-term sustainability of its search.

Regulatory and competitive push-back

A critical look at the impacts of AI in search would not be complete without mentioning the regulatory and competitive factors that can slow down or steer Google. Google’s dominance in search already attracts the attention of antitrust authorities – in the U.S. a major antitrust trial is under way regarding Google’s monopoly position in search (though that concerns exclusive deals rather than SGE directly). The European Union has advanced legislation to protect publishers in recent years (copyright directive, e.g. the “link tax” for Google News) and is generally very active in tech regulation (DMA, DSA and the upcoming AI Act). Regulators are likely to examine the effects of generative search on the market and society. If data show that Google significantly harms the media industry or exploits content without compensation, we may see moves to remedy this. This could mean legislation requiring AI profit-sharing with publishers or rules to ensure transparency and mandatory source attribution. Some publishers already argue that AI answers are derivative works that should fall under the original creators’ copyright – and that Google should pay licence fees similar to how radio stations pay composers for public performance of music. These legal questions are not yet resolved, but the pressure is growing.

The mentioned threat of a collective lawsuit by publishers in the U.S. (NYT & co.) worth $1 billion shows that big players are willing to go to direct confrontation. Even if the court route fails, it sends a signal to lawmakers. Similarly, authors’ lawsuits over training AI on their books without permission could set a precedent for website text. Regulators therefore have levers to force Google to adapt: from fines and binding codes of conduct to the extreme possibility of separating certain services (though that is not currently on the table specifically for AI, rather for monopoly issues in general).

Besides regulators, there is also competition, although currently limited. Microsoft with its Bing Chat was actually first to integrate GPT-4 into search (February 2023) and is trying to grab market share with innovative features. However, as The Decoder aptly noted, “even the crazy hype around ChatGPT did nothing to make people use Bing.” Bing’s share remained small (around 3–4 %) even after AI deployment – people are used to Google and did not migrate en masse. This gives Google some freedom to experiment without risking user flight. Still, competition exists: besides Bing, start-ups like Neeva (now defunct), You.com or Perplexity AI (my personal favourite) tested new search models. Most failed to gain significant audiences. If, however, Google fails in users’ eyes (e.g. AI frequently erring or users disagreeing with how Google treats sources), an opportunity opens for a competitor with a proclaimedly more ethical or community-driven approach. At the very least Apple is on the horizon – it has long been rumoured that Apple is developing its own search engine. Apple emphasises privacy and might launch AI-powered search that respects sources and user privacy more. If that happened, Google would face not only regulatory but also market pressure to do better.

Also worth mentioning is the competitive environment within the information ecosystem: platforms such as Meta, Amazon or TikTok. Each has its own search (many young people today search directly on TikTok or Instagram instead of Google). These platforms do not operate on the open web, but they compete for time and attention. If Google disappoints content creators, more content might go exclusively to other platforms. For example, publishers might decide to invest more in social-media content or build communities outside Google’s reach. That would indirectly harm Google, as it would have less quality content to index. Competitive push-back thus works in that creators have alternative channels to which they can shift their efforts. Google is not the only gateway to an audience – it is very influential, but a creator can bet on other paths (with different compromises).

Regulation and competition together create an environment that can force Google to be more considerate. We already see hints: Google says SGE “will be how search works”, but also that it “will send valuable traffic and support a healthy web” – language tailored to deflect allegations of monopolistic behaviour. Google knows that if it sinks its partners (publishers) completely, it risks reactions from Washington, Brussels and the public. Likewise, Google’s effort to negotiate with selected publishers about licensing content (rumoured programmes where Google would pay media for AI content, similar to Google News Showcase) signals it wants peaceful coexistence and to pre-empt harsh regulation.

Competition for Google is not an immediate threat in general search, but could emerge in a specialised form – perhaps a search engine that guarantees open-web principles. In the open-source community there are projects like Haystack or Marginalia that experiment with alternative approaches (e.g. non-commercial search favouring small websites). These are not mass-market, but show that some people are actively looking for an alternative to Google’s hegemony. Google therefore feels the wind at its back in the sense: “Watch out you don’t overdo it, or else…”. And that is good for the ecosystem – this counter-pressure can help ensure that the development of AI in search proceeds more responsibly and with regard to wider impacts.

