SOURCE

An Anonymous Source Shared Thousands of Leaked Google Search API Documents with Me; Everyone in SEO Should See Them

Properties1
tagsseo, google, leak, navboost, click-data, chrome, brand-signals, rand-fishkin
AUTHOR
Rand Fishkin
PUBLISHED
2024-05-27
ORIGINAL
view original

5 min read

An Anonymous Source Shared Thousands of Leaked Google Search API Documents with Me

Author: Rand Fishkin (NEW: rand-fishkin) | Published: 2024-05-27 | Source: SparkToro · article

Summary

Rand Fishkin’s May 2024 SparkToro post is the primary public disclosure of Google’s leaked Search Content Warehouse API docs: 2,500+ pages, 14,014 attributes, committed to GitHub 2024-03-27 and taken down 2024-05-07. An anonymous source (identified 2024-05-28 as SEO practitioner Erfan Azimi) brought Fishkin the leak plus claims that Google collects clickstream, Chrome, and related signals despite years of public denials. Ex-Googlers told Fishkin the docs “look legit”; Mike King (iPullRank) reviewed them and published a companion technical analysis.

Fishkin is explicit about the limit: the docs show what Google stores and can call, not weights, and not proof every field is used in live ranking. Deprecated flags exist; unmarked fields were likely still in the docs as of ~August 2023. AI Overviews do not appear. “Don’t point to a feature and say SEE, that’s proof it ranks.” Still, it is the largest Search-division leak in 25 years and it corroborates DOJ testimony (Pandu Nayak on NavBoost/Glue).

Five early findings: (1) NavBoost / Glue store good/bad/lastLongest clicks, impressions, squashed vs. unsquashed clicks, geo- and device-segmented click data, and host-level quality evaluations — click/CTR/long-vs-short-click user data exists as named features. (2) Chrome clickstreams appear in metrics used for things like sitelinks (chrome_trans_clicks / top URLs). (3) Whitelists for travel, COVID local authorities, and election authorities. (4) Quality-rater / EWOK ratings show up in modules; some notes say evaluation pipelines only. (5) Click data helps bucket links into low/medium/high-quality indexes: no-click URLs can have their links ignored; high verified Chrome clicks can put a URL in the high-quality link graph so PageRank/anchors can flow.

Strategic takeaways Fishkin owns as interpretation: brand/entity popularity beats classic on-page SEO; E-E-A-T may be “80% propaganda, 20% substance” (almost no topical-expertise field; authors-as-entities exist); NavBoost/user-intent can outweigh links and content; PageRank still exists in multiple evolved forms but has waned; page titles still matter; for most SMBs, SEO returns are poor until navigational demand exists. He urges the trade press to stop uncritically repeating Google statements.

Key Claims

  • Leak authentic in appearance: internal API style, ex-Googler review, King analysis; 14,014 attributes. (reported; later confirmed by Google outside this extract)
  • Stored ≠ ranking. Docs are an inventory, not a ranking recipe. (Fishkin’s explicit caveat)
  • NavBoost stores click types, impressions, long vs. short clicks, geo/device splits, host-level quality. (from the docs + DOJ overlap)
  • Chrome user clickstreams are available as features (e.g. sitelink top URLs). (from the docs)
  • Click data tiers the link graph; low-click URLs’ links can be ignored rather than trusted. (source claim + cited modules)
  • Whitelists for travel / COVID / elections. (from the docs)
  • EWOK rater data exists in the warehouse; some modules mark it evaluation-only. (from the docs)
  • Fishkin interpretation: brand and navigational demand dominate; E-E-A-T is oversold as a direct factor. (opinion)

Notable quotes

“I would urge readers not to point to a particular API feature in this leak and say: ‘SEE! That’s proof Google uses XYZ in their rankings.’ It’s not quite proof. It’s a strong indication, stronger than patent applications or public statements from Googlers, but still no guarantee.”

Why citable: The methodological rule for using this leak in the wiki. Stored attributes evidence capability, not live weight.

“We already know, one signal could be more powerful than the whole big system on a given metric. For example, I’m pretty sure that NavBoost alone was / is more positive on clicks … than the rest of ranking”

Why citable: Google VP Alexander Grushetsky, quoted by Fishkin from a 2019 internal email. Independent-of-Caleb evidence that click/user systems can dominate.

“I’m a bit worried that E-E-A-T is 80% propaganda, 20% substance.”

Why citable: Fishkin’s read of the leak vs. the public quality story. Pair with official “E-E-A-T is not a specific ranking factor” — they rhyme, from opposite sides.

Connections

Entities mentioned: Google · Rand Fishkin (NEW: rand-fishkin), Mike King / iPullRank (minor), Erfan Azimi (minor) Concepts referenced: Brand & User Signals as Ranking Drivers, Local Link Building & Authority, Entity-Based SEO, GEO / AEO (Getting Recommended by AI) · E-E-A-T (NEW: e-e-a-t)

Contradictions / Tensions

  • Independent source for Brand & User Signals as Ranking Drivers. Dooley claims traffic/CTR/dwell are #1. Google’s public line has long downplayed click-as-ranking. This leak shows named click, impression, long-click, and Chrome-clickstream attributes exist. That is not the same as “CTR is the ranking factor,” and Fishkin forbids that leap. Concept-page update: user-signal data collection is evidenced; weight remains contested. Post-publication, Google confirmed authenticity and said not to assume every attribute is used — which is the same caveat Fishkin already printed.
  • Vs. A guide to Google Search ranking systems page-level ranking. Leak has host-level NavBoost quality. Official: site-wide signals exist and do not lift every page. Compatible if host scores are classifiers, not a blanket boost.
  • Vs. official “raters don’t rank” (Creating helpful, reliable, people-first content). Leak: EWOK ratings are stored; some modules say evaluation pipelines. Do not collapse into “raters directly rank.”
  • Vs. “links are dead” (Dooley) and “links are the post-2023 quality filter” (Caleb). Leak: links still exist, but click-qualified into tiers; unclicked links may be ignored. Closer to “links that users actually use” than to either slogan.
  • E-E-A-T 80/20 vs. industry E-E-A-T productization. Aligns with official “not a specific ranking factor.”

Notes

Primary leak-disclosure source. Do not ingest weights or “siteAuthority as proven factor” from this page alone — Fishkin does not document a siteAuthority field here; industry catalogs of specific attribute names (including siteAuthority) live in King’s companion analysis, not in this extract. Date: 2024-05-27; source identified 2024-05-28.