Some interesting points in Google Internal Emails that were made public by DOJ recently

Recently Department of Justice has made public many emails and other internal communications showing what Googlers were discussing about different things. Also there are some interviews of Googlers too like Pandu Nayak, H.J. Kim who for years has been working in Google’s Search engineering teams.

I read through all these documents yesterday and posted what I found interesting on my X account.

Here are all of those X posts

Table of Contents

What Googler Pandu Nayak said in the interview

Googler Pandu Nayak's interview notes which DOJ published on 2th May https://t.co/KLAiNGuwDX have some really interesting points.

– Q* (pronounced "Q star"), Google's measure of quality of a document.

– Google have an internal debugger window which they can use to see signal… pic.twitter.com/VmqfQcJtnR
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Definition of Navboost from Googler Pandu Nayak

Navboost definition via Googler Pandu Nayak 👀 https://t.co/pIut1B9ZXu pic.twitter.com/1yXl9oxDvE
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Google avoids predicting clicks said Googler Pandu Nayak

INTERESTING!! Google avoids simply "predicting clicks" because clicks are easily manipulated and are a poor proxy for enhancing user experience https://t.co/pIut1B9ZXu pic.twitter.com/Tz0D21URMv
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Google is currently re-thinking their search stack from the ground-up wth LLM taking a more prominent role

wow Googler Pandu Nayak told DOJ – "Google is currently re-thinking their search stack from the ground-up with LLM taking a more prominent role. They are thinking about how fundamental components of search (ranking, retrieval, displaying SERP) can be reimagined given the… https://t.co/pIut1BaxN2 pic.twitter.com/s4mowXi0UM
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Impressions and clicks was an early signal but then a Google Engineer developed modification of it for avoiding bias created by link position in search results

Impressions/clicks ratio – early signal but Google found its biased by link position in results. So they developed a modification to avoid bias by link position. Which means there do exist a signal which is thinking of impressions to clicks ratio but its independent of link… https://t.co/pIut1BaxN2 pic.twitter.com/3XJ2l0VTqB
— Gagan Ghotra (@gaganghotra_) May 13, 2025

How Google interpret query seeking specific information via Googler HJ Kim

Interesting!!!! https://t.co/332T21BRdN pic.twitter.com/FgPz1OVz98
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Googler HJ Kim explains three fundamental ranking signals ABC

HJ Kim – there are three fundamental ranking signals ABC.
Anchor (A) – a source page pointing to a target page (links).
Body (B) – terms in the document.
Clicks(C) – historically, how long a user stayed at a particular linked page before bouncing back to the SERP …. Things like… https://t.co/332T21Cp3l
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Googler HJ Kim kind of said raters data is directly used in ranking for determining signal thresholds

Google has been telling us we don't use raters directly for ranking signals meanwhile HJ Kim says in performing regression for majority of signals label data from human raters is used. https://t.co/332T21BRdN pic.twitter.com/XSzYAcp6VO
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Q* page quality and Q is largely static and largely related to the site rather than the query says Googler HJ Kim

Q* page quality and Q is largely static and largely related to the site rather than the query.
No wonder why we have Forbes & others ranking for everything because trustworthiness, authority from Google's perspective is largely at site level rather than it being query relevance… https://t.co/332T21Cp3l pic.twitter.com/UDLdlNkY6h
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Googlers in Gemini team were talking about using anon navboost queries search specific data for training models

🆕Googlers in Gemini team were talking about using anon navboost queries search specific data for training models.
"(e.g. anon navboost queries, which could be huge, given we see mmlu increases from aquarium queries)" said one Googler.

PS – this conversion is from Aug 2024 and… pic.twitter.com/H0QR0SmN4p
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Internal communications show while exploring ways to integrate Gemini in Google Search ‘preserve ads revenue’ was a principle

LOL while exploring ways to integrate Gemini in Google Search one of the principles that Googlers were discussing internally is "Preserve ads revenue" 😄

PS – source an internal presentation from 16th July 2024 and was made public by DOJ on 7th May 2025 https://t.co/iV3D2FHdS9 pic.twitter.com/rTtxD5VgOH
— Gagan Ghotra (@gaganghotra_) May 13, 2025

July 2024 internally Googlers were discussing to have Gemini as a tab in search results

Internally in July 2024 Googlers were discussing having a tab for Gemini in Google Search. https://t.co/LqwgdcejIN pic.twitter.com/zwPgVDy1tv
— Gagan Ghotra (@gaganghotra_) May 13, 2025

What is ‘Search Grounding’ explained one Googler to another in an internal email thread

For Gemini internally Googlers defined 'Search grounding' as check search results before generating the answer. SORRY I've many times said its generate the answer then check results because that's what is highlighted in SGE patent from Google. Probably (just guessing) Google for… https://t.co/CK9AEmDCDP pic.twitter.com/Q87KiPu1tQ
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Why may be Forbes and other large sites rank for everything?

Q* page quality and Q is largely static and largely related to the site rather than the query.
No wonder why we have Forbes & others ranking for everything because trustworthiness, authority from Google's perspective is largely at site level rather than it being query relevance… https://t.co/332T21Cp3l pic.twitter.com/UDLdlNkY6h
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Google Marketing Live (2024) emails talks about pMax

🚨 Google Marketing Live (2024) and emails now made public by DOJ reveals internal frustration about Pmax.

These internal emails are 🍿 #ppcchat

via DOJ https://t.co/ml8ZI8j9nT made public on 9th May 2025. pic.twitter.com/gL8aWRrnjL
— Gagan Ghotra (@gaganghotra_) May 13, 2025

April 2024 Google was discussing a partnership to have Gemini as a preferred model in Apple products

April 2024 Google was negotiating preferential treatment of Gemini in Apple products. "For any generative Al access points on Apple's products …. where Apple chooses to use a third party generative Al service, Apple will refer users to Gemini in response to user prompts".

made… pic.twitter.com/qHbRmqduDa
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Googlers discussing using search data for model training

More about using Search data in training runs for Gemini ….. Googler "we should push search data as much as we can with ranking and other things, And see if we can push our other models without those specifics. Will be extremely interesting to see what ….. with search data… https://t.co/7rEpIM8Fu3 pic.twitter.com/ad1yNWsqid
— Gagan Ghotra (@gaganghotra_) May 13, 2025

Internal slide from Sep 2024 shows launch of Gemini Live in Google Chrome

I'm excited about Google I/O 2025 because an internal slide (Sep 2024) made public by DOJ (on 26 Apr 2025) shows Googe I/O 2025 point and says "Bring Google's personal Al to Chrome desktop w/ Gemini Live".

Slide via DOJ https://t.co/erOPhr6maF pic.twitter.com/R7URnse66J
— Gagan Ghotra (@gaganghotra_) May 13, 2025