New angles and opportunities

Despite all the question marks around AI Overviews, there are also positive perspectives and new opportunities that this shift could bring. Crises and changes often give rise to innovation and qualitative improvements. In this chapter we focus on several viewpoints that offer constructive visions: how generative search could lead to better content, what new monetisation and distribution models are emerging, and what adaptation strategies (including AIO, GEO and multimedia) actors can use to their advantage.

Qualitative shift in content

One paradoxical possible impact of generative search could be a general qualitative rise of online content. Historically, optimisation for search engines sometimes tempted creators towards a quantity-driven, utilitarian approach: producing as many articles as possible focused on popular keywords, often at the expense of depth or originality, just so the site covered every topic people search for. This led to the rise of content farms or many similar “How-to” articles recycling the same information. If AI now takes on the role of answering the most generic queries, it could paradoxically free creators from the race for quantity. The content business may have to shift from quantity to quality.

Imagine AI Overview reliably covering basic queries like “how to boil an egg hard”. The operator of a recipe website would no longer get much traffic from such a query, so writing the twentieth article on the same topic with minor variations would make no sense. Instead, they can focus on unique content that AI cannot fully replace – for example original recipes, personal food stories, video tutorials, niche cuisines, etc. E-E-A-T (experience, expertise, authority, trustworthiness) will be a key guide for creators: the more of their unique know-how and perspective they put into content, the better the chance users find and value it (and perhaps AI prefers it as a source). Thin content generated just for SEO positions will lose its purpose, because AI will commoditise it and users have no reason to read the original.

This could lead to a shift from “SEO-driven” to “user-driven” content. Creators will have to ask: what truly brings value to my audience? They may listen more to readers, engage them and build a community around their content. Content may become less about pleasing the search engine and more about genuine informational value and experience. For example, instead of dozens of average articles, a site might publish fewer but richer pieces – and rather become an authority in a particular field. Authorities and established brands are likely to profit from this shift: they will be the sources AI cites (because it algorithmically trusts them) and that users seek when they want something beyond a generic summary.

In addition, creativity in form may flourish. If plain text no longer attracts the masses as before, creators can experiment with form – interactive stories, data journalism with visualisations, series and narratives, deeper analytical essays for demanding audiences. These users may then come directly (bookmarks, newsletters) instead of through the search engine. Thus arises a kind of renaissance of traditional creation of valuable content, driven by the fact that added value is the only way to justify the existence of an article alongside an AI summary.

There is also the role of AI as a tool for creators – generative models can help in the creation process itself (drafting text, transcribing interviews, summarising research). That could save creators time on routine parts of the job and allow them to devote more capacity to analysis, verification and creative tasks. The result could be better-processed content – AI as an assistant, editor or inspiration, while the human delivers the final quality and uniqueness.

In short, generative search may push ballast off the web and highlight the importance of truly quality content. For users this would be beneficial – even if they use AI for a quick answer, when needed they still have rich, deep content from human creators. And for creators who can adapt, it can be a chance to stand out, because competition from “me-too sites” will weaken. Of course this is an optimistic scenario and it will depend on whether a sustainable financial model for this qualitative shift emerges, which relates to the next point.

New monetisation and distribution models

If generative AI disrupts the existing model largely based on ads from traffic, the ecosystem’s logical response will be seeking new monetisation models. Even before SGE’s arrival many publishers diversified revenues – for example by building paywalls and subscriptions. The New York Times is a shining example: digital subscriptions became its main revenue source and thanks to that it is relatively resilient to advertising market declines. Generative search can accelerate this trend. If earning from anonymous Google traffic becomes harder, motivation will grow to convert part of the audience into paying customers. This may take the form of a full paywall (content behind login), a freemium model (some content free, the rest paid), or memberships where paying fans get extra perks (bonus content, community, merchandise, etc.).

