Corsac
Use Corsac
All evals
Google

Eval directory · Search & Knowledge

Evals for Google

Eval coverage for Google, mapped from its public product surface.

About Google

Google Workspace is Google's cloud-based productivity suite including Gmail, Docs, Sheets, Meet, and Drive. Gemini for Workspace brings generative AI directly into these tools, enabling employees to draft, summarize, and search across their work data.

Employees

~182,000

Industry

Cloud Productivity & AI

Headquarters

Mountain View, CA

Website

workspace.google.com

Use the eval library for Google

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

Generate your own →

Related in Search & Knowledge

All evals →

More Search & Knowledge eval libraries

Coverage map

What would you measure for Google?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

  • Adversarial robustness
  • Workflow quality
  • Safety gates
  • Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

01

Search entry points and AI Mode

The Search surfaces reachable from google.com and about.google/products: classic web search, AI Mode responses with follow-ups and web links, Lens camera/image/screen search, Circle to Search on Android, and the How Search Works explainer.

“Circle, highlight or tap to search anything on your Android device.” about.google

Mapped capabilities

4 capabilities

  • AI Mode response with follow-ups and source links

    Answer is returned, follow-up questions continue the same thread, and web links remain available alongside the AI response.

  • Lens visual search across camera, image and screen inputs

    Each documented input mode (camera, uploaded image, on-screen content) is offered and produces a search result path.

  • Circle to Search invocation on Android

    Circle, highlight and tap gestures are presented as documented entry gestures on a supported Android device.

  • Classic search and Images navigation

    google.com and Google Images expose search plus documented links: Advanced search, Advertising, Business Solutions, About Google.

Illustrative example

Input
In AI Mode, ask "what should I look for in a used electric car battery?" then follow up with "what about in cold climates?"
Expected behavior
The follow-up is answered in the same thread with the earlier used-EV-battery context carried over, and the response still surfaces web links alongside the AI answer rather than replacing them.

02

Gemini assistant workflow

The Gemini assistant as marketed on about.google/products: chat with a personal assistant, Gemini Live conversation about surroundings or screen content, and Canvas for going from prompt to a working prototype.

Mapped capabilities

4 capabilities

  • Multi-turn chat continuity

    Assistant carries context across turns within a conversation started from the Try it now entry point.

  • Gemini Live conversational session

    A voice conversation can be started and scoped to what is around the user or on their screen, per the page's description.

  • Canvas prompt-to-prototype

    Canvas accepts a prompt and produces one of the documented artifact types: game, app or infographic.

  • Entry from product pages into the assistant

    Try it now, Talk it out and See how it works CTAs resolve to the corresponding Gemini surface.

03

Generative creation tools

AI creation surfaces named in the context: image generation from ideas, video generation from photos, and Flow for cinematic clips, scenes and stories, presented under new AI products and experiments.

“Create cinematic clips, scenes and stories with our AI filmmaking tool.” about.google

Mapped capabilities

4 capabilities

  • Text-to-image generation

    A text idea yields images through the Create entry point.

  • Photo-to-video clip generation

    A user photo is transformed into a short video clip as the page describes.

  • Flow filmmaking flow

    Flow accepts a creative prompt and produces clips, scenes or a story sequence.

  • Experimental-status framing

    Tools listed as new products and experiments are presented as such rather than as settled products.

04

Business Profile setup and upkeep

The free Google Business Profile funnel at business.google.com: creating a profile, controlling how it appears on Search and Maps, and the ongoing data-management work of hours, contact info, photos, posts and reviews.

“Take control of how your free Business Profile appears on Google Search and Maps.” business.google.com

Mapped capabilities

4 capabilities

  • Create-your-profile entry and claiming

    Create your profile leads into the documented setup path for appearing on Search and Maps.

  • Business information accuracy and updates

    Hours, contact information and other business details can be edited and reflected on the profile.

  • Reviews and response handling

    Customer reviews are surfaced, can be responded to, and ratings can be highlighted.

  • Vertical-specific merchandising

    Retail, services and restaurant paths expose the documented options: products in stock, ordering, reservations, bookings, pickup or delivery.

06

Devices, research and policy pages

Trust and breadth surfaces from about.google and google.com footers: the Android ecosystem and Pixel portfolio (phones, watches, earbuds, tablets), Maps updates, AI model and research pages including open-sourced WeatherNext 2, and the Terms and Privacy policy destinations.

“Our WeatherNext 2 AI model demonstrated a massive leap forward in predicting cyclones.” about.google

Mapped capabilities

4 capabilities

  • Android and Pixel portfolio browsing

    See what's new and Shop now reach Android feature and Pixel lineup pages covering the listed device categories.

  • AI model and research explainers

    Model capability and research pages, including WeatherNext 2 cyclone forecasting and its open sourcing, are reachable from About Google.

  • Terms and Privacy resolution

    Footer Privacy and Terms links resolve to policies.google.com equivalents.

  • Accessibility and account support entry

    Product support, Google Account and accessibility support links are reachable from the products page.

Coverage is mapped from Google's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Google test?+

The coverage map is generated from Google's own public product surface (consumer search, AI assistant and advertising platform): 6 scoring areas — Search entry points and AI Mode, Gemini assistant workflow, and Generative creation tools, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Google evals scored?+

Every case generated for Google — across Search entry points and AI Mode and Gemini assistant workflow and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Google library include?+

The full Google library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, AI Mode response with follow-ups and source links and Lens visual search across camera, image and screen inputs under Search entry points and AI Mode); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Google or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Google areas and set them up in a Corsac workspace, where you can run every test case against Google or your own agent with your own data.