Build an AI Policy Brief in Gemini

Create a citation-checked AI policy briefing with Gemini Notebook.

Build a Responsible AI Policy Brief with Gemini Notebook

30 Second Summary

A confident answer can sound convincing even when the evidence behind it is thin. That becomes risky when a school or workplace uses it to decide how people may use AI.

In this project, you will build a responsible AI briefing pack for a school or workplace learning team. You will use source grounding to expose an evidence gap before comparing guidance from three public institutions.

What You'll Build

Picture yourself showing a learning team a policy briefing whose recommendations open straight to the public evidence behind them in Gemini Notebook.

By the end of this project, you'll have:

A citation-checked policy comparison that lets you open inline citations to inspect supporting passages from NIST, UNESCO, or the U.S. Department of Education.
A source-coverage demonstration that makes the one-source evidence gap visible. A repaired answer then draws on all three organizations.
A review-ready briefing pack with a written briefing, an explorable mind map, and a short Audio Overview you can play.
Secret Mission: Generate a formal audio debate and fact-check one claim against the original source passage.

Are there any prerequisites?

You need a Mac with internet access plus a work or school Google account that has Gemini Notebook enabled. Safari is included with macOS, so you do not need to install anything.

Before We Start

Your project environment is ready in Safari with Gemini Notebook access through your work or school account. This step is your moment to commit to the briefing purpose before the hands-on research begins.

Create Your Evidence Notebook

Your briefing purpose now identifies its audience. It also names the responsible-AI decision they need to make.

A reliable briefing needs a focused workspace. It also needs a known starting source before any synthesis begins.

Gemini Notebook uses source grounding to keep its responses tied to the evidence you provide. You will create the workspace in Safari with the NIST AI Risk Management Framework as its starting source.

In this step, get ready to:

Access Gemini Notebook with your work or school account.
Create a dedicated notebook for the policy briefing.
Add the NIST framework as the first processed source.

Open Gemini Notebook in Safari

Your organizational account controls access to Gemini Notebook. Using the account you confirmed earlier keeps this project within your existing access.

Press Cmd+Space (macOS) or the Windows key (Windows) to open your search bar.
Type Safari and press Enter to open it.
Enter https://notebook.google.com/ in Safari's address bar.
Press Enter to load Gemini Notebook.
Sign in with your work or school Google account if prompted.
Confirm that Gemini Notebook shows your work or school account as signed in.

Gemini Notebook loads under your organizational account. Your recorded briefing purpose remains available to guide the research.

Can't Access Gemini Notebook?

Confirm that you selected the same work or school Google account whose access you checked earlier.
Contact your administrator if your managed account does not have access.

Ask for help diagnosing the account access problem.

Start a dedicated policy workspace

A dedicated notebook keeps the evidence for one policy decision together. This makes the notebook's source coverage visible from the start.

Select Create new notebook.
Select Upload a source.

The source upload dialog opens for your new policy briefing notebook.

Add the NIST starting source

A web source lets the notebook ground its responses in an institution's published material. The source title creates a visible record of the evidence currently in scope.

Choose the option for adding a website.
Paste https://www.nist.gov/itl/ai-risk-management-framework into the URL field.
Use the dialog's confirmation control to add the source.

Source processing can take a moment. A temporary processing state is expected.

Keep the notebook open until the source finishes processing.

Before you check, what would prove that the notebook has finished turning the NIST page into usable evidence?

Inspect the Sources panel for the processed source.
Inspect the Chat panel for the generated source summary.

You will see AI Risk Management Framework in the Sources panel. The Chat panel displays its generated source summary.

Why Start With NIST?

The NIST page provides broad risk-management guidance from a public institutional source. It gives your policy research a controlled evidence baseline.

Don't See the NIST Source?

Confirm that the URL matches https://www.nist.gov/itl/ai-risk-management-framework exactly.
Retry the import if the source never appears in the Sources panel.
Wait for processing to finish before checking the Chat panel again.

