AI-Assisted Access Manual - Accessible Learning Center

First-time keyboard guide

Use this short routine before following the NotebookLM, Be My Eyes, and AI-Assisted Access Manual procedures. It explains exactly how the keyboard directions in this manual work.

Press a single key

Find the named key without pressing it if possible.
Press and release the key one time. Do not hold it unless the step specifically says to hold it.
Stop and listen to the full screen-reader announcement before pressing another key.

Press a keyboard combination

When a command contains a plus sign, such as Ctrl+S, hold down the first key.
While continuing to hold the first key, press and release the final key once.
Release the first key. The plus sign is not a key and should not be pressed.
Listen for the expected result described in the procedure. If nothing changes, verify that the correct application or webpage has focus and try the command once more.

Press a sequence of keys

When keys are separated by commas or the word “then,” press and release them in order instead of holding them together.
After each key, listen for a menu, tab, control, or new screen before continuing.
If the announcement does not match the step, press Escape once to close the current menu or dialog and restart that procedure.

Know which navigation key to try

Tab and Shift+Tab: move forward and backward among buttons, links, fields, and other controls.
Arrow keys: move inside menus, lists, groups of choices, text, and many screen-reader reading views.
Enter: opens the selected item or activates the focused command.
Space: activates many buttons and checks or clears checkboxes. In a text field, Space types a space.
Escape: cancels or closes the current menu, list, or dialog without intentionally accepting a change.

Verify before moving on

Listen for the window, page, or application name so you know where focus is.
Perform only one numbered step.
Compare what the screen reader announces with the “What should happen” information in the procedure.
If the result is wrong, stop. Use the procedure’s recovery directions, Escape to cancel, or Ctrl+Z to undo when that command is supported.
Continue only after you can identify the result. Save completed work with Ctrl+S whenever the application supports saving.

First complete independent workflow

This rehearsal joins existing procedures into one uninterrupted session without repeating their directions. Follow the linked procedure when a step names it, then return here for the next verification.

Safety check: Confirm the active account, workspace, audience, and privacy level before entering, uploading, sharing, or sending information.

Open NotebookLM, Be My Eyes, and AI-Assisted Access Manual and wait until its name, starting screen, or connected-device status is announced.
Choose a fictional, public, disposable, or instructor-provided practice item. Do not use private client, student, medical, financial, or account-recovery information.
Open 3.1 Create a Notebook. Read its “What you will learn,” “Before you begin,” keyboard-command, and expected-result information before pressing anything.
Complete that procedure one numbered action at a time. After each key, stop and identify what changed before continuing.
State the result aloud or record it in a practice note. If it differs from “What should happen,” use that procedure’s recovery directions before advancing.
Continue directly to 3.2 Add Several Related Sources without closing the application or device session. First confirm that focus, account, file, book, conversation, or project context is still correct.
Complete the second procedure and verify the resulting name, state, location, permission, value, or content. Do not rely only on a success sound.
Save or synchronize when the product supports it. Wait for the completed state before disconnecting a device, closing a browser, or switching accounts.
Continue to 5.1 Ask a Grounded Question. Use the exact command listed there; if the command varies by model or version, use Keyboard Help, the product’s shortcut list, or the linked official reference instead of guessing.
Introduce one harmless recoverable error, such as opening the wrong menu or moving focus away. Press Escape once, identify the active window or device area, and return to the correct step.
Review the finished practice item from its beginning. Confirm its title or filename, structure, content, destination, sharing state, and saved or synchronized status as applicable.
Close and reopen the item or return to it from the product’s main screen. The workflow is complete only when you can locate it and verify the same result independently.

Success standard: Complete all three linked tasks, explain the commands used, recover from the practice error, and locate the finished result without another person directing the next key.

Source-Grounded Research, Visual Assistance, Screen Readers, Braille, Verification, Privacy, and Human Escalation

Current product name: Google renamed NotebookLM to Gemini Notebook on July 16, 2026. This manual keeps NotebookLM in the title for familiarity while teaching the current Gemini Notebook product.

Purpose: This manual teaches source-grounded research with Gemini Notebook, visual assistance with Be My Eyes and Be My AI, and a practical framework for safe AI-assisted access.

Core principle: AI can increase independence, but it is not an infallible authority. Verify important details and escalate to an authoritative source or trusted human when an error could cause significant harm.