Another path is micropayments or tipping systems – e.g. platforms like Patreon, Ko-fi, Pickey in the Czech Republic, where readers can support creators directly. If ad revenue drops massively, even larger media houses may turn to direct reader support (similar to Wikipedia or some non-profit newsrooms like NPR or ProPublica with fundraising).

This ties into changes in content distribution: creators will want a more direct relationship with their audience so they aren’t dependent on middlemen. A newsletter renaissance is expected – an email newsletter delivers content straight to fans and bypasses algorithms (NYTimes reports millions of newsletter subscribers). Likewise podcasts and RSS feeds may surge again – channels Google doesn’t control that provide direct creator-audience connections. Publishers also invest in mobile apps and notifications to bring users directly. Analysts advise publishers to rely less on one “gatekeeper” and diversify traffic sources – i.e. drive readers through social networks, referrals, mailing, mobile push, SEO on other platforms (YouTube search, Bing, Seznam in Czechia, etc.). The more diverse the traffic mix, the better a site is protected when one channel (Google) monopolises visits.

New monetisation models may also relate directly to AI: the idea of licence payments from AI companies is emerging. For example, a consortium of big publishers could negotiate with Google or OpenAI for flat fees for using their content in training or answers. Something similar already works in music (Spotify pays labels) or news (Google News Showcase pays some media for news). If this became common, it would be a new revenue stream for content creators – essentially a data-licensing economy. It is not reality for most sites yet, but pressure is growing.

Some media also try alternative business models: organising conferences, workshops, selling data or analytical reports, running e-shops with merchandise, etc. EMarketer recommends publishers to add, for instance, job boards, events or experiential activities, because relying on the comeback of good old display ads in full strength is naïve. Thus media companies become more diversified businesses.

An important new area is also AI as a service: some sites may deploy their own AI chatbot that users subscribe to as an expert assistant trained on that site’s content. For instance, a specialist magazine could offer paid access to AI that answers questions based on its article archive. This monetises content indirectly – not through ads, but through a value-added service. Today we already see tools such as Vault AI (a chatbot trained on a company’s own data). If publishers deployed something similar publicly, they could compete with generic SGE and charge a fee.

Content distribution may also shift to a multi-platform strategy. Creators will want audiences to find them where generative AI does not reach. For example, video platforms: Google does show videos, but AI can hardly replace the full value of video content (not everyone is satisfied with a text transcript). Websites may therefore invest more in video production (YouTube, TikTok), audio production (podcasts, audiobooks), interactive formats (mini-apps). The goal is to avoid dependence on a single distribution channel. If one gate closes (Google limiting visits), others should open.

As a result, all of this could lead to a more stable and resilient ecosystem, even if Google Search plays a smaller role as the central hub. Publishers that diversify their revenue and strengthen direct ties with their audiences can prosper even with less traffic from search. And users will be able to support their favorite creators directly, or consume content in various forms according to their preferences.

Adaptation through AIO, GEO, and a multimedia strategy

As outlined earlier, one of the big opportunities is adapting to the new conditions—those who orient themselves quickly may even gain a competitive advantage. From an SEO and web-development perspective, now is the time to experiment with AIO (AI Optimization) and GEO (Generative Experience Optimization) methods. What does that mean in practice?
AIO means optimizing content so that artificial intelligence can process it easily and correctly. For example:

• Careful use of structured data: mark up reviews with schema.org (rating, author, date), use FAQ schema for frequently asked questions, etc., which can help AI directly extract Q&A formats or key parameters. Google already sometimes shows tables or bullet points in generative answers that are taken from well-structured pages.
• Clear and factual phrasing of key information: if an article contains an important definition or number, it makes sense to state it unambiguously (e.g., “The number of X is Y.”), because AI can identify that information more easily. The website should offer snippets that AI can quote.
• Consistent style and terminology: to avoid AI errors, it is good to use established terms and avoid contradictory wording. If, for example, a recipe is being described, stick to the same units and step-by-step instructions—AI can then correctly list part of the procedure for the user.
• Marking authorship and expertise: Google still factors authority into account, so it is important to show an author with a bio, references, and information sources. AI may not pass these elements on to the user, but they can influence whether Google selects the site as a source (trustworthiness).

