Google pipes Preferred Sources into AI Overviews

What happened

Per Search Engine Journal, Google users have now selected 345,000 Preferred Sources, and Google is extending that personalization signal into AI Overviews and AI Mode. The feature, which lets searchers tell Google which publishers they want to see more often in Top Stories, will now influence the sources cited inside generative answers and a new set of link carousels.

Search Engine Journal reports that the expansion arrives with refreshed UI surfaces inside AI Overviews, including link carousels that visually elevate Preferred Sources when they are relevant to a query. The opt-in is user-driven: people pick the outlets they trust, and Google biases its outputs accordingly.

This is the first time Google has wired an explicit user preference signal directly into the citation logic of its generative answers. That detail matters more than the 345,000 number.

Why it matters for your brand

Preferred Sources changes the citation game from a pure ranking problem into a subscription problem. Until now, getting cited in an AI Overview was a function of authority, freshness, and topical match. Now there is a second path: convince a user once to add you to their Preferred Sources, and you carry a persistent advantage into every future generative answer they trigger. That is closer to a newsletter list than a search ranking.

For financial services brands, this reframes what "thought leadership distribution" means. If a CIO at a pension fund adds the FT, Bloomberg, and your house view publication to their Preferred Sources, your research gets pulled into AI Overviews for queries you would never have ranked for organically. The implication: the audience capture event is no longer the click, it is the opt-in. Treat Preferred Sources status the way you currently treat newsletter subscriptions, with a dedicated CTA and a dedicated metric.

For multilaterals and policy institutions, the calculus is sharper still. UN agencies, the World Bank, IMF, and OECD do not compete on SEO volume; they compete on being the cited authority when a policymaker asks an AI assistant a substantive question. Preferred Sources gives these institutions a way to lock in citation share with the exact audiences they care about: ministers, regulators, development finance officers. A 50,000-person opt-in base inside a policy elite is worth more than 5 million monthly organic visits from a general audience.

Industrial groups have a different problem. Holcim, Siemens, ArcelorMittal: these brands rarely show up in consumer-facing AI Overviews, but they show up constantly in B2B procurement, ESG, and engineering queries. The buyers running those queries are a small, definable population. If your sales engineers and your procurement counterparts add your technical content library to their Preferred Sources, you have effectively created a private citation channel inside Google's public AI surface. That is a defensible moat in a category where every competitor is publishing similar whitepapers.

Philanthropic and policy institutions should read this as a permission structure. Foundations have long struggled to get cited in mainstream search results for the issues they fund, because the SEO-optimised content farms outrank them. Preferred Sources lets a foundation say to its grantee network, its policy partners, and its donor base: add us once, and our analysis becomes the default reference inside your AI search. That is the closest thing to earned distribution that Google has shipped in a decade.

The content strategy implication is concrete. Stop optimising every page for the broadest possible query. Start producing a smaller set of authoritative reference pages that justify the opt-in, and run an explicit campaign to drive Preferred Sources additions among your named accounts. Measure it. The CMOs who win the next 18 months will be the ones who can report Preferred Sources opt-in counts to their boards the way they currently report MQLs.

The signal in context

Google is doing two things at once with this release. First, it is responding to publisher complaints that AI Overviews cannibalise traffic without rewarding source quality, by giving publishers a way to be selected directly. Second, it is borrowing a page from the personalization playbooks of ChatGPT and Perplexity, both of which already let users pin or favour specific sources. The 345,000 figure is small in absolute terms (Google handles billions of queries a day), but it represents a self-selected, high-intent cohort. These are the users who care enough about provenance to configure it. That is exactly the audience B2B brands sell into.

The broader trend is the splintering of the citation surface. A year ago, "getting cited in AI" meant winning on a single set of model-side ranking signals. Now there are at least three parallel mechanisms: training-data presence, real-time retrieval ranking, and explicit user preference. Each requires a different content investment and a different distribution motion. Brands that treat all three as one problem will underperform brands that staff and measure them separately.