What Actually Changed, and What It Means for the Open Web
Type a question into Google today and there is close to a coin flip chance you will not click on anything. Instead, an AI generated summary appears at the top of the page, pulling together an answer from several sources, and for a huge share of users, that answer is enough.
No scrolling, no comparing links, no visiting the sites that actually did the research.
This is AI Overviews, and two years after its launch it has quietly become one of the most consequential changes to how information travels across the internet since Google itself first indexed the web.
AI Overviews is Google's AI generated summary that appears above traditional search results for a large share of queries. Instead of showing ten blue links and letting the user pick, Google's model reads across multiple sources, synthesizes an answer, and presents it directly in the results page, with small citation links tucked beneath or beside the text.
It launched in 2024 as an experiment and has since scaled into something close to a default experience. As of early 2026, it appears on roughly 48 to 50 percent of all tracked Google searches in the US, up from just 6.5 percent in January 2025, a near eightfold jump in about a year. It now runs in more than 200 countries and over 40 languages, and at Google's I/O 2026 keynote, the company said AI Overviews and its more conversational sibling, AI Mode, reach more than 2.5 billion monthly users combined.
The rate at which it triggers depends heavily on the type of question being asked. Health related queries trigger an overview roughly 88 percent of the time, education around 83 percent, while entertainment queries sit much lower at about 37 percent. Broadly informational questions, the kind that used to send people straight to a blog post or a Wikipedia page, trigger it about 40 percent of the time.
Here is the number that has kept publishers and SEO teams awake for two years: when an AI Overview appears, clicks to the traditional top ranked result fall sharply. Different research groups have measured this differently, with figures ranging from a 34 percent drop to as much as 58 to 61 percent, depending on the study, the query type, and the time period measured. Whatever the exact figure, the direction is consistent and has held steady since the feature's earliest days.
The story did get slightly less grim in 2026. After organic click through rates on AI Overview queries fell as low as 1.3 percent in December 2025, they climbed back to around 2.4 percent by February, a meaningful rebound even if still far below what a normal search result page delivered before the feature existed. Meanwhile, queries without an AI Overview saw their own click through rates rise over the same period, suggesting users are increasingly sorting themselves into two very different search experiences depending on what they are looking for.
The single most important shift for anyone publishing content online is this: ranking well in traditional search results is no longer enough on its own. What matters increasingly is whether your page gets cited inside the AI generated answer itself.
The data on this is striking. Sites cited within an AI Overview earn substantially more organic clicks per impression than uncited sites appearing on the exact same query, with some studies putting the gap above 100 percent. Being the source Google's model chose to quote or reference has effectively become a new, separate battleground running parallel to classic search rankings.
This has also scrambled the old assumption that citation follows ranking. Analysis using large scale citation datasets has found that pages ranking well outside the traditional top ten, sometimes between positions 11 and 20, still get pulled into AI Overviews if they answer the underlying question clearly and directly. Google's own summarization system appears to be evaluating clarity and topical relevance somewhat independently of the traditional ranking signals that decide where a page lands on a standard results page.
One pattern that stands out across recent research is how strongly AI Overviews favor recently updated content. Pages refreshed within the last 30 days receive roughly three times more AI citations than older material covering the same topic. For any publisher, this argues for a habit that used to be optional and is now closer to essential: revisiting and updating existing articles, not just publishing new ones.
This connects directly to the broader shift Google confirmed through its own guidance in 2026, stating plainly that optimizing for generative AI search is simply optimizing for the search experience as a whole, and that publishers do not need a separate strategy for traditional SEO versus what some in the industry have started calling AEO or GEO, meaning answer engine or generative engine optimization. According to Google, it is still one discipline, just with new terrain to cover.
The data also makes clear that AI Overviews is reshaping the web unevenly. Recent core updates through 2026 have consistently rewarded original brands, official sources, and destinations rich with proprietary data, while pages that primarily aggregate or summarize content already published elsewhere have lost the most ground. One March 2026 core update was severe enough that nearly 80 percent of top three search positions shifted, and roughly a quarter of previously top ten ranking pages fell entirely out of the top 100 results.
Citation patterns tell a similar story. Certain large, well established platforms, particularly Wikipedia and YouTube, continue to dominate as sources AI Overviews pulls from most often. At the same time, the share of citations going to brand owned websites, meaning a company's own domain rather than a third party writing about them, has been climbing, moving from around 26 percent to 31 percent over the past year, a small but genuinely encouraging trend for independent publishers willing to put in the work.
For anyone publishing content today, a few conclusions hold up consistently across the data. Clear, direct answers near the top of an article tend to get cited more than long narrative buildups before the point is made. Content that stays current, revisited and updated regularly rather than published once and left alone, earns a meaningful citation advantage. Original data, direct analysis, and firsthand information outperform pure summary or aggregation, because that is precisely the kind of content Google's own systems now have the least reason to rewrite in their own words when a source has already done that work well.
None of this makes traditional search obsolete, and it clearly has not replaced the fundamentals of good, well structured writing. What it has done is add an entirely new layer that publishers now have to think about deliberately, since roughly half of all searches today are answered before a single link ever gets clicked.
How to Get Cited in AI Overviews
- Answer the question in the first two or three lines, without a slow build up.
Summarization systems prioritize fragments that answer directly, not articles that open with historical context or storytelling before making the point. If your headline is essentially an implicit question ("what is X?"), the first sentence of the body should almost answer it already.
- Update existing content, not just publish new articles.
The data is clear on this: content refreshed within the last 30 days receives 3.2 times more AI citations than older material on the same topic. Revisiting and refreshing older articles, with new data, updated dates, or an added paragraph, can matter more than writing something new from scratch.
- Use a structure that is easy to extract.
Numbered lists, comparison tables, one sentence definitions, subheadings phrased as real questions people search for. A long narrative paragraph is far harder to cite in a short fragment than a clear, self contained sentence.
- Bring something that isn't already on fifty other pages.
This is the pattern the August core update confirmed: content that merely summarizes what others already say loses ground to content with original data, verified figures, named and reasoned opinion, or genuine analysis. It is exactly what you are already doing in pieces like the VCSEL versus MicroLED comparison, so you are already headed in the right direction.
- Schema.org markup (JSON-LD).
Well implemented Article, FAQPage or HowTo markup helps the system understand unambiguously what question each page answers. It is an added technical signal, not a substitute for the content itself, but it helps.
- Watch the question to answer ratio in your headings (H2/H3).
If an H2 reads close to what someone would actually type into a search bar ("How does X work", "What is the difference between X and Y"), that section has a better chance of being pulled as the source for the answer.