How to Use This Manual

Practice first with public documents, ordinary household objects, and non-sensitive photos.
Use Gemini Notebook citations to return to the original source rather than trusting a summary alone.
Do not use Be My AI for sensitive personal or financial information.
Use a live volunteer, company agent, trusted person, or professional when AI uncertainty matters.
Protect privacy before sharing a screen, camera view, clipboard image, document, speech output, or braille display.
Ask AI what it directly observed versus what it inferred.
For high-stakes information, the authoritative source or qualified human is the final verifier.

Part 1 - AI-Assisted Access: The Big Picture

Tool	Best role
Gemini Notebook	Source-grounded research and learning from documents you select.
Be My AI	AI-generated descriptions of photos, screens, image files, and clipboard images.
Be My Eyes volunteer	Live human visual assistance through one-way video and two-way audio.
Screen reader / braille / OCR	Primary access tools that can work alongside AI.

Independence: Choosing the right tool is part of independence. AI is not automatically better than OCR, a screen reader, or a human.

Part 2 - NotebookLM Is Now Gemini Notebook

Google renamed NotebookLM to Gemini Notebook on July 16, 2026.
It remains a standalone research product.
Existing notebooks and shared links continue through the rebrand.
Gemini Notebook chat is grounded in notebook sources; Gemini app conversations using notebooks may also use web search and other tools.

Part 3 - Create a Notebook and Add Sources

3.1 Create a Notebook

What you will learn: Build a research workspace from selected sources.

Before you begin: Use a Google account and non-sensitive practice material.

Exactly how to do it

Open the current Gemini Notebook website.
Activate Create notebook or the new-notebook control.
Give the notebook a descriptive name.
Find Add source.
Choose a source type.
Upload or add the source.
Wait for processing to complete.
Review the Sources panel and source guide.

What should happen: The notebook contains a processed source that can be used in chat.

If that does not happen: If import fails, check file size/type, permissions, copy protection, or source availability.

Practice: Create a notebook with one public PDF and one public webpage.

What you will learn: Create a stronger research collection.

Before you begin: A notebook is open.

Exactly how to do it

Activate Add source.
Add another relevant source.
Repeat until you have several sources.
Rename unclear source titles when the interface allows.
Keep unrelated topics in separate notebooks.

What should happen: The notebook contains a focused source set.

If that does not happen: If answers become unfocused, deselect irrelevant sources or split the project into multiple notebooks.

Practice: Build a three-source practice notebook.

Part 4 - Source Types and Limits

Current supported sources include PDF, DOCX, TXT, Markdown, CSV, PPTX, Google Docs, Google Slides, Google Sheets, images, audio, web URLs, public captioned YouTube videos, pasted text, ePub, and eligible Google Play Books ebooks.
Current published local-upload limit is up to 200 MB and 500,000 words per source.
Free accounts currently support up to 50 sources per notebook; paid plans may allow more.
Google Slides currently supports up to 100 slides as a source.
YouTube import uses the transcript/captions, not the visual video content.

Rights: Only upload or import content you are authorized to use.

Part 5 - Chat with Sources and Use Citations

5.1 Ask a Grounded Question

What you will learn: Get an answer tied to source material.

Before you begin: At least one source is processed.

Exactly how to do it

Move to the chat prompt.
Ask a specific question about the source material.
Send the question.
Read the response.
Move to an inline citation.
Activate the citation.
Read the cited passage and surrounding context.

What should happen: You can trace important claims back to the source.

If that does not happen: If the question is outside the selected sources, Gemini Notebook may say it cannot answer.

Practice: Verify three claims through citations.

Screen-reader citation strategy: Do not depend on visual hover previews. Activate the citation and read the source passage directly.

Part 6 - Source Selection, Notes, and Source Guides

6.1 Limit a Question to Selected Sources

What you will learn: Compare documents or reduce irrelevant context.

Before you begin: The notebook has multiple sources.

Exactly how to do it

Review source checkboxes in the Sources panel.
Select only the sources relevant to the question.
Return to chat.
Ask the question.
Repeat with another source subset and compare the answer.

What should happen: The response reflects the selected source set.

If that does not happen: If unwanted sources remain active, recheck selection state.

Practice: Compare two sources on the same issue.

Source Guides summarize individual sources.
Notes can contain your own text or saved chat responses.
Current Google Help says written notes are editable; saved-response notes are not editable once created.
Notes can be converted into sources when you intentionally want future chat grounded on them.

Part 7 - Audio and Video Overviews

7.1 Generate an Audio Overview

What you will learn: Create a podcast-style spoken summary.

Before you begin: Use a notebook with suitable sources.

Exactly how to do it

Open Studio.
Find Audio Overview.
Select sources/customization options if offered.
Enter a focus prompt if desired.
Generate the overview.
Play it.
Use play/pause, skip, and speed controls.
Return to source citations to verify surprising claims.

