Randomized Experiment Finds Google AI Overviews Cut Outbound Link Clicks by 40%, as AI Summaries Now Appear in 61% of Search Results
Google's automatically created summaries at the top of search results are changing how much traffic websites receive. A randomized controlled experiment found that when AI Overviews appear in search results, the share of users who click through to outside websites falls by 40%. Researchers hid the feature from some desktop users in the United States and compared their behavior with a control group that could still see the summaries. This 40% difference comes from an experiment, not simply from observing a possible connection between two things.
In simple terms, when an AI answer appears at the top of the search page for a keyword, even the website ranked first in regular search results gets about 40% less traffic than before. The reason is straightforward. Google puts AI Overviews above its regular results and includes links to the pages used to create the summaries. After reading the AI answer, most users do not scroll down the page. Search engine optimization (SEO) still helps people find a page, but being found and getting visitors are now two different things when an AI answer appears first.
Another important issue is how quickly AI Overviews are spreading. DataForSEO's monthly study of 20 million keywords found that the share of search results containing AI Overviews rose from 5.93% in September 2024 to 61.43% in August 2026. In less than two years, the share grew from below 6% to above 61%.
| Time point | Share of search results with AI Overviews |
|---|---|
| September 2024 | 5.93% |
| August 2026 | 61.43% |
Note: The data comes from DataForSEO's monthly study of 20 million keywords.
This change has led to the idea of GEO, or Generative Engine Optimization. Its goal is to get content included in AI-generated answers. A practical way to understand GEO is to think of it as SEO plus two extra steps. First, your content needs to appear on the outside websites that AI answers use as sources. AI answers do not come from nowhere. They use information from sources they trust, so getting included among those sources matters more than simply writing a good page on your own website. Second, your pages need to be easy for AI systems to quote. This means using a clear structure, putting complete arguments in separate paragraphs, and presenting facts that can easily be taken out and used in an answer.
Beyond these points, there do not appear to be any other new rules. This is important because GEO has not created a completely new set of methods. Instead, it has shifted attention from how highly your website ranks to whether AI systems trust your content. The skills content creators have built over the past decade, such as choosing keywords, getting links from other websites, and improving page loading speed, are still useful. But these methods mainly help answer the question, "Will users click through to my website?" When the answer is already shown on the search page, users may have no reason to visit the website. The biggest unanswered question is what will happen to the remaining 39% of search results. Now that more than 60% contain AI answers, how the rest will change is one of the most practical questions for the next two years.
At the same time, Google has changed the models available with its paid plans. The company released Gemini 4 Neon on November 18 and started adding it to the Gemini app on the same day. In the Gemini 4 family, Neon sits between Hydrogen, which focuses on speed, and Carbon, the most powerful model. For paying users, this also means there are now two fewer models to choose from.
Google AI Pro costs $19.99 per month in the United States and €21.99 (about $24) in Germany. Its model menu now has three choices: Auto, Hydrogen and Neon. Argon and the older Pro models are no longer included in this plan. Neon has taken Argon's place directly. Google is adding the change gradually, so some accounts may receive it later. Carbon, the most powerful model, is still only available through the AI Ultra plan, which starts at $99.99 per month.
There is one setting that is easy to miss. If you choose a model yourself but do not select a thinking level, Gemini "will default to Low." For more difficult tasks, Medium or High is a better choice, but using these levels consumes your usage allowance more quickly. The model and thinking level stay the same throughout a conversation, even if you open that conversation again later.
Neon is not available to free users or subscribers to Google AI Plus, which costs $4.99 per month. AI Plus subscribers can still choose Hydrogen. Free users can only use Hydrogen when Auto selects it for their prompt. On November 13, AI Plus also lost access to Pro models. Developers must pay to use Neon too: Google's Gemini API pricing page does not show a free plan for this model.
In the app, Google says Neon produces clearer writing and is better at reading tables, PDFs and photos. The company also says it searches the web more often, which can help with tasks that need current information, such as comparing prices. Google says tasks in Gemini can now run on Neon, and multi-step tasks use less of the usage allowance than they do with larger models. Neon scored 61 on the Artificial Analysis Intelligence Index. In a comparison table selected by Google, this puts it ahead of Claude Opus 5.5, which scored 59, and GPT-6.2 Sol, which scored 57.
For developers using the API, Neon costs $3 for every million input tokens and $15 for every million output tokens. It has a context window of 1 million tokens. Until November 30, Carbon and Argon cost $2 and $10 per million tokens, respectively. From December 1, their prices will rise to $4 and $20. Developers who currently use Argon should work out how much switching to Neon will cost before the end of the month.
These two developments point to the same larger change: updates to both Google's search product and its AI models are putting pressure on the advantages that outside content creators and developers have relied on. On the traffic side, AI answers keep users on the search page instead of sending them to websites. On the developer side, changes to models and API prices are changing how much it costs to build and run products. For websites that depend on search traffic, getting their content included among the sources cited by AI is now just as important as competing for high search rankings.