GEO is an even broader concept: Generative Experience Optimization means that content should be prepared for all possible environments in which generative AI will appear. For example:

• Voice assistants: With SGE expanding, similar generative answers are likely to reach voice search (Google Assistant, Alexa). GEO therefore includes voice SEO—making sure AI can read the answer aloud clearly, that the content includes pronunciations of difficult names (the <phoneme> tag can help), and that the brand is mentioned (e.g., “According to Example.com, …” – Google may name the source aloud).
• Various devices and formats: AR/VR interfaces, smart glasses—if generative search extends there, the content should be ready. This can mean providing 3D models of products for an e-shop (so AR can display them) or having well-described images (AI may soon generate visual answers that combine images).
• Multimedia: Google is already working on search that combines text and video. GEO therefore means optimizing videos (add captions, chapters) and images (alt texts, captions) so that AI can link to a specific time in the video or describe an image with a reference to the author.

In short, a multimedia strategy is crucial: websites should not rely on text alone. AI snapshots are still mostly text-based, but Google already shows images and may soon include short clips or audio. If a publisher offers high-quality multimedia content, it can appear in generative results (e.g., as a recommended video for a query or an audio answer). And even if not directly in AI Overview, multimedia content is harder to “scrape”—AI can paraphrase text, but it is difficult to generate a unique video or infographic without the user having to visit the source.

Adaptation also means using AI for your own benefit. Many publishers already use AI to improve SEO (e.g., identifying content gaps), personalize content for users, automatically tag articles, and so on. This internal use of AI can increase efficiency and allow creators to focus on what matters. Some sites are experimenting with chatbots for users: for instance, an e-shop might have an AI advisor answering product questions on the site; a university might have an AI guide for students. These tools keep users engaged directly on the site and provide added value.

Summary of adaptation strategies: “Be wherever your user is and offer what a generic AI overview cannot replace.” That can be:

• Depth and expertise (see qualitative shift),
• Personalization and interaction (community features, tailored chat),
• Multimedia and experiential elements (video, audio, AR),
• Technical readiness (markup, structure, data),
• Brand building (so users recognize and seek out your content directly).

Those who embrace these principles in time will become pioneers of the new era of digital marketing and publishing. SEO experts are already developing methodologies and tools for monitoring SGE visibility, testing what type of content gets into AI answers (some SEO tools such as Rank Ranger and Semrush Labs have already reported which queries have generative answers and which sources are cited). The learning curve is wide open—it’s a chance to innovate and outpace competitors who are waiting on the sidelines.

Strategic recommendations for key players

Based on the analysis above, we now present concrete strategic recommendations for different roles within the digital ecosystem. The goal is to offer practical guidance on how to approach the new situation—to minimize risks and leverage potential benefits. The recommendations are grouped by target audience—separately for content creators, digital marketers, web analysts and developers, and regulators and policymakers.

For content creators (publishers, bloggers, media)