What should happen: A spoken overview is generated from your sources.

If that does not happen: If an important claim seems wrong, verify it in the source rather than trusting the audio.

Practice: Generate one overview and verify three claims.

Video Overviews: Video Overviews are visually rich. For nonvisual access, use source-grounded chat, available text, or request a separate description of the visuals.

Part 8 - Reports, Study Aids, Data Tables, Slide Decks, and Infographics

Output	Use
Reports	Briefing documents, FAQs, study guides, or custom reports.
Flashcards / Quizzes	Practice and recall from notebook material.
Data Tables	Structured rows and columns that can be exported to Sheets.
Slide Decks	Presentation-style source summaries.
Infographics	Visual summaries that may need a textual alternative.

Verification: Generated study aids and reports can still contain errors. Return to the original sources when correctness matters.

Part 9 - Mind Maps and Accessible Alternatives

9.1 Turn a Mind Map into Structured Text

What you will learn: Use the conceptual relationships without depending on a visual diagram.

Before you begin: A Mind Map exists.

Exactly how to do it

Ask chat for the main branches as headings.
Ask for subtopics as nested lists.
Ask it to state cross-branch relationships in sentences.
Verify the concepts against the selected sources.

What should happen: You receive a linear, screen-reader-friendly representation.

If that does not happen: If a relationship is missing, ask for an explicit list of connections.

Practice: Convert one mind map into a nested outline.

Part 10 - Exporting and Mobile Use

Reports can be exported to Google Docs.
Data Tables can be exported to Google Sheets.
Edits in exported Docs/Sheets do not synchronize back to the notebook.
Sharing permissions do not automatically carry over to exported files.
The mobile app supports a subset of desktop features. Google recommends desktop for the full experience.
Mobile supports key source types such as PDF, websites, YouTube, audio, and copied text, plus Audio/Video Overviews; some Studio features remain desktop-only.

Part 11 - Gemini Notebook with Screen Readers and Braille

Use headings and landmarks to move among Sources, Chat, and Studio when exposed by the web app.
Use button/form-control navigation for Add source, checkboxes, prompt fields, generation controls, and menus.
JAWS/NVDA users may need to switch between browse and focus/application modes for interactive controls.
VoiceOver users may need to interact with grouped web controls.
Braille works through the active screen reader; test output, input, and routing.
Prefer textual reports and citation-linked source passages when a visual artifact is inefficient to access.

Recovery: Press Escape to close a popup, return to a top-level heading or landmark, then re-enter the panel you need.

Google states that files added to notebooks and chats in Gemini Notebook are not used directly to train foundational generative AI models unless you provide feedback.
Data handling can differ when a notebook is used inside the Gemini app.
Shared notebooks expose permitted notebook content to collaborators.
Exported Docs/Sheets have independent sharing settings.
Deleted notes currently cannot be recovered.
Do not upload confidential or regulated information unless your organization authorizes it.

Part 13 - What Be My Eyes Is

Be My Eyes connects blind and low-vision users with live sighted volunteers, participating company support, and Be My AI.
Volunteer calls use one-way video from the user and two-way audio.
Be My AI accepts photos and returns detailed AI-generated descriptions.
Be My AI is available on iOS, Android, and Windows desktop.
The Service Directory can connect users with participating companies for product-specific help.

Part 14 - Call a Sighted Volunteer

14.1 Get Live Visual Help

What you will learn: Use a human when real-time camera guidance is useful.

Before you begin: Choose a non-sensitive task and stable internet.

Exactly how to do it

Open Be My Eyes.
Open Get Support.
Choose Call a Volunteer.
Wait for the volunteer to answer.
State the task clearly.
Tell the volunteer how you are holding the phone.
Move the camera slowly in response to directions.
Ask for confirmation of important details.
End the call when finished.

What should happen: A volunteer provides live verbal visual assistance.

If that does not happen: If the volunteer cannot help, end politely and call again or use a specialized company service.

Practice: Practice centering a household label.

Part 15 - Use Be My AI on Mobile

15.1 Take a Picture for Description

What you will learn: Get an AI description of an ordinary object or scene.

Before you begin: Use a non-sensitive object.

Exactly how to do it

Open Be My Eyes.
Open the Be My AI tab.
Activate Take picture.
Point the rear camera at the subject.
Hold still and take the picture.
Wait for the description.
Read or listen to the result.
Use Ask more for a specific follow-up.

What should happen: Be My AI returns a description within the conversation.

If that does not happen: If it is vague, retake with better light, distance, framing, or angle.

