Google just made AI Overviews a popularity contest

Google announced yesterday that Preferred Sources — the feature that lets users mark specific publishers as their favourites — now affects which links surface inside AI Overviews and AI Mode responses. Users have selected over 345,000 sources, up from 90,000 in December. Google says people click Preferred Sources at twice the rate of other links inside AI responses.

Read past the announcement framing and what Google has actually done is bolt a follow-graph onto AI search. Quietly. With almost no industry reaction yet.

This is bigger than the press release makes it sound. It's the first time Google has admitted that the AI Overviews citation set is not purely an algorithmic output. It's now partly a function of who the user has subscribed to. Which means the question every business should be asking — how do we get cited in AI answers? — just acquired a second answer that has almost nothing to do with the technical SEO conversation the industry has been having.

The answer is: get people to follow you.

What Preferred Sources actually does now

The mechanics are simple. Users can star sources they trust inside Google Search. Those starred sources get a visible label when they appear, and Google now confirms they also appear more prominently inside AI Overviews and AI Mode responses.

Google has been careful to say this doesn't override ranking systems. John Mueller addressed exactly this recently. The feature works "alongside" quality signals rather than replacing them. Fine. But "alongside" is doing a lot of work in that sentence. If two articles are roughly equivalent on traditional ranking signals and one is in the user's Preferred Sources list, the one the user has followed wins the citation slot. That's what "twice the click rate" implies in practice.

The 345,000 figure also matters more than it looks. Quadrupling in roughly six months is not a feature gathering dust. It's a feature publishers are actively promoting to their audiences — and audiences are using.

The quiet platform shift nobody named

Look at what's actually happened across the last fortnight if you stack the announcements together.

AI Overviews is no longer purely an algorithmic output. It's an algorithmic output plus the user's follow-graph plus the user's inbox.

iPullRank tested Google's Personal Intelligence and found that Gmail signals lifted brand citations in AI Mode by 46 percentage points. Brands seeded into a user's inbox went from appearing in 23.9% of responses to 66.8%.

Now Preferred Sources joins it. Explicit follow signals. Two layers of personalisation, both feeding the citation engine, both running on private user data the rest of the web can't see or measure.

That's not a tweak. That's a different kind of search.

AI Overviews is no longer purely an algorithmic output. It's an algorithmic output plus the user's follow-graph plus the user's inbox.

The thing that's strange about this is how little anyone is treating it like the architectural shift it is. We've spent eighteen months arguing about schema and structured data and llms.txt files. Meanwhile Google has been quietly grafting a personalisation layer onto AI citations that makes a chunk of the conversation moot. Your beautifully optimised page can lose to a worse page that the user happens to follow.

What this means for how AI search actually works

The implications fan out in three directions, and none of them are getting enough airtime.

The discovery question is now partly an audience question. If Preferred Sources doubles your citation odds inside AI responses, then growing a base of users who follow you is now a direct AI search optimisation tactic. Not an indirect brand-building benefit. A direct, measurable lift on whether you appear in answers. This is the kind of thing the industry has been quietly assuming for years — that brand strength matters in AI citations — but it's now mechanised. Coded into the product. You can point at it.

The measurement problem just got worse. I've written before about how the AI search measurement layer is being built by outsiders because Google isn't supplying one. Preferred Sources makes the measurement gap structurally worse. Two users running the same query through AI Mode can now get different citation sets based on private follow data that no third-party tool can see. Rank tracking for AI search was already a polite fiction. It's now a fiction that varies by user in ways that can't be reconstructed from the outside.

The "follow us on Google" prompt is back. Sites used to badger users to follow them on Google News. That faded. Expect it to return, this time pointed at the broader Preferred Sources flow. Publishers are already running these promotions — Google literally cites this as the reason for the 4x growth in selections. Watch this become a standard footer module on any site that takes AI visibility seriously.

The honest read on this

Two things can be true at the same time.

First, Preferred Sources is genuinely a useful feature for users. The web is full of slop. Letting people elevate the publishers they trust is the kind of thing Google should have built years ago. The 2x click-through rate isn't a manipulation — it's evidence that personal trust signals carry real information.

Second, it shifts the AI search optimisation problem in a direction that favours brands with existing audiences. If you already have a base of engaged readers, you can convert that base into AI citation lift by getting them to add you as a Preferred Source. If you're a new entrant with no audience, the playing field just tilted further away from you.

This is consistent with the pattern across the entire AI search transition. The big keep getting bigger. The brands AI systems have heard of get cited more. The publishers users have followed get cited more. The Reddit threads get cited disproportionately because they're loud. None of this is news. What's new is that Google is now openly building product features that codify the advantage.

What to actually do about it

This is a small piece of a larger discipline, so don't let it dominate your strategy. But three things are worth doing this week if you publish anything that could conceivably appear in AI Overviews.

Add a Preferred Sources prompt to your site. Google has documentation on how to encourage selections. A clean, non-intrusive nudge in a sidebar or footer is the bar. If your readers value you, some of them will click. Each click is, increasingly literally, an AI citation vote.

Audit whether you have any audience worth prompting. This is the harder question. Most B2B sites in particular have functionally zero direct audience — they have search traffic that comes and goes. Direct audience, the kind that opts to follow you, is now a meaningful AI search asset. If you don't have one, building one moves up the priority list.

Stop treating AI search visibility as a purely technical problem. Schema markup, semantic HTML, clean information architecture — all still matter, and I'd argue they matter more than most people pretend. But the citation layer above all of that is increasingly being shaped by audience signals that no technical audit will surface. The machine-readability work is foundational, but it's not the whole job anymore.

The bigger pattern

Step back and what you're watching is the slow conversion of AI search from a content-evaluation system into a content-plus-relationship-evaluation system. Gmail signals. Photos signals. Preferred Sources. Sundar Pichai hinting at subscription-based source preferences. Each one a private input that the open web can't see, can't measure, and can't optimise against directly.

The brands that win this transition will be the ones that built audiences before the audience itself became a ranking signal. Everyone else gets to compete on the technical surface and hope it's enough.

That's the loop. And Google just made it visible.