Ask for help troubleshooting the source import.

That is your evidence baseline in place. Your notebook can now ground its next response in the NIST source.

Your evidence notebook is ready. Next, you will ask a grounded policy question and trace its claims back to the source.

Ask and Verify a Grounded Question

Your Gemini Notebook now contains the NIST AI Risk Management Framework. Its generated summary gives your policy briefing a starting point.

A useful AI answer is not enough on its own. You need to trace its claims to the source before trusting it.

Inline citations connect an answer to supporting passages. You will verify one of those links before preserving the response.

In this step, get ready to:

Ask a focused question about the NIST framework.
Trace an inline citation to its supporting passage.
Save the checked response as a citation-linked note.

Ask a source-grounded question

Source grounding keeps the answer tied to material inside your notebook. Your prompt also requests a citation for every point so each claim can be checked.

Ask Gemini Notebook for cited NIST guidance by pasting this prompt into Chat:

From the NIST AI Risk Management Framework source, identify the framework's main purpose and three trustworthiness considerations. Cite every point.

What does this prompt test?

The request for the main purpose checks whether Gemini Notebook can summarize the framework's overall goal.
The request for three trustworthiness considerations gives the response a specific evidence target.
The instruction to Cite every point makes each claim easier to audit.

Submit the prompt by pressing Enter.

You'll see an answer describing the framework's purpose plus three trustworthiness considerations. Each point should include an inline citation.

No Citations in the Response?

Confirm that AI Risk Management Framework remains listed in the Sources panel.
Paste the exact prompt again if the citation request was omitted.

Ask for help checking why the response has no citations.

Trace a claim to its source

A citation preview reveals the quoted evidence behind a claim. Opening the citation shows that evidence inside its wider source context.

Hover over at least one inline citation in the response to reveal its quoted text.
Select the same citation to inspect the supporting NIST passage in context.

You'll see the supporting NIST passage in context. This proves that the answer's claim has a traceable source.

Citation Passage Not Opening?

Return to the response in the Chat panel.
Select the citation marker attached to the claim.

Get help inspecting an inline citation.

Save the checked response

Saving the response preserves its original format. Its clickable citations remain attached to the research note.

Return to the citation-linked response in Chat.
Select Save to note.

Before the final check, predict whether the saved note's inline citations will still open the NIST source.

In the Studio panel, select the note created from your Chat response.
Select one inline citation inside the saved note.

You'll see the complete response preserved in the note. Selecting its citation takes you back to the supporting NIST passage in context.

Saved Note Missing?

Return to the original response in Chat.
Select Save to note again.

Get help finding the saved citation-linked note.

That trust check is complete. Your first policy research note now carries evidence you can reopen.

Your NIST findings are now preserved with their evidence. Next, you will test what happens when the notebook lacks the sources needed for a broader comparison.

Expose the Source Gap

Your saved NIST note proves that Gemini Notebook can use source grounding to connect a focused answer to a real passage. A comparison across three institutions demands broader source coverage.

Every organization in a trustworthy comparison needs supporting evidence from that organization. This step tests whether one broad framework can meet that standard.

In this step, get ready to:

Keep NIST as the only selected source.
Test a three-organization comparison against one-source evidence.
Save the missing evidence as a source-gap note.

Limit the evidence to NIST

A source checkbox controls whether Gemini Notebook can use that source in its answer. Keeping one source active creates a clear evidence boundary for the comparison.

Return to the Sources panel in the responsible-AI policy briefing notebook from earlier.
Confirm that the checkbox beside AI Risk Management Framework is selected.
Confirm that no other source checkbox is selected.

You are in control of the evidence boundary now. Gemini Notebook can draw only from the NIST page for the next answer.

Run the unsupported comparison

Inline citations reveal whether each claim has evidence behind it. A three-organization request gives you a direct way to test the notebook's current coverage.