• Focus on quality and uniqueness of content: In the era of AI-driven search, the rule “content is king” applies more than ever. Create content with added value—original research, exclusive information, deep expertise, or personal perspective. Content that is interchangeable and easily replaced by a summary will have little chance of winning visitors. Conversely, authoritative content with high E-E-A-T will likely be preferred by Google as a source even in AI overviews. Keep your articles factually accurate and up to date so your brand earns a reputation as a reliable source—raising the likelihood AI will cite you or that users will come to you for more detail.
• Build a loyal audience and community: Reduce reliance on incidental search traffic by strengthening relationships with your readers or viewers. Launch newsletters that regularly deliver your content directly to fans. Use social media for real engagement (not just one-way link drops). Consider creating a community forum, Discord/Slack group, or FB group where your readers can gather—this boosts loyalty. A loyal reader might then go straight to your site (or use your internal search) instead of Google. A success story here is the NY Times, with millions of subscribers and a strong brand largely insulated from search traffic drops. Building a brand and community is a long-term investment that pays off in uncertain times.
• Diversify revenue streams: Prepare for potential declines in banner-ad revenue. Explore other monetization paths: introduce premium content behind a paywall, offer online courses, e-books, or a paid newsletter. Some sites have successfully launched membership programs (e.g., community access, special events). If you have a strong community, you can get support via crowdfunding or donations. Also consider e-commerce integration—sell merchandise or relevant products or services linked to your content. The goal is to avoid relying solely on advertising whose performance is beyond your control. Experts advise adding new revenue sources—events, job boards, etc.—instead of counting on display ads to come roaring back—they won’t. Diversified revenue means greater stability.
• Optimize content for AI (technically and editorially): Adapt your publishing workflows so your content is as readable and usable for AI systems as possible (AIO strategy). In practice that means:
  • using structured data (mark up articles, recipes, products with schemas—this can increase the chance AI takes your content into account and cites it correctly),
  • stating key information clearly (e.g., do not start an article with pages of fluff—summarize the result/test right at the top; provide definitions where the reader expects them),
  • including FAQ sections in content (Google likes to use these for supplementary questions in SGE),
  • adding relevant images with captions (AI may include them in or next to its answer, and users see your image with a citation, which boosts click-through),
  • maintaining site speed and UX: if a user clicks from AI Overview to your site (e.g., via the “Learn more” button), they likely expect fast, high-quality loading. A slow site or poor mobile usability could drive them back to the AI answer. Keep the site technically polished.

For digital marketers (SEO specialists, content strategists, PPC managers)

• Revise your SEO strategy for the SGE era: Audit the keywords and topics you target and categorize them by how AI Overview affects them. For queries where an AI answer already appears (or you expect it soon), don’t expect the same click-through rate and re-evaluate KPIs. It may be sensible to focus more on mid-tail and long-tail queries (where users still want detail the AI may lack) or on head queries where you want to be the cited source. Update your content plan: you can stop producing some generic articles (AI covers them) and instead identify gaps or new questions that arise from AI interaction (e.g., what follow-up questions might users ask after reading an AI summary? Produce those). Also consider optimizing meta titles and descriptions—AI doesn’t need them, but if your link sits beneath an AI answer, it should attract. A catchy title may get more clicks from AI Overview than from classic SERPs, where there were many links; now you might be one of three sources—stand out.
• Watch for new ad opportunities in AI environments: Google has signaled that ads will remain a “native part of SGE” and is already testing them. PPC managers must keep up: AI snapshots currently show product listings (Shopping Ads) labeled Sponsored. Make sure your Merchant Center data, product imagery, etc. are high quality because they may appear right in the generative answer. Experiment with different ad formats: if Google offers, say, a text ad mid-AI answer, tweak campaigns (perhaps different creative tailored to capture attention in AI text, not the old headline+description). Optimize for visual impact—Google recommends richer image and multimedia extensions since generative search is more visual. Retail advertisers should keep adopting Performance Max campaigns and feeds that Google calls the backbone of the new search experience.
• Focus on quality traffic and conversions: Success may not hinge on raw traffic volume as before but on the quality and intent of visitors. You may get less organic traffic, but those who come will be further along in the decision process—AI-qualified traffic with a higher chance to convert. Adjust tactics: landing pages should assume the visitor already knows the basic answer from AI. If you sell financial products and someone arrives after a generative overview of options, don’t re-explain general terms—offer specific reasons to choose your product, or an interactive calculator. Optimize the conversion funnel for a better-informed customer. Also double down on retention and retargeting: if you win someone organically, capture them (cookie, email, push) and guide them to return directly.
• Monitor and leverage search-trend data: SGE is changing queries—users in conversational mode issue longer, more specific searches. This can generate new long queries that weren’t common. Regularly analyze Search Console and other tools for emerging phrases. Mark Irvine (SearchLab) advises using more broad match and phrase match keywords in PPC to catch these new formulations while managing negatives carefully.
• Continuously optimize campaigns and explain changes: Keep in mind that falling organic traffic doesn’t necessarily mean SEO failure but structural market change. Educate colleagues or clients—set expectations that metrics may evolve differently (e.g., fewer visits but higher lead quality). Adjust KPIs, focus on conversion rate and ROI instead of just visit counts. Be ready to increase brand-marketing investment—if organic wanes, brand and direct traffic grow in importance.

