How to research with Hypothesis Generation

Use Hypothesis Generation, a multi-agent AI system, to develop a research proposal, accelerate the manual literature review process, and generate novel ideas that you can adapt to your expertise and challenge.

Learn more about the Intended Use Policy for Science experiments by Google Labs.

Access Hypothesis Generation

Access to Hypothesis Generation is currently limited to our interest form.

To express your interest, sign up at labs.google.com/science.
- Important: Submitting your interest does not guarantee access.
If you’re selected, you’ll receive an email notification with further instructions.

If you get access, for the best experience with Hypothesis Generation, be sure to:

Log in to Hypothesis Generation.
Have a stable internet connection.
Use a desktop device.
Use Google Chrome.

Start a new research goal

Translate your research challenge or hypothesis into a structured goal.

On your computer, open Hypothesis Generation.
In the prompt box, explain your research challenge or hypothesis.
- You can use case studies as coreferences for structuring your research goal and understanding expected outputs. Learn how to find case studies.
Click the arrow .
Complete a conversational interview with the Hypothesis Generation agent.
- On the right, under “Interview Progress,” you can check your progress through the specific elements of your research goal. Learn about the elements of effective research goals.
- This agent-based flow guides you through building your run specifications, which makes it easier to iteratively refine your goal.
- When you complete all the elements of your research goal, Hypothesis Generation will prepare your final specifications.
To review your finalized research plan, in the overview widget, click Open.

Find case studies

On your computer, open Hypothesis Generation.
Scroll down to “Explore our Case Studies.”
Click an example you’d like to use.

Tip: We are working directly with Scientists to bring more case studies into Hypothesis Generation.

Understand the elements of effective research goals

The key to effective use of Hypothesis Generation is a well-defined research goal. Frame your requests as precisely as possible. Think of it as a conversation with a highly competent collaborator.

Your research goal will include these elements:

Research Challenge: This is the high-level objective of your study. It should define the problem you are solving, the specific targets you want to hit, and what "success" looks like. Research challenges can take many different forms. For example:
- “What would happen next …”
- Looking for a mechanism of action
- Evaluating a rationale
- Evaluating the top candidates to test
- Looking for a strategy to reverse a biological outcome
- Looking for a rationale of why something would fail
- Requesting a design
- Suggesting an alternative strategy
Focus Area: These are the specific pillars or domains of research that the system will explore to solve your challenge. They act as "lenses" to narrow down the search for solutions.
Preferences: This section defines how you want the research to be conducted and any specific constraints on the solutions. It guides the system toward your preferred technical approach.
Title: Optional but helpful for organizing your goals.
- After you configure a run and define your research challenge, Hypothesis Generation automatically suggests a title.
- For quick visual identification of specific goals, add an emoji.
- To change the title, click More Rename.

Run your research

Important: You’re limited to 3 Standard Runs and 1 Advanced Run in progress concurrently. Hypothesis Generation uses significant computational power per run.

On your computer, open Hypothesis Generation.
In your research goal, at the top right, click Configure Run.
Select your run type.
- Standard Run: Get quicker results when testing and refining your research goal.
- Advanced Run: Get more comprehensive and in-depth analysis to uncover novel breakthroughs. You may also get more nuanced or diverse suggestions.
Click Start run.
- Once the report is ready, you’ll receive an email.

Tip: Your run might take several hours to complete.

Troubleshoot AI credit refunds for failed runs

Important: A run is considered failed only if its status displays as "Failed" under recent runs.

If a run fails to complete, any AI credits used on that run may be refunded to your account. If you were charged for a failed run, reach out to Google One Support to initiate the refund process.

Tip: A run's status may take a moment to sync. If your report doesn't appear immediately after completion, try refreshing your browser.

Review goal reports & ideas

After you define your research goal and start your run, Hypothesis Generation runs an intensive literature review, generates novel ideas through a competitive tournament, and synthesizes potential research directions in a goal report.