Enter this comparison request in Chat:

Compare the NIST, UNESCO, and U.S. Department of Education recommendations for student privacy, human oversight, and equity. Use at least one cited point from each organization.

What Does This Prompt Test?

The request requires a cited point from each named organization.
The three policy themes give the comparison a specific scope.
The citation requirement provides an observable test of the selected evidence.

Before you submit it, do you think one selected source can satisfy every part of the request?

Submit the request by pressing Enter.
Review the completed response.
Look for an inline citation attached to a UNESCO claim.
Look for an inline citation attached to a U.S. Department of Education claim.

The exact wording may differ. The response can support NIST points from the selected page.

The response cannot supply cited UNESCO or U.S. Department of Education evidence. This missing coverage is the intended shortfall.

Does the Response Appear Complete?

A fluent answer can still reach beyond its evidence. Check where each citation leads before treating the comparison as grounded.

Return to the Sources panel.
Confirm that only AI Risk Management Framework is selected.
Select any citation attached to UNESCO or U.S. Department of Education wording.

A citation that opens a NIST passage does not provide organization-specific evidence from UNESCO or the U.S. Department of Education. Help me audit whether this comparison is grounded in the selected NIST source.

Save the source-gap evidence

A failed evidence test becomes useful when you preserve it. The saved response records exactly where the current research base falls short.

Select Save to note beneath the comparison response.

Before you open the saved note, which organizations do you expect to have no matching cited evidence?

Select the note you just saved under Notes in the Studio panel.
Confirm that the note preserves the three-organization comparison request.
Confirm that the note records the missing UNESCO evidence.
Confirm that the note records the missing U.S. Department of Education evidence.
Confirm that only AI Risk Management Framework remains selected in the Sources panel.

You should see a saved comparison note that lacks cited evidence from two of the three named organizations. Only the NIST source should remain active.

You have turned an incomplete comparison into evidence about the research design. The source gap is now visible and recorded.

You have proved that answer quality depends on source coverage. Next, you will expand the evidence base with sources that directly address international ethics and education.

Add Evidence and Repair the Answer

Your source-gap note proves that the first comparison lacks UNESCO evidence plus education-specific evidence. A trustworthy policy brief needs those perspectives before it can support a three-way comparison.

Source coverage determines which claims Gemini Notebook can support with citations. This step repairs the gap by adding the two missing institutional sources.

In this step, get ready to:

Import the two missing institutional sources.
Activate the complete three-source evidence set.
Save a citation-checked policy synthesis.

Add the missing sources

The UNESCO source contributes international guidance on AI ethics. The U.S. Department of Education report contributes education-specific recommendations.

Select Add sources in the Sources panel.
Choose the source option that accepts web URLs.
Add both missing sources by pasting these URLs on separate lines:

https://www.unesco.org/en/artificial-intelligence/recommendation-ethics
https://www.ed.gov/sites/ed/files/documents/ai-report/ai-report.pdf

What Do These Sources Add?

The UNESCO source adds evidence about privacy, fairness, accountability, plus human oversight.
The U.S. Department of Education PDF adds guidance about humans in the loop, equity, safety, plus education guardrails.

Use the dialog's confirmation control to start the import.
Wait for both new sources to finish processing.

Your evidence base is broader now. The Sources panel lists the NIST AI Risk Management Framework, UNESCO Recommendation on the Ethics of Artificial Intelligence, plus Artificial Intelligence and the Future of Teaching and Learning: Insights and Recommendations.

Sources Not Processing?

Confirm that each URL begins with the complete web address shown above.
Try the import again with the URL that did not process.

Ask for help diagnosing the failed source import.

Select all three sources

A source checkbox controls whether Gemini Notebook can use that source in its next answer. All three organizations must be active for the repaired comparison.

