Google Changed How AI Search Picks Its Sources
By the Numbers
345,000 sources picked. 2x the clicks when labeled
Half of Google searches now show an AI summary
Only 16% of brands track AI search performance.
44% now call AI search their primary source.
On May 27, 2026, Google quietly changed how its AI search results pick and show their sources. It made three changes, and none of them got much attention. All three affect whether your brand still gets seen when someone asks AI a question instead of clicking a link. Here is what happened and why it matters.
The Problem
If you run marketing for a company somewhere between $10M and $100M, you have probably felt this without being able to name it. You built the content. The rankings came, and a few of your pages still sit on page one. And the traffic keeps slipping anyway.
Here is why that feels confusing. For most of the last decade, the deal was simple. You ranked, someone clicked, they landed on your site. Rankings and visits moved together. If you could see the ranking, you could roughly predict the visit.
That deal is breaking.
When someone asks Google a question now, they often get an answer at the top of the page written by AI. Google calls these AI Overviews, and the fuller conversational version AI Mode. The answer is built from many sources, and it sits there before the regular blue links. A lot of the time, the person reads it and never clicks anything.
The scale is not small. McKinsey reports that about half of Google searches already show an AI summary, and projects that figure to pass 75 percent by 2028. McKinsey also found that 44 percent of people who use AI search now call it their primary source of information, ahead of traditional search at 31 percent. For queries that now trigger an AI answer, some sites have watched organic traffic fall in the 20 to 30 percent range.
So the ranking can hold while the visit disappears. That is the part that does not show up cleanly in your reports. Your position looks fine. Your traffic does not. And most analytics tools were never built to show you why.
Here is the harder part. Even inside the AI answer, your own website is usually a small piece of what the AI reads. McKinsey found that brand-owned pages often make up only 5 to 10 percent of the sources an AI uses to build an answer. The rest comes from publishers, review sites, forums, and social discussions. And only 16 percent of brands track how their content performs in AI search at all.
Now Google has added new rules about which sources get pulled forward and labeled inside those AI answers. That is the thing worth understanding, because it changes who gets seen.
The Insights
Google framed the update around one idea. In its own words, the goal is to surface:
"High-quality content and firsthand perspectives." — Google
There are three separate changes. They are related, but they are not the same thing, so it helps to take them one at a time.
Preferred Sources now appear inside AI answers
Preferred Sources is a feature you have to opt into. A signed-in person goes into their Google Search personalization settings, picks the websites and creators they want to see more of, and Google then favors those sources for them. Any site that publishes fresh content is eligible.
Until now, that preference only changed the Top Stories section. As of this announcement, it also changes AI Overviews and AI Mode. When a source you picked shows up inside an AI answer, it carries a label so you can spot it.
Duncan Osborn, a Product Manager on Google Search, described it plainly:
"Easily spot links in AI responses from the sources you've already selected." — Duncan Osborn, Google Search
A few numbers make this concrete:
People have already selected more than 345,000 unique sources, up from roughly 90,000 when Google first expanded the feature.
Google says people are about twice as likely to click a result when it is labeled as a Preferred Source.
The feature is live globally and works in all languages.
Read that middle number again. A label roughly doubles the click rate. That means the label itself is now a ranking factor of sorts, and it is one your audience controls, not Google's algorithm alone.
A new carousel for perspectives
The second change is about how Google shows articles and opinions on topics that are still developing.
For some searches on a topic someone is following, Google will now show a prominent carousel of timely articles near the top of the AI response, instead of tucking a few links at the bottom. That carousel will also highlight your Preferred Sources. The point, in Google's framing, is to make recent articles visible on a wider range of queries and give people a few options before they decide where to read.
There is a second carousel coming. For searches where someone wants to hear from other people rather than from a publisher, Google says they will soon see a carousel of "helpful perspectives" pulled from online discussions, forums, and social media. Google used the words "you'll soon see," which tells you this part has not fully rolled out yet.
Both carousels are showing up inside AI Mode and AI Overviews.
Here is the quiet shift inside this one. Forum posts, community threads, and social discussions now have a formal seat in the AI answer. Not as a workaround. As a built-in slot Google designed on purpose. For years, the goal was to get a blue link on page one. Now there is a separate lane for first-hand, lived-experience content, and it sits right next to the editorial links.
