Google discloses the vulnerability of its search index, rankings, and live results

The head of Google Search has warned a federal court that forcing the company to share its search index, ranking data, and live results with competitors would cause “immediate and irreparable harm” to Google, its users, and the open web.

The warning comes from an affidavit filed by Elizabeth Reid, Google’s vice president and head of Search, that was filed with Google’s proposal to halt key antitrust remedies while it appeals the final decision in the DOJ’s search case alone.

The filing specifies what Google considers its most sensitive Search assets and why sharing them would expose proprietary systems, enable reverse engineering, and fuel spam.

Disclosure of Google’s web search index

Fighting: Section IV of the final decision will compel Google to provide “qualified competitors” with a one-time dump of core web index data at a lower cost. That data will include:

• All URLs in the Google web search index
• DocID-to-URL mapping
• Clear time data
• Spam scores
• Device type flags

Google Conflict: This can provide competitors with the results and insights gathered from over 25 years of targeting work.

Reid defined indexing as the product of proprietary crawling, annotation, and indexing systems that determine which pages enter Google Search:

• “The selection of web pages in the Google search index is the result of more than twenty-five years of continuous investment and comprehensive engineering efforts.”

He warned that simply knowing which URLs Google indexes will allow competitors to skip large parts of the crawl and analysis altogether:

• “Getting a list of URLs in Google’s index will allow Relevant Competitors to stop crawling and analyzing the vast web, and instead focus their efforts on searching only a fraction of the pages that Google has included in its index.”

Metadata such as crawl frequency will reveal how Google prioritizes novelty and demand, he added:

• “Information about Google’s transparency schedule will give competitors insight into Google’s new ownership signals and index structure.”

Included in the affidavit is this image, “Google’s Web Crawling Process and Indexes: Results,” which shows that Google labels most web pages as “Spam, Duplicates, and Low Quality Pages.”

• Google specified the changed number of pages in the billions. As of 2020, Google’s index contained about 400 billion documents, according to the testimony of Pandu Nayak, Google’s CEO.

Risk of spam, harassment, and reputational damage

Anxiety: Google argues that disclosing spam scores, even indirectly, would weaken its ability to fight webspam.

Effective spam fighting depends on stealth, Reid stressed:

• “Anti-spam depends on secrecy, as outside knowledge of anti-spam methods or signals destroys the value of those methods and signals.”

If spam points are leaked or compromised, bad actors can use them to bypass Google’s defenses, Reid warned:

• “Spammers … can bypass Google’s spam detection technology and strengthen Google in its anti-spam efforts.”

That can push very low-quality and misleading content into the search results, users ultimately blame Google:

• “Mandatory disclosure is likely to cause more spam and misleading content to appear in user queries, endangering users’ safety and tarnishing Google’s reputation as a trusted search engine.”

Disclosure of user-side search data (Glue and RankEmbed)

What you need is a judgment: Ongoing sharing of “user-side data” used to run Google Glue and RankEmbed models. Reid says the data includes:

• Questions
• Location
• Time to search
• Clicks, scrolls, and other interactions
• All the results and the search feature displayed, and their order

Glue captured 13 months of US search logs, according to the affidavit.

Google Conflict: This would equate to a large, continuous exposure to Google’s ranking output at scale.

• “The disclosure of Glue’s training data is similar to the disclosure of Google’s intellectual property, because it reveals the output of Google’s search technology in response to all queries issued by a user in the United States over a 13-month period.”

He also warned that the data could be reused directly.

• “Eligible Competitors can easily use the exposed Glue and RankEmbed data as training data for a large language model.”

On privacy, Reid insists that Google will not control the final decisions on anonymity.

• “Google will not have the final authority to make decisions about the anonymization and privacy-enhancing methods that will be applied to user data before it is shared.”

Users will still hold Google accountable for any failures, Reid predicted.

• “Google users are however likely to blame Google for any privacy or security issues arising from data disclosure.”

Marketing of Google search results and features

What is required: Article V would force Google to license and sell core search results to competitors for up to five years, including:

• Organic web results (“ten blue links”)
• The question is being rewritten
• Location, Maps, Photos, Video, and information panels

Google warning: This will expose the live results of its search systems to competitors—and more.

• “The search results and features required to assemble Qualified Competitors are the product of decades of continuous engineering and innovation efforts and many billions of dollars in investment.”

Even with contractual restrictions, Google will lose control, Reid said:

• “Google does not have the ability (as it normally does) to refuse to sell to Qualified Competition.”

Competitors can store, analyze, or reward the data — and third parties can scrape it, Reid warned.

• “Any third party may ‘remove’ marketed results and features from Eligible Competitors’ sites and thereby obtain Google’s own results and features.”

Document. Read it here.

• What it is: Affidavit of Elizabeth Reid (Document #1471, Attachment #2)
• Completed: Jan. 16 at 3:46 pm ET
• Case: United States of America v. Google LLC, No. 1:20-cv-03010 (DDC)
• Purpose: It supports Google’s motion to stay part of the antitrust remedies pending the appeal

Reid testified earlier at the settlement hearing and said the affidavit reflected his personal knowledge as the chief executive officer of all Google Search.