Get a free AI visibility report
.png)
Google has moved Preferred Sources out of Top Stories and into AI Overviews and AI Mode. A reader's saved preferences can now shape which sources appear inside an AI-generated answer, and carry a visible label when they do.
Preferred Sources started as a niche Top Stories filter. It has quietly become something closer to a trust layer inside Google's AI answers. Anyone tracking how a brand shows up across ChatGPT, AI Overviews, Gemini, Copilot, and the rest of the answer engine landscape should understand this shift in detail, not just as a headline.
This article covers what the feature does, the timeline that got it here, why it matters for AEO, and the exact code to add the button, including a Webflow walkthrough.
Preferred Sources lets a signed-in Google user mark specific publications as trusted, go-to sources. Once selected, Google prioritizes content from those sites. It shows that content more prominently across the surfaces where the feature is live.
The mechanic is simple from the reader's side. A user searches for a topic, opens the source preferences tool, and adds a handful of publications they already trust. From that point on, Google is more likely to surface content from those sites. It labels the result with a "Preferred" badge so the reader recognizes it.
What makes this different from a normal ranking signal is that it is entirely personalized. Two readers can search the same query and see two different sets of preferred badges. Each set depends on which sites that specific reader selected. Google has described this as a personalization layer that runs in parallel to the core ranking algorithm. It is not a universal boost that lifts a site for every visitor.
The feature has moved fast, from a small test to a fixture across several Google surfaces.
Google launched Preferred Sources as a Search Labs beta in June 2025. It reached general availability in the United States and India in August 2025, limited to the Top Stories module. The feature expanded to English-language users worldwide in December 2025. It rolled out globally, across all supported languages, on April 30, 2026.
The turning point came on May 27, 2026. Search Engine Journal reported that Google had brought Preferred Sources into AI Overviews and AI Mode. A reader's saved choices could now carry a visible label inside an AI-generated answer for the first time.
On August 20, 2026, Google published a company blog post from Mrinalini Loew, General Manager of the Google Search Ecosystem. It announced an embeddable button that publishers can add to their own pages, alongside two other personalization updates to Discover and Google News audio briefings. TechCrunch reported that the update builds on earlier Google findings. People are about twice as likely to click through to a source once they have marked it preferred.
That August update matters because it removed the main source of friction in the original flow. Before the button, a reader had to leave the page and navigate to a separate Google settings screen. The button keeps the reader on the page and returns them to where they left off once the selection is made.
Most AEO advice focuses on making a site easier for a model to crawl, parse, and cite. Preferred Sources works on a different lever: reader trust, built over repeated visits, then explicitly registered with Google.
That distinction lines up with how citation share actually forms. In practice, a small set of domains earns most of the citations for any given topic. Earned mentions on trusted third-party pages typically outweigh direct owned citations to a brand's own domain. Preferred Sources adds a second, reader-driven trust signal on top of that pattern. A site does not need to change its content or its technical setup to benefit. It needs enough loyal readers who choose to follow it.
For a brand tracking source mention rate, this is one more reason the metric matters. A domain that already earns frequent citations on a topic is also the kind of domain a reader is likely to mark preferred. That can compound the advantage over time. Cognizo's Answer Engine Insights module tracks source mention rate alongside citation share. The goal is to show a brand which third-party domains AI platforms already trust on its topic, so outreach and content partnerships can target those domains instead of guesswork.
It is worth being precise about scope here. Preferred Sources is a Google feature. It shapes what a Google user sees in Top Stories, AI Overviews, and AI Mode. It has no bearing on how ChatGPT, Claude, Perplexity, or Copilot decide which sources to cite. Each of those platforms runs its own retrieval and ranking logic on its own index, fully separate from Google's.
Gemini sits in a different spot. It is not one of the three named surfaces, but its Google Search grounding option draws on the same search infrastructure that powers AI Overviews. Google has not stated whether a reader's preferred sources carry over into grounded Gemini answers, so that connection is unconfirmed rather than ruled out. Either way, a brand that invests heavily in Preferred Sources adoption is optimizing one mechanism inside Google's ecosystem, not the full set of platforms that typically matter for AI visibility. Treat it as an addition to a cross-platform strategy, not a replacement for one.