Practice: Describe a package and ask where a control or label is located.

Internet: Be My AI requires an internet connection.

Part 16 - Follow-Up Questions and Multiple Pictures

16.1 Ask More

What you will learn: Turn one description into an interactive visual task.

Before you begin: A mobile Be My AI description is open.

Exactly how to do it

Activate Ask more.
Ask one specific question.
Send it.
Read the answer.
Add another picture if the mobile interface offers it.
Compare what is directly visible across the images.

What should happen: The conversation stays focused on the visual subject.

If that does not happen: If the model starts guessing, ask: What can you see directly and what are you inferring?

Practice: Use two angles of the same harmless object.

Desktop difference: Current Windows desktop guidance uses one image per chat; start a new desktop chat for a new image.

Part 17 - Be My AI on Windows Desktop

Task	How / What to know
Describe your screen	Captures the current screen after hiding the Be My Eyes app.
Describe an image file	Select a supported image from the computer.
Describe clipboard	Describe an image copied to the clipboard.
Take a picture	Use webcam or linked phone camera when available.

17.1 Describe a Harmless Screen

What you will learn: Use AI as a supplement to screen-reader navigation.

Before you begin: Close unrelated private windows first.

Exactly how to do it

Open Be My Eyes desktop.
Choose Describe your screen.
Allow the capture.
Read the description.
Ask a focused question such as: Which dialog is open and what buttons are visible?
Return to the original app.
Use your screen reader to locate the named control.

What should happen: The AI description gives orientation while the screen reader remains the control/navigation tool.

If that does not happen: If the screen contains sensitive information, cancel and use a safer method.

Practice: Describe a Windows Settings screen and locate one named button with your screen reader.

Part 18 - Service Directory and Be My Eyes Workplace

Service Directory connects users with participating organizations for specialized product/service assistance.
Availability depends on region and business hours.
Be My Eyes Workplace is enabled by an employer or organization.
Current Workplace features include Workplace AI, Workplace Reader, and Workplace Connect with trusted company assistance/screen sharing where configured.
Follow employer confidentiality and data-handling rules.

Part 19 - Be My Eyes Privacy and Sensitive Information

Do not show passports, driver licenses, or other government ID.
Do not share bank, credit-card, insurance, or other financial information.
Do not expose Social Security or government identification numbers.
Do not share usernames, passwords, access keys, or security-question answers.
Avoid sensitive medical information in ordinary Be My AI use.
Remember that live volunteers can see the camera view and hear the conversation.

Current data statement: Be My Eyes states that it does not train AI models on Be My AI submissions, questions, or responses and does not allow its AI provider to train on that data. Current help says ordinary AI descriptions are retained for 30 days unless the user chooses to save specific images/conversations.

Part 20 - Choosing AI, OCR, or a Human

Task	Preferred method
Printed menu	OCR or AI; verify allergens/prices when important.
Clothing matching	AI or volunteer; human help for subtle shade matching.
Appliance panel	AI for layout; official manual/human for safety-critical operation.
Medicine label	Accessible pharmacy/professional or trusted human; not AI alone.
Bank statement	Accessible bank channel; avoid general visual-assistance upload.
Government ID	Do not share through general AI/volunteer workflows.
Photo composition	Be My AI or volunteer.
Complex chart	AI description plus original data/source verification.

Part 21 - Better Image and Screen Descriptions

21.1 Ask Better Visual Questions

What you will learn: Get actionable spatial information instead of a generic description.

Before you begin: Use a harmless image.

Exactly how to do it

Ask for an overall description.
Ask for spatial layout: top/bottom, left/right, foreground/background.
Ask for exact visible text.
Ask which details are unreadable or uncertain.
Ask where the object/control is located relative to a tactile landmark.
Retake the picture if glare, angle, distance, or cropping limits the answer.

What should happen: The answer becomes more useful for independent action.

If that does not happen: If the AI changes its answer, treat the disputed detail as uncertain and escalate.

Practice: Describe one control panel using orientation language.

Part 22 - Documents, Charts, Forms, and Tables

Use Gemini Notebook when a document can be uploaded and cited.
Use OCR/document readers for straightforward text recognition.
Use AI to explain chart/layout structure, but verify exact values against the data.
For forms, prefer accessible electronic forms and official assistance; do not let AI infer legal or financial answers.
Ask AI to convert visual tables into row-by-row structured text.
Verify signatures, dates, totals, account numbers, dosage, and legal terms directly.