For web analysts and developers

• Track new metrics and trends: Analysts should monitor how impression-to-click ratios shift on key queries. Identify where CTR drops—that may signal AI answered the query (the user didn’t click). If Google adds AI-related metrics in Search Console or GA4 (e.g., AI-Overview impressions), incorporate them immediately. Also use external SEO tools that detect SGE presence on certain queries (BrightEdge, Semrush, etc.) and try to quantify impact on your site. These data will be invaluable for future strategy and making the case to leadership for new measures.
• Implement technical measures for the AI era: Developers should check robots.txt and meta tags regarding new AI directives. Google recently introduced attributes such as Google-Extended for opting out of training data—decide whether to use them. Also keep an eye on schema markup standards and any new formats (Google may introduce special markup for identifying cite-able passages, etc.). Keep site code clean and semantic—proper headings, lists, tables, image descriptions—all make AI extraction easier. Consider exposing APIs for your content (e.g., a public data API) through which AI systems could access your data under your control.
• Innovate the site experience: When a visitor does arrive, offer value the AI overview didn’t. Developers can add interactive features—on a travel article, show a map, cost calculator, comment section, etc. You could also integrate your own AI chatbot that answers using your content (similar to Microsoft’s site chat or expert bots built on private knowledge bases). That keeps users longer and offers a consistent on-site experience. Don’t forget speed and performance—AI may increase load (e.g., on-the-fly answers), so optimize back-end, caching, and CDNs to keep things smooth.
• Test and adapt continuously: Run A/B tests of AI-era tweaks—for instance, add short summaries at the start of some articles and see whether scroll depth changes (you may help AI-conditioned readers find the main points and stay for details). Or test different headline formats for queries where you’re cited in SGE and measure CTR impact. The analytics team should work closely with SEO/content teams when assessing what works. Be ready to adapt site and tracking quickly as Google refines SGE (e.g., if Google enables interactions in AI Overview, make sure you can track that traffic).

For regulators and policymakers

• Update copyright and data legislation: Assess whether current laws sufficiently cover the use of website content in generative AI models. It may be necessary to clarify the status of training AI on web texts—is it fair use, or should it require a license? Regulators might mandate licensing agreements between large AI operators and content publishers (similar to laws compensating news publishers on platforms). That would ensure content creators get a fair share of the value AI generates from their data.
• Enforce transparency and accountability: Require companies like Google to report transparently how AI Overviews affect site traffic and content economics. For example, they could publish aggregate data showing the percentage of queries answered without clicks or how much traffic goes to sources. This would help objectively assess impact and calibrate regulation. Also consider standards for labeling generated content—users should know when a machine is answering and where it draws from. Some proposals say AI should cite sources much like humans do.
• Oversee competition: Google’s dominance means any search change has huge effects. If it appears Google favors its own content or services in generative answers (e.g., prioritizes YouTube, Google Shopping, etc., and suppresses external sites), that has antitrust implications. Authorities should watch whether Google abuses its position by choking off outgoing traffic or closing the ecosystem. Regulators could, if needed, require that a certain percentage of generative content come from external sources, or in extreme cases consider structural measures.
• Support open technology and innovation: The public sector can help balance big-platform influence by supporting open AI models and independent search tools. This could involve grants for public-interest AI research or collaboration with European AI initiatives. Regulators can also facilitate dialogue between Google and publishers—e.g., a forum to regularly discuss data and impact (similar to content-moderation discussions on social networks). International cooperation is key: AI is global, and coordination (e.g., under the EU AI Act) can stop global platforms from playing by their own rules regardless of local markets.
• Protect user rights and information pluralism: Policymakers should consider the broader societal impact—generative AI in search affects how citizens get information. Ensure the new system doesn’t spread disinformation or lock users into a single version of the truth. It may be necessary to regulate AI bias—e.g., require independent audits of algorithms for skew. Maintaining a plurality of sources serves the public interest; supporting public-service content or favoring nonprofit information sources in AI environments could help (so they don’t disappear in favor of SEO-optimized commercial sites). The aim should be to keep the internet open and diverse even as it undergoes AI transformation.