Find past research goals

On your computer, open Hypothesis Generation.
Find your research.
- Find recent research: Scroll down to “Recent research.”
- Find all research: At the top left, click Target .
Click a research goal.
- You’ll find your draft, in-progress, and past research goals along with their creation dates.

Review the literature

On the goal report page, click one of these tabs:

Ideas: Review the full leaderboard of generated research proposals.
- To help you identify the highest-quality candidates, proposals are ranked by Elo rating.
- Proposals are also categorized into two actionable buckets: High Potential and Non-Viable.
Knowledge Base: A centralized repository of in-depth technical documentation, research-backed insights, and detailed data designed for quick reference and comprehensive analysis.
Summary: A synthesized overview of the entire research effort
Run Specifications: Review details about the specific parameters and constraints that define a research “run.”

Tip: To dive deeper into specific ideas or evaluate potential trade-offs, on the right, click Open Agent .

Export or share goal reports

To collaborate with colleagues or further analyze your findings, export or share your goal reports.

Export goal reports in NotebookLM

Open your goal report directly in NotebookLM to interact with the findings, ask follow-up questions, and synthesize the report with other research documents.

On your computer, open Hypothesis Generation.
On the left, click Target .
Click a research goal.
At the top right, click Open in NotebookLM.

Share goal reports

Generate and copy a unique link to your goal report.

On your computer, open Hypothesis Generation.
On the left, click Target .
Click a research goal.
At the top right, click Share .
Turn on Enable public sharing.
Click Copy.

Download goal reports

Save a local copy of your goal report to your device for offline review or archiving.

On your computer, open Hypothesis Generation.
On the left, click Target .
Click a research goal.
At the top right, click Download .
Select a download option.

Cite Hypothesis Generation when you publish results

If you use Hypothesis Generation in your research, please use this citation:

@article{Gottweis2026,

title = {Accelerating scientific discovery with Co-Scientist},

ISSN = {1476-4687},

url = {http://dx.doi.org/10.1038/s41586-026-10644-y},

DOI = {10.1038/s41586-026-10644-y},

journal = {Nature},

author = {Gottweis, Juraj and Weng, Wei-Hung and Daryin, Alexander and Tu, Tao and Sirkovic, Petar and Myaskovsky, Artiom and Glowaty, Grzegorz and Weissenberger, Felix and Orlandi, Alessio and Popovici, Dan and others},

year = {2026},

month = May

}

To acknowledge Hypothesis Generation usage in your work, in your methods section, we recommend you include this text:

AI Tool Usage: Hypothesis Generation system was used during hypothesis and experimental design generation and prioritization [date used]

Citation in other formats

APA format: Gottweis, J., Weng, W.-H., Daryin, A., Tu, T., Sirkovic, P., Myaskovsky, A., Glowaty, G., Weissenberger, F., Orlandi, A., Popovici, D., Palepu, A., Rong, K., Tanno, R., Saab, K., Zhang, F., Blum, J., Carroll, A., Kulkarni, K., Tomašev, N., … Natarajan, V. (2026). Accelerating scientific discovery with co-scientist. Nature. https://doi.org/10.1038/s41586-026-10644-y
MLA format: Gottweis, Juraj, et al. "Accelerating Scientific Discovery with Co-Scientist." Nature (2026), https://doi.org/10.1038/s41586-026-10644-y.
Chicago format: Gottweis, Juraj, Wei-Hung Weng, Alexander Daryin, Tao Tu, Petar Sirkovic, Artiom Myaskovsky, Grzegorz Glowaty, et al. 2026. "Accelerating Scientific Discovery with Co-Scientist." Nature, May 2026. https://doi.org/10.1038/s41586-026-10644-y.
Harvard format: Gottweis, J., Weng, W.-H., Daryin, A., Tu, T., Sirkovic, P., Myaskovsky, A., Glowaty, G., Weissenberger, F., Orlandi, A., Popovici, D. et al. (2026) 'Accelerating scientific discovery with Co-Scientist', Nature. doi: 10.1038/s41586-026-10644-y.