Knowledge Graph facts now reach Google AI Mode as full biographies without URLs, according to an analysis of Google Earth code and AI Mode responses. The finding matters because a biography can supply an answer while leaving readers without a page to check. In one test, AI Mode answered that Dan Petrovic was “47 years old” and showed a biography result with no URL.
The result contained a short profile, including a birth year, birthplace, nationality, professions and education. Google’s public Knowledge Graph APIs held a thinner record for the same person, including the description “Australian digital marketer” and an old URL. The article’s authors say the AI Mode biography was richer and more current than the public API record.
Knowledge Graph Facts Appear Without a Link
AI Mode listed two sources: a Wikipedia result with a URL and the biography result without one. Its only citation, a Wikipedia page about a footballer, appeared beside the sentence about Petrovic’s age.
A query about owayo returned an “Organization Biography” with a founding year, three founders, headquarters, product range and relevant webpages. The company record appeared at position 0.12.1. Separately, hidden page data tied Petrovic’s answer to a person entity.
The authors report that the biography result appeared in two separate incognito sessions. They also compare their observation with earlier testing by Natzir Turrado, who found his Knowledge Graph data in AI Mode in 10 of 10 quoted-query tests and 0 of 10 unquoted tests. In the newer test, the Petrovic query had no quotation marks, yet the biography appeared.
What the Code Shows About Knowledge Graph Facts
The team examined Google Earth’s October 15, 2026 build. Registered result types grew from 539 to 542 compared with October 8, and QRS providers rose from 97 to 99. The newer build added translate and cehelpcenter providers.
The researchers counted 142 result types from 99 providers and 268 per-query signal types, including Navboost click data, RankEmbed and AI Mode-related types. The build contains fact-provenance fields for source categories, citation requirements, access controls and license switches.
Those fields describe possible restrictions, not a verified rule for any specific fact. The article says a fact may carry multiple language-model restrictions because the restrictions are stored as a number. It also says sensitive personal facts need a public-information certificate with an attribution URL and last-verified date before display.
The team stresses that the records returned by AI Mode as raw source metadata were written by the model and repeated across sessions. The schemas were bundled as dependencies in the Earth binary. Their presence shows that the data structures exist; it does not prove Google Earth uses them or reveal which flags apply to particular facts.
What Site Owners Can Check
The authors recommend checking the facts associated with an entity and correcting them at their source. They point to Natzir Turrado’s height and Petrovic’s birthplace as details that appeared in the biography but on no cited page. They also suggest reviewing the “Relevant webpages” attached to an entity: the Petrovic record listed dejan.ai, Hugging Face and LinkedIn, while the public APIs listed dejanseo.com.au.
They note that a fact can carry a GOOGLE_EXTENDED_OPT_OUT flag and that a fact may carry REQUIRES_CITATION. The article does not establish that either flag applied to the tested biography. It does describe a practical visibility concern: when Knowledge Graph facts have no link, an AI answer may repeat it without directing readers to the page that supplied it.
For readers comparing how search systems handle entity information, this report connects with coverage of Knowledge Panel reliability, brand credibility in AI results and SEO beyond rankings. The source’s code analysis and AI Mode examples are detailed in the original Knowledge Graph analysis.