Conclusion – The future of the open internet in the AI-search era

The arrival of Google AI Overview is a milestone in search evolution—“over time this will just be how Search works,” Sundar Pichai said. It is therefore very likely that generative AI will become an integral part of how people discover information online. The impact on the digital ecosystem is complex: on one hand, there is the promise of faster, smarter user experiences; on the other, a threat to the mechanisms that fund and sustain content on the open web.

For the future of the open internet, it is crucial to find a new equilibrium. If everything is left to run wild, extreme scenarios loom—either the internet is flooded by AI-generated content spam (a vicious circle of poor quality) or information is oligopolized by a few platforms, leading to the decline of independent sites. Neither is desirable. A more optimistic scenario, supported by arguments in this essay, is that players adapt and build a sustainable coexistence model: an AI-powered search engine offers user convenience but also “sends valuable traffic and supports a healthy, open web,” content creators focus on quality, innovation, and new revenue streams, and regulators ensure fair rules.

In practice, a partial correction is already under way—Google is experimenting with adding links and citations directly into AI text, which it says has increased clicks to sources. It has also pledged to keep adjusting based on publisher feedback. This suggests the path isn’t predetermined—SGE/AIO may evolve to minimize negative effects. Google certainly won’t risk killing content: it will need to keep the ecosystem alive, either through product tweaks or financial participation in content creation (licenses, funds, etc.).

For creators and marketers alike, the open internet remains full of opportunities; only the routes to the audience are changing. Those who now invest in adaptability—from AIO/GEO techniques to new monetization models—can gain an edge in the coming years. Users need not lose out either: quality content will not disappear, it will just be packaged and distributed differently. Indeed, the pressure of AI may even cleanse online content, and users could ultimately find richer information (whether via AI or direct interaction with an expert site).

The future of the open internet therefore depends on our ability to innovate and collaborate. Google AI Overview need not be the destroyer of the open web—it can be a catalyst for its next evolution. But it will require active participation from all stakeholders: tech firms, creators, advertisers, users, and lawmakers. If we can ensure that quality, truthfulness, and fair compensation stay at the center, generative search could elevate the whole ecosystem.

At this pivotal moment, we must not forget the core principle of the open internet: the free flow of information and the ability for anyone to contribute and to draw from it. AI should serve as a tool that facilitates this flow, not one that closes it. Ultimately, the future is not predetermined—we shape it through the decisions we make now. This essay has presented arguments both for caution and for actively seizing new opportunities. As one commentator aptly noted, Google may see SGE as a “springboard into the world of the web,” but it is up to all involved to ensure that the springboard leads to the further development of the open internet, and not to its neglect. If we succeed, generative AI can become not a threat but the next chapter in the web’s evolution, in which both knowledge-seeking users and the creators who produce that knowledge continue to thrive.

Sources:

• Sanchez, M. (2024). What Google’s new AI Overviews means for Raptive creators and publishers. Raptive Blog.
• Goldman, J. (2024). Google’s AI-driven SGE could slash publishers’ traffic by 20%–60%. Insider Intelligence / eMarketer.
• Konstantinovic, D. (2024). Google’s AI Overviews aren’t hurting some publishers’ traffic yet. Insider Intelligence / eMarketer.
• Ostwal, T. (2024). Publisher Traffic Spiked Thanks to Google’s Gen AI Overview Tweaks. Adweek.
• Pichai, S. (2024). Quarterly earnings call – remarks on Search Generative Experience. (Cited in The Decoder).
• Goodwin, D. (2024). Search is evolving toward SGE, says Sundar Pichai. SearchEngineLand.
• Glover, R. (2024). Google’s Search Generative Experience: What AI on the SERP means for you. WordStream.
• Patel, N. (2024). SEO died in 2024. GEO is born. Bookspotz.
• Stein, R. (2024). Generative AI in Search: Let Google do the searching for you. The Keyword – Google Blog.