Preferred Sources rewards a domain readers already trust. The most useful way to find those domains isn't guesswork, it's looking at who already earns citations in AI answers for a given topic.
In Cognizo's own citation tracking across the AI-visibility and AEO topic set we monitor, a handful of domains stand out by source mention rate, our measure of how often a domain gets cited relative to every other domain in the same prompt set. Over the roughly three months from June 1 to late August 2026, developers.google.com, the same page that hosts the Preferred Sources implementation guide linked above, accounted for 0.79% of all citations we tracked in this space. Search Engine Land sat at 1.08%, TechRadar at 1.33%, and G2 at 0.90%.
None of that measures Preferred Sources adoption directly. Google doesn't expose which domains readers have marked preferred. But it points at the same underlying signal: a domain that already earns steady citation share has already done the harder work of becoming trusted. Adding the button is a small step after that. For a domain still building citation share, the button alone won't manufacture trust it hasn't earned yet, the citation work has to come first.
Google's Search Central documentation lists three implementation paths, in increasing order of technical control. All three require a domain-level or subdomain-level site. A subdirectory such as example.com/blog is not eligible. Check eligibility at google.com/preferences/source before writing any code.
This is Google's recommended path. It needs only two lines of HTML, and it renders an automatically localized, Google-styled button.
<script async src="https://news.google.com/swg/js/v1/publisher.js"></script><div google-add-preferred-source-btn data-theme="dark"></div>The data-theme attribute accepts light or dark. A data-lang attribute can force a specific display language, overriding the visitor's browser default.
The same library exposes a programmatic API for a custom-designed trigger instead of Google's default button. This API can be tied to any element on the page.
(self.PREFERRED_SOURCE = self.PREFERRED_SOURCE || []).push(function (preferredSource) { preferredSource.init({ theme: 'light', lang: 'en' }); const button = document.querySelector('#myButton'); button.addEventListener('click', () => { preferredSource.addPreferredSource(); });});This path suits a CMS or design system that does not render Google's auto-generated markup cleanly. It also suits a team that wants the trigger tied to a specific on-page action.
A plain link accomplishes the same thing without any script. It works in emails, social posts, and newsletters, as well as on a page.
<a href="https://www.google.com/preferences/source?q=example.com"> Add us as a preferred source</a>Replace example.com with the actual domain.
A large share of marketing and publisher sites run on Webflow. The practical steps look like this:
WordPress has two practical paths, depending on how comfortable a site owner is with touching code.
Manual, using a header and footer plugin. Install a script-injection plugin, the same category of tool most sites already use for Google Analytics or similar tracking snippets. Paste the standard script tag into the header field, site-wide. Add the div snippet through the same plugin's body or footer field, or drop it into a post using an HTML block in the block editor. For it to appear on every article automatically, add the div through the theme's single post template instead of one post at a time. A plugin with per-post-type targeting works too. As with Webflow, the button only renders on the live page, not inside a draft preview.
Turnkey, using a plugin. Dedicated plugins exist in the WordPress plugin directory that wrap this setup into a settings screen, typically with a banner or inline card, position and color controls, and sometimes click analytics. None of them are built or endorsed by Google. Search the plugin directory for "preferred source" to compare current options, and check the install count and last-updated date before relying on one.
Preferred Sources influences each reader individually. It does not apply a blanket ranking boost, so there is no single dashboard metric that captures its full effect.
A few things are worth watching instead. Track referral traffic patterns from Top Stories and Discover over time. Note any visible "Preferred" labeling on AI Overview appearances where it shows up. Watch the broader trend in AI-driven visibility, since this signal now runs alongside the usual citation and ranking factors.