The Highly Cited label gets bigger
The third change is about original reporting.
Google first introduced the Highly Cited label back in 2022, on Top Stories for mobile. It marks an article that a lot of other stories have referenced, so a reader can tell they are looking at the primary source instead of a piece that just summarized someone else's work.
Two things are changing now:
The label is expanding to more web article links across the regular search results page, not just Top Stories.
Google is adding a second signal that shows when an article explicitly references a Highly Cited source. So a reader can see both the original reporting and the follow-up coverage that points back to it.
One detail matters for planning. Google said this expansion runs across Google Search and is not specific to AI Mode or AI Overviews. But the underlying idea, who cites whom, is exactly the kind of signal an AI answer leans on when it decides which source to trust and pull forward.
What these three changes add up to
Put them side by side and a pattern shows up. Google is naming, and labeling, three kinds of source that earn a spot in AI search:
Sources a person has personally chosen to favor. That is Preferred Sources.
First-hand and community content from forums, discussions, and social. That is the perspectives carousel.
Original reporting that other people cite. That is the Highly Cited label.
So the question "which channel wins" no longer has one answer. Visibility in AI search is now split across three different levers:
One lever your audience controls, by picking you as a preferred source.
One lever Google's AI controls, by deciding which documents to read and trust.
One lever the open web controls, through who links to and cites whom.
That is the kernel of it. "Preferred channel" used to mean the spot you fought for in the rankings. Now it is partly chosen by the user, partly routed by the AI, and partly earned through the structure of citations and conversation across the web. You can influence all three. You cannot win any of them by treating this like 2022 SEO.
And the stakes are not abstract. McKinsey put a number on it:
"By 2028, $750 billion in US revenue will funnel through AI-powered search." — McKinsey
That is the economic floor under all of this. AI search is not a feature to monitor. It is a channel where real money is already moving.
The Takeaway
1: Your rankings can stay flat while your traffic falls, because the AI answer now resolves the question before the click happens. The problem is not your position. It is that the decision is being made on a surface you do not control yet.
2: Visibility in AI search is no longer one thing you optimize for. It is three. Your audience can choose you as a Preferred Source, the AI can pull you into an answer, and the web can mark you as a Highly Cited original. Different work earns each one.
3: Your own website is a minority of what the AI reads. If you only invest in your owned pages, you are competing for 5 to 10 percent of the answer and ignoring the rest. The brands that show up are present across publishers, communities, and original reporting, not just their own blog.
The Action
Action 1: Ask your audience to make you a Preferred Source. Preferred Sources roughly doubles click-through, and it is opt-in by the user. Add a short, plain instruction to your newsletter and your best content telling readers how to set you as a Preferred Source in their Google Search personalization settings. This is the rare lever where a loyal audience directly raises your AI visibility, so treat it like a list-building goal, not an afterthought.
Action 2: Find out if you are even in the answer. You cannot fix what you cannot see. Run your top 15 buying-intent questions through AI Overviews, AI Mode, and a couple of standalone AI tools, and write down whether your brand appears and which sources got cited instead. Do this monthly. This is your AI visibility baseline, and most of your competitors do not have one. You can do a quick check - https://tools.camino5.com/
Action 3: Build the first-hand content the perspectives carousel wants. Google is creating a dedicated slot for forums, discussions, and social perspectives. Generic, written-by-committee content does not fill that slot. Real customer stories, honest reviews, founder takes, and active community threads do. Pick two topics where your customers have genuine first-hand experience and publish content that sounds like a person, not a brand template.
Action 4: Earn citations as the original source, not the summary. The Highly Cited label rewards primary reporting that other people reference. Once a year, publish something only you can: your own data, a customer survey, an internal benchmark, a point of view no one else has. Original work gets cited. Summaries of other people's work do not.
Action 5: Add an AI-search line to your reporting. Stop judging organic health on rankings and clicks alone. Add a simple tracker that answers three questions each month: are we cited in AI answers for our key topics, are we labeled, and is our share of citations growing. Pair it with Google Search Console to watch where impressions hold but clicks drop, which is the early signature of the AI answer absorbing your traffic.