 

Glossary

• AI Overviews: AI-generated summary answers displayed by Google above traditional search results, combining information from multiple sources.
• Featured Snippet: A highlighted excerpt of text, image, or list from a single web source that appears at the top of Google’s search results as a direct answer to a user’s query.
• E-E-A-T (Experience, Expertise, Authority, Trustworthiness): A set of evaluation criteria used by Google to judge the quality of content and the credibility of websites and their creators.
• Zero-click search: A search in which the user finds the answer directly on the search results page (e.g., in an AI Overview or featured snippet) and doesn’t need to click any external link.
• User Intent: The intention or need a user has when entering a search query (e.g., to get information, buy a product, find a location).
• Schema Markup (structured data): Code (e.g., Schema.org) added to web pages to help search engines better understand the content and display rich snippets.
• DSM Directive (EU 2019/790): An EU directive on copyright in the digital single market that, among other things, regulates exceptions for text and data mining (TDM) and introduces an opt-out mechanism.
• Text and Data Mining (TDM): An automated analytical technique for extracting text and data to uncover patterns, trends, and correlations, often used for training AI models.
• Opt-out: A mechanism allowing copyright holders to explicitly prohibit the use of their works for TDM, usually via machine-readable means.
• EU AI Act: Forthcoming EU regulation to establish harmonized rules for artificial intelligence, imposing copyright-compliance duties on providers of general-purpose AI models.
• GPAI (General-Purpose AI Models): AI models usable for a wide range of tasks, not limited to a single purpose (e.g., large language models used in AI Overviews).
• Google-Extended: A mechanism offered by Google for websites to reserve rights against their content being used to train future Google AI models (e.g., via a robots.txt directive).
• AI bias: Systemic error or prejudice in an AI model’s output arising from skew in training data or algorithm design.
• YMYL (Your Money, Your Life): Topics that could affect a user’s future happiness, health, financial stability, or safety (e.g., medical, legal, financial advice). Google applies stricter E-A-T standards and often limits AI Overviews for such queries.
• AIO (AI-first Optimization): A content-optimization strategy aimed at making content easily processed and understood by machine-learning algorithms used in AI search.
• Generative Experience Optimization (GEO): A broader concept that goes beyond SEO and focuses on crafting the overall digital experience for users in conversational and multimodal AI environments.
• Cannibalization (of content/traffic): A situation in which one form of content presentation (e.g., AI Overview) “eats” the traffic or visibility of another form (e.g., the original web page) from which the content is drawn.
• Contextual information: Data that give meaning and context to a situation or dataset (e.g., user location, time, search history) and that AI uses to personalize and improve answer relevance.
• Snapshot (AI Snapshot): A one-off, holistic answer generated by AI to a given query, often displayed at the top of search results.
• Conversational mode (Chat with Google / SGE Overviews): An interactive search mode where the user can ask follow-up questions based on the previous answer, making search a smoother dialogue.
• Paywall: A mechanism on a website that restricts or blocks access to content for users who have not paid a subscription or created a paid account.
• Newsletters: Regular email messages sent to subscribers, containing news, articles, or other content—a direct distribution channel.
• Robots.txt: A text file on a web server that provides instructions to web robots (including search crawlers) about which parts of the site they may index and which they may not.
• Meta tags: HTML tags in a page’s header that provide metadata about the page’s content. Some (e.g., noindex, nosnippet, noarchive) influence how search engines, including AI, process the page.
• Needs Met: A Google search-quality rating criterion assessing how well a search result satisfies a user’s informational need or intent for a given query.

I have also written an article with links to further studies and updates: https://mareklecian.cz/en/google-ai-overviews-and-ai-mode-top-studies-analysis-and-recommendations/