This is exactly the kind of change that argues against periodic, one-off audits. A reader-driven signal can move a brand's standing on a given surface between quarterly checks, without any warning. Cognizo's AI Traffic Analytics module pairs crawler activity, including GPTBot and OAI-SearchBot, with human referral sessions. A shift in AI-referred traffic then shows up against the crawler behavior that produced it, rather than as an isolated number.
Preferred Sources sits closer to earned media than to technical optimization. Building the reader loyalty that leads to a preferred selection looks a lot like the work that already earns citations elsewhere. It takes consistent, useful content, active promotion, and a presence on the third-party domains a topic's audience already trusts. Those are the answer engine optimization fundamentals that apply across every AI surface, not just Google's.
That is the same groundwork covered when a technical signal like llms.txt gets added to a crawlability strategy. No single mechanism replaces the underlying work of being genuinely citable. Each one is a lever worth pulling once the fundamentals are in place.
None of this replaces watching how citations actually move. Whether a mention comes from an AI Overview badge, a ChatGPT citation, or a Perplexity source list, the underlying question stays the same: is the brand showing up where its buyers are asking, and is that presence growing or shrinking. Cognizo's Autopilot plan already folds fast-moving mechanics like Preferred Sources into its ongoing research and content loop. A change like this becomes one more input the system accounts for, rather than a manual research task added to someone's backlog. Continuous tracking of brand mentions across platforms is what turns a single feature update into an actionable part of a strategy, instead of a news item that gets read once and forgotten.
No, with one caveat. Google's documentation names the badge on Top Stories, AI Overviews, and AI Mode only. ChatGPT, Claude, Perplexity, and Copilot run entirely separate retrieval systems, with no connection to a Google user's saved preferences. Gemini is less clear-cut: its Google Search grounding option pulls from the same search infrastructure that powers AI Overviews, but Google has not confirmed whether the preferred-source signal itself carries into grounded Gemini answers. Treat that link as unconfirmed rather than ruled out, and keep tracking across platforms regardless.
No. Google describes it as a personalized signal, not a universal ranking factor. It changes what an individual opted-in reader sees across Top Stories, AI Overviews, and AI Mode. It does not raise a site's position for readers who have not selected it, and it runs alongside the standard ranking algorithm rather than replacing any part of it. A site can be highly ranked without a single preferred selection, and vice versa.
No. Only domain-level and subdomain-level sites are eligible. A site at example.com or blog.example.com can qualify, but a subdirectory such as example.com/blog cannot appear as a separate preferred source in the tool. Check eligibility directly at google.com/preferences/source by searching for the site's domain before building any implementation, since a subdirectory setup will need a different approach entirely.
Readers can still select any eligible site as a preferred source through Google's own settings page, even without the button on that site. The button removes a friction point by letting the request happen on the publisher's own page, instead of sending the reader to a separate Google screen. That tends to improve adoption, but the button itself is not a requirement for eligibility or for readers to add the site.
Not necessarily. The badge reflects a personalized signal tied to one reader's saved preferences and one specific query. Its appearance depends on relevance to that query as much as on the preference itself. A site being marked preferred increases the likelihood of prominent placement and labeling, but it does not guarantee a badge shows up on every single mention that site earns going forward.
Structured data and llms.txt are technical signals that help a crawler or model understand and access a page's content. Preferred Sources is a reader-driven trust signal, set through Google's own interface, unrelated to a site's code or markup. The two work at different layers of the same problem. One helps a model retrieve and understand content, and the other reflects human trust that Google then surfaces back inside its own results.
No. Google has not published a dashboard or Search Console dimension that reports individual selections or aggregate counts specific to one site. A publisher can track button clicks through its own analytics tooling, but anything downstream of that click happens inside Google's own systems. None of it is currently exposed back to the site, so a publisher is working from click volume alone, not confirmed selections.
The standard implementation loads asynchronously, through the async attribute on the script tag. That prevents it from blocking the rest of the page from rendering. As with any third-party script, it still adds a network request and a small amount of execution time. It is worth testing page speed before and after adding it, particularly on templates already close to their performance budget.
.svg)
.svg)