Leave the checkbox beside AI Risk Management Framework selected.
Select the checkbox beside the UNESCO source.
Select the checkbox beside the U.S. Department of Education source.
Confirm that all three source checkboxes show selected indicators.

Your active evidence set now covers general risk management, international AI ethics, plus education-specific guidance.

Re-run and verify the comparison

The repaired answer must prove its coverage through inline citations. You will inspect one citation from each organization before preserving the synthesis.

Before you send the new question, consider whether all three organizations will appear with citations.

Ask Chat using this prompt:

Using all three selected sources, compare their recommendations for privacy, human oversight, and equity. Include at least one cited point from NIST, UNESCO, and the U.S. Department of Education, then propose five guardrails for responsible AI use in education.

What Does This Prompt Test?

The comparison tests whether the three active sources support the same policy themes.
The guardrails turn the cited synthesis into concrete proposals for responsible AI use in education.

That closes the gap you exposed: the response now draws cited points from all three organizations. You should also see five proposed guardrails for responsible AI use in education.

Hover over one NIST citation to reveal its quoted evidence.
Select the NIST citation to open its supporting passage in context.
Return to the comparison response.
Hover over one UNESCO citation to reveal its quoted evidence.
Select the UNESCO citation to open its supporting passage in context.
Return to the comparison response.
Hover over one U.S. Department of Education citation to reveal its quoted evidence.
Select the U.S. Department of Education citation to open its supporting passage in context.

Each selected citation should open a passage from the matching organization. This confirms that the comparison is grounded across the complete evidence set.

Return to the comparison response.
Select Save to note.

You should see the synthesis preserved as a note with clickable inline citations. Your original NIST research note plus the source-gap note remain available.

Missing One Organization's Evidence?

Confirm that the missing organization's source checkbox is selected.
Submit the same comparison prompt again.
Inspect the replacement citation before saving the new response.

Help me find why one source is missing from the comparison.

Your three-source synthesis is now grounded in the complete evidence set. Next, you will turn that evidence into a briefing pack for different review styles.

Generate Your Briefing Pack

The source gap is repaired. Your notebook now holds a citation-checked comparison across NIST, UNESCO, and the U.S. Department of Education.

Decision-makers need evidence in formats that fit different moments. Gemini Notebook's Studio can package the same sources for scanning, exploring relationships, or listening.

In this step, get ready to:

Review an AI-generated report built from your three selected sources.
Explore policy relationships in a mind map.
Listen to a short Audio Overview focused on shared guardrails.

Generate the briefing document

A briefing document gives your audience a written synthesis they can scan quickly. Studio generates it from the sources already selected in your notebook.

Select Reports in the Studio panel.
Choose briefing document as the report type.
Start the report generation.
Open the completed briefing document.

You'll see the completed briefing document open in Studio. Its content draws from the three selected sources.

Your audience now has a written version of the evidence that is ready for a quick policy review.

Report not finishing?

Confirm that all three source checkboxes remain selected.
Start the report generation again after every source shows as processed.

Help me diagnose the report generation problem.

Map the policy relationships

The written report presents a sequence of ideas. A mind map reveals how the policy themes connect across the evidence base.

Return to the Studio panel.
Select Mind map.
Start the mind map generation.
Open the completed mind map.

You should see branches connecting privacy to human oversight. You should also see equity linked to risk management.

That gives you a visual route through the policy themes before a discussion or review.

Mind map missing a policy theme?

Confirm that the NIST source remains selected.
Confirm that the UNESCO source remains selected.
Confirm that the U.S. Department of Education source remains selected.

Help me troubleshoot the missing mind map theme.

Create the audio briefing

The final format turns the evidence into a short briefing for someone who prefers listening. The Brief uses one speaker to deliver the key takeaways.

Return to the Studio panel.
Select Audio Overview.

You'll see the Audio Overview setup with controls for the format and focus.