Part 23 - Everyday Visual Tasks

23.1 Labels and Packaging

Take a straight, close photo with good light.
Ask what product is shown.
Ask for exact visible text.
Ask where information appears on the package.
Photograph another side instead of expecting one image to show everything.

23.2 Clothing

Ask for garment type, color, pattern, orientation, and visible damage.
Use a second photo when matching items.
Exact color is lighting-dependent; use a human when shade matching matters.

23.3 Appliances

Ask for the arrangement of buttons/dials.
Identify tactile landmarks before pressing anything.
Use the official manual for safety-related controls.
Do not let AI guess burner, electrical, gas, medical-device, or other hazardous settings.

Part 24 - High-Stakes Limits and Human Escalation

Medication, dosage, pills, medical devices, or medical records.
Financial accounts, payment instructions, cards, checks, or tax documents.
Legal forms, contracts, court documents, or government identification.
Electrical panels, gas controls, fire/smoke hazards, traffic, machinery, or dangerous tools.
Identity verification or accusations about a person.
Anything where an error could be difficult to reverse.

Escalation order: Use an authoritative accessible source first, then a trained professional/company representative or trusted human. AI may help formulate questions but should not be the final verifier.

Part 25 - Deepfakes, Misidentification, and False Confidence

AI can misidentify people, objects, text, colors, locations, and relationships.
Generated or edited images can look authentic.
Voice cloning can imitate relatives, executives, or public figures.
Do not use AI image description as proof of identity.
Ask what is directly observed versus inferred.
Verify important media through original sources, direct contact, and independent evidence.

Speech can be overheard; use headphones for sensitive work.
Braille cells can expose information to nearby braille readers.
Clipboard history may retain copied descriptions or document text.
Screen-description tools capture what is visible; close unrelated private windows first.
Camera assistance can reveal mail, addresses, people, screens, and surroundings unintentionally.
Lock devices when finished.

Pre-capture question: What else is visible that I did not intend to share?

Part 27 - Troubleshooting

27.1 Gemini Notebook Answer Seems Wrong

What you will learn: Return to the source rather than arguing with the summary.

Before you begin: A claim conflicts with your understanding.

Exactly how to do it

Open the citation.
Read the cited passage and surrounding context.
Check which sources are selected.
Repeat the question with only the most relevant source.
Ask the model to separate direct statements from inference.

What should happen: The source or interpretation problem becomes clear.

If that does not happen: If the source itself is unreliable, add a stronger source.

Practice: Correct one ambiguous answer.

27.2 Gemini Notebook Cannot Answer

What you will learn: Check whether the information exists in the selected sources.

Before you begin: Chat says the answer is unavailable.

Exactly how to do it

Review selected sources.
Search for the key term in the source.
Rephrase using the source terminology.
Add a relevant source if the information is genuinely missing.

What should happen: Chat answers when appropriate source material exists.

If that does not happen: Do not ask source-grounded chat to invent missing information.

Practice: Ask one answerable and one unanswerable question.

27.3 Be My AI Is Vague or Inconsistent

What you will learn: Improve the image and narrow the question.

Before you begin: The visual answer is not trustworthy enough.

Exactly how to do it

Ask which detail is uncertain.
Retake with better light.
Change distance.
Change angle.
Ask one question at a time.
If inconsistency remains, call a volunteer or trusted person.

What should happen: The task becomes clearer or is escalated safely.

If that does not happen: Repeated confident answers are not proof.

Practice: Retake one image three ways.

27.4 Volunteer Cannot See the Detail

What you will learn: Use deliberate camera orientation.

Before you begin: A live volunteer says the view is unclear.

Exactly how to do it

State how you are holding the phone.
Ask for one direction at a time: left/right/up/down/closer/farther.
Move slowly.
Pause after each movement.
Ask when the target is centered.

What should happen: The volunteer can guide the camera more effectively.

If that does not happen: Stop if the task requires unsafe one-handed operation.

Practice: Practice centering a printed label.

What you will learn: Return to structural navigation.

Before you begin: An AI web panel or popup changed focus.

Exactly how to do it

Press Escape to close menus/dialogs.
Use heading navigation.
Use landmarks/regions if supported.
Use edit-field navigation to find chat.
Switch browse/focus/application mode if direct interaction is required.

What should happen: You return to a known part of the interface.

If that does not happen: After major product updates, explore labels instead of relying on old visual position.

Practice: Practice opening and closing a Studio panel.

Part 28 - Independent Practice Course

Level 1 - Tool Choice

☐ Explain when to use Gemini Notebook.

☐ Explain when to use Be My AI.

☐ Explain when to call a volunteer.

☐ Explain when to use an official/professional source.