Google’s Preferred Sources
Selects the Site. Not the Byline.
Google’s Preferred Sources lets any reader set standing favorites among the sources they trust inside Top Stories, AI Overviews, and AI Mode — more than 600,000 unique sources selected so far. The developer documentation is precise about what qualifies: domains and subdomains. A subdirectory isn’t eligible.
And a byline is not a unit
the tool can hold at all.
On August 20, 2026, Google handed publishers a button. Per the developer documentation, it takes two lines of HTML, and it turns a reader’s loyalty into a standing setting: click, and, in the words of Google’s own announcement, “it adds the site as a Preferred Source on Google and immediately directs them back to where they left off on the publisher’s page.” Preferred, until the reader says otherwise, in Top Stories, in AI Overviews, in AI Mode. The same announcement counts the constituency the program has already built: “So far, people have already selected more than 600,000 unique sources.” A reader can now tell Google’s answer layer which sources they value most, and the layer will remember. Which makes the specification for whom (the exact shape of what can be preferred) worth reading as closely as anything this record has filed.
The specification is public, plainly written, and current: Google’s developer documentation for the feature, last updated August 20, 2026. It says: “Only domain-level and subdomain-level sites are eligible to appear in the source preferences tool.” It even furnishes the example: www.example.com qualifies, code.example.com qualifies, “but the subdirectory https://www.example.com/blog isn’t eligible.” And there the categories end. This record searched the document for a third one: author, byline, journalist, individual, person. No such eligibility path appears anywhere on the page. The tool built to let a reader choose whom to trust has two units it can hold: a domain and a subdomain. Neither of them is a human being.
Now place a working professional against that rule. A licensed clinician who has published for years under her own name and credential on a major psychology platform writes, architecturally, at a subdirectory: the platform’s domain, then /contributors/, then her name. And that arrangement is the common one: a byline on a publication the writer doesn’t own typically resolves to a path on the publisher’s domain. Psychology Today’s public contributor directory shows the same shape, page after page: contributors — psychologists, psychiatrists, researchers, other credentialed professionals — many with credential initials carried in the page address itself. Under the eligibility rule, those contributor pages cannot be selected separately as preferred sources. If the platform appears in the tool, selecting it records a preference for the publishing site. It does not isolate the clinician whose work prompted the choice.
The specification, read closely
What the tool does. Google’s help documentation describes it directly: “You can customize your Top Stories, AI Mode and AI Overviews experience by managing your source preferences” — “highlight content in your AI Mode and AI Overviews experience from sources you value most.” Selections persist across browsers when the reader is signed in. This is a saved reader preference.
Who qualifies. The documented rule selects sites: “Only domain-level and subdomain-level sites are eligible to appear in the source preferences tool.” Google also warns that sources without regular updates may be unavailable — “Sources that aren’t updated regularly may not be available.” Regular publication does not make an individual contributor page separately selectable.
What that means clinically. In this record’s clinical vocabulary — the mapping is ours, not Google’s — this is Signal Fragmentation: the chain that should run from license to byline to a source a reader can follow breaks at the eligibility rule, so a reader’s preference for that contributor cannot be saved separately from the publishing site. Signal Fragmentation is the primary finding here; the contributing one, distinct from it, is Trust Transfer Failure. The reader’s preference attaches to the publishing site. The setting cannot isolate the contributor whose work prompted that preference.
What this rule is not
Precision matters here, so the boundaries go on file with the finding. This is architecture, not misconduct: nothing is hidden. The exclusion sits in public developer documentation, and this record is quoting it. The claim is scoped to that documentation: the source-preferences tool itself is interactive and session-bound, and this record verified the rule as written, not every behavior of the live tool. The feature is young and Google iterates; the documentation was touched the same day the button shipped, and if the eligibility rule is ever extended, that revision will belong on this record too. And one boundary cuts the other way, and it is the one worth sitting with: the rule’s target is narrower than it first appears. What it excludes is the byline. A practitioner whose work lives at its own domain or subdomain — her own site, even a hosted subdomain — can qualify the way any site qualifies. A contributor page beneath a publisher’s domain cannot qualify separately under this rule. And even then, look at what the tool is holding: her site. The practice can be preferred. This feature selects the site; it offers no separate author-level preference.
Edition No. 056 examined a different instrument: Agentic Resource Discovery v0.91, a specification for finding callable resources. It defined no native person entry, while allowing extensions. That was resource discovery; this is reader preference. The comparison concerns what each documented format represents directly.
Why this reaches past the news tab
The professionals this record is written for mostly do not think of themselves as news sources, and Top Stories is a news surface. But the preference does not stay there: by Google’s own description, it rides into AI Overviews and AI Mode — the surfaces where questions with money and judgment behind them increasingly get answered. So the questions worth testing against a real practice are concrete. Where does your best work actually live — on a unit this tool can hold, or on a subdirectory it cannot select separately? If one of your readers wanted Google to remember trusting you tomorrow morning, is there anything they could check? And when a publisher places the button beside a contributor’s work (part of an update Google frames as “helping publishers build stronger, direct connections with their readers”), every preference it collects will be, by the rule’s own terms, a preference for a site.
That is the honest close. When the button sits beside work written by people whose names the tool cannot hold, the preference it collects can only ever land on the platform. The documented rule provides no separate selection for a contributor page; the form simply has no field where the name would go. The reader meant the practitioner. The record shows what the checkbox meant.
Google Search Central, “Guide to Preferred Sources in Google Search for Web Publishers,” developer documentation, last updated August 20, 2026 — developers.google.com. Eligibility rule, subdirectory example, surfaces statement, and embeddable-button implementation (“by adding only two lines to your HTML”) quoted verbatim; document retrieved and archived September 6, 2026. The document was additionally searched for any author-, byline-, or person-level eligibility path; none appears.
Google Search Help, “Preferred Sources in Google Search” — support.google.com. Product description, signed-in persistence, and the freshness condition (“Sources that aren’t updated regularly may not be available”) quoted verbatim.
Mrinalini Loew (GM, Google Search Ecosystem), “Personalize what you see on Google Search, News, Discover,” Google blog, August 20, 2026 — blog.google. Button behavior, the “more than 600,000 unique sources” figure, and the publisher-connection framing quoted verbatim.
Method note: the contributor-page pattern is cited at the platform level (the public contributor directory at the /us/contributors/ path); no individual contributor is named in this edition, by design. Claims about the source-preferences tool are scoped to Google’s published rule and descriptions; the interactive tool itself was not independently exercised. The clinical mapping in the boxed analysis is this record’s own.