What happened
On August 20, 2026, Google announced new personalization features across Search, Discover, and Google News. The Search update gives publishers an interactive Preferred Sources button they can place on their own pages. When a reader selects it, Google says the site is added as a preferred source and the reader returns to the publisher page.
Google says Preferred Sources can affect Top Stories, AI Overviews, and AI Mode for users who choose a source. Its Search Central documentation, last updated August 20, says domain-level and subdomain-level sites are eligible, while subdirectories such as a standalone blog path are not. The same documentation gives publishers a standard JavaScript button, an advanced implementation, and a deeplink option.
Google also announced a Discover change. In the coming days, users will be able to tap the three-dot menu in the Google app and describe the topics or links they want to see more or less of. Google says the feed will adjust and remember those requests.
For Google News on Android, Google says users can now curate daily audio briefings by choosing topics, with source attribution and links to full articles. The Verge reported the same day that the Discover feature uses a chatbot-style interface and that the Preferred Sources button lets readers add a publication without leaving the page.
Those are the reported facts and primary-source statements. REC's read is about expert publishing. Google is giving readers more explicit ways to choose sources, tune feeds, and listen to AI-shaped briefings. That makes source quality and source memory more visible parts of distribution.
Preference is earned before the button
A Preferred Sources button is useful only after the reader has a reason to prefer the source. The button can reduce friction. It cannot create trust, voice, usefulness, or recall by itself.
For a publisher, creator, consultant, or founder, the practical question is smaller than a Google strategy deck. What would make someone want this source to reappear in Search, Discover, an AI answer, or an audio briefing?
A reader may prefer a source because it explains a topic clearly, names evidence, corrects itself, has a useful point of view, or keeps showing up with material that feels hard to replace. Those qualities start inside the editorial workflow before they reach any feed.
This is where expert content gets interesting. The source is often a person with real judgment, not a newsroom brand or an anonymous content operation. A founder explains why a product choice was hard. An operator explains the constraint behind a process. A researcher separates a result from an inference. A creator explains the taste or method behind a recommendation.
If that thinking is captured only as a polished post, the source can become thin. If it is captured as a recorded answer, transcript, research brief, source links, and approval note, the team has something worth reusing across search, video, audio, and social formats.
AI feeds separate claims from formats
Google's announcement touches three surfaces at once: Search, Discover, and News audio. The same source may show up as a link in Top Stories, a highlighted source in AI Overviews, a Discover recommendation, or a spoken briefing with attribution.
That shift matters because the finished format is no longer the whole experience. A written article may become part of an AI answer. A video may be discovered through its transcript or surrounding page. A newsletter idea may later become a cited web page. An audio briefing may point back to the original source.
REC's analysis is that teams should make the source record strong enough to survive that movement. The reader may encounter the claim in one place and inspect it somewhere else. The claim needs a person, a context, evidence, and a clear boundary.
AI can help prepare that work. It can collect background links, draft better interview questions, clean transcript punctuation, propose headings, identify candidate clips, and adapt an approved answer into a summary. Those uses are support roles.
The claim still needs to trace back to the person or source that supplied it. If an AI-shaped feed repeats a sentence, the team should be able to find the recording, transcript passage, public source, or review note behind it.
What expert teams should do
First, publish answers that deserve repeat attention. Do not ask readers to prefer a source that only produces interchangeable commentary. Give them clear explanations, specific examples, practical limits, and source links.
Second, connect video to crawlable text. If the useful idea starts in a recorded interview, turn the answer into a page with a clean title, transcript-backed sections, source references, and an embedded or linked clip where the face and voice add context.
Third, keep the internal record. Save the raw recording, transcript, research brief, source list, AI role, edits, and approval note. That record helps the team reuse the answer without changing the meaning.
Fourth, make source ownership obvious. If the point came from the expert's experience, say so plainly. If it came from reported facts, cite the source. If the final section is analysis, label it as analysis instead of letting it sound like a reported fact.
Fifth, treat preference as a relationship, not a hack. A reader who adds a source is making a small bet that future material will be worth seeing. The right response is consistency, not volume.
What to avoid
Do not confuse a follow button with a content strategy. The button can help a reader act on trust that already exists. It cannot make thin material useful.
Do not optimize only for AI summaries. Google's own materials describe links, badges, Discover preferences, audio briefings, source attribution, and full articles. The system still points back to source material that humans can inspect.
Do not publish source-free opinions and expect personalization to rescue them. A strong point of view still needs evidence, context, and responsibility.
Do not let AI rewrite the expert into a generic publisher voice. If the source is a person, preserve the person's claim, cadence, caveat, and reasoning. Clean the answer without replacing it.
Do not measure only whether a page appeared in an AI answer. Ask whether the cited page represents the real source, whether the reader can verify the claim, and whether the piece gives someone a reason to come back.
The practical takeaway
Google's August 20 update is timely because it turns source preference into a more explicit product action across AI-shaped discovery surfaces. Readers can name sources they want to see. Publishers can make that choice easier. Feeds can respond to described interests.
For REC's world, the useful response starts before distribution. Record the human answer. Keep the transcript. Attach the sources. Mark what AI did. Review the final wording against the original answer. Then publish the material in a format that people can read, watch, cite, and choose again.
A preferred source needs a source record behind it.