Choose The Brief as the format.
Set the audio focus by entering this prompt:

Summarize the strongest shared guardrails, identify one important difference between the sources, and remind the listener to verify recommendations against the cited material.

How Does This Prompt Shape the Briefing?

The shared-guardrails instruction keeps the audio focused on common recommendations.
The difference instruction preserves an important point of disagreement or emphasis.
The verification reminder keeps the generated briefing connected to its cited evidence.

Audio generation can take a couple of minutes. A short wait here is expected.

Start the Audio Overview generation.

When generation finishes, you'll see The Brief ready to play.

Audio Overview not finishing?

Confirm that all three source checkboxes remain selected.
Start generation again after every source shows as processed.

Help me diagnose the Audio Overview problem.

Before you press play, which shared guardrail do you expect the briefing to introduce first?

Play the generated Audio Overview.

You'll hear the responsible-AI policy topic introduced in a short single-speaker briefing. The audio should direct the listener back to the cited material for verification.

Your briefing pack is ready for its audience across three review formats.

Secret mission

Debate and Fact-Check the Policy

A polished audio debate can still overstate the evidence behind a policy position. Generate a formal debate from your three sources. Then audit one claim and preserve the verdict as a citation-linked fact-check.

Clean Up Your Resources

Choose whether to keep your briefing pack in Gemini Notebook, pause your browser session, or remove the notebook under your organization's retention policy. This project creates no new paid subscription or metered cloud resource beyond your existing work or school access.

Resources you used:

One responsible-AI policy briefing notebook containing three selected public sources.
Four saved research notes covering the NIST guidance, the source gap, the three-source synthesis, and the audio claim fact-check.
One generated briefing document.
One generated responsible-AI mind map.
One Audio Overview using The Brief format.
One Audio Overview using The Debate format.

Keep everything running

No action is needed. Choose this option if you want the responsible-AI evidence pack available for future decisions.

Keep the responsible-AI policy briefing notebook in your work or school account.
Return to notebook.google.com whenever you want to review the briefing pack.
Reuse the citation-linked notes as a checked evidence base for future policy discussions.
Use the briefing document, mind map, and Audio Overviews when presenting the evidence in different formats.

Pause - I'll come back to this later

Pausing keeps every artifact in the notebook while ending the current Safari session. Your saved work remains available through the same organizational account.

Close Safari.
Return to notebook.google.com in Safari when you are ready to continue.
Use the same work or school Google account if Gemini Notebook asks you to sign in.
Select the responsible-AI policy briefing notebook to resume with its sources, notes, and Studio outputs intact.

Delete - I don't want to use this again

This option removes the notebook when your organizational retention policy does not permit keeping project material. The notebook holds every source, note, Studio output, and Audio Overview listed above.

Check Your Retention Policy

Deleting the notebook removes the complete briefing pack from your account. Your organization's data-retention policy determines whether the notebook should be kept.

Return to notebook.google.com.
Locate the responsible-AI policy briefing notebook in your notebook list.
Open the notebook's options menu.
Choose the option that deletes the notebook.
Confirm the deletion when Gemini Notebook asks you to approve it.

The responsible-AI policy briefing notebook should no longer appear in your notebook list. Cleanup is complete for every artifact stored inside it.

Nice Work!

You did it! Your Gemini Notebook now turns three public sources into a citation-checked responsible-AI briefing pack for a school or workplace learning team.

You've learned how to:

Test source grounding by exposing the limits of a one-source comparison. Repair the evidence gap with UNESCO guidance. Add the U.S. Department of Education report.
Trace inline citations to their supporting passages. Save a three-source policy synthesis with five responsible-AI guardrails.
Turn selected evidence into a briefing document for scanning. Use a mind map to explore policy relationships. Play The Brief Audio Overview to hear the policy topic introduced.
Secret Mission: Generate The Debate Audio Overview. Classify one debate claim against the evidence. Save your verdict in a citation-linked fact-check note.

Ready to quiz yourself?