How to Upload a PDF and Ask an AI Assistant a Question With NVDA and Firefox

The access method and the points where it can break

The access method in this guide is NVDA with Firefox, using a Windows keyboard and no mouse or touch screen. The task is to attach a PDF to an AI chat, submit a question about its contents and retrieve the response. The most important potential failure points are the attachment control, the file-selection dialog and the transition between uploading a document and making it available to the conversation.

AI tools for blind users need to expose more than readable answer text. A usable workflow also depends on meaningful control names, keyboard focus, status announcements and a logical reading order. A chat window may contain perfectly readable messages while still making its file attachment feature difficult to discover. The browser, application interface and operating system each contribute to the experience.

• NVDA: screen-reader output and keyboard navigation.
• Firefox: browser navigation and application interaction.
• Windows: file selection and document location.
• An AI chat service that accepts PDF attachments.

Prepare the PDF and the browser session

Start with a PDF whose location and filename are known. A short, descriptive filename such as project-guidelines.pdf is easier to recognize in a file list than a string of unexplained numbers. Keep the document in a folder that is straightforward to reach, such as Documents. If the PDF is scanned imagery, its contents may require optical character recognition before an AI service can extract meaningful text.

Open Firefox and navigate to the AI service through its usual sign-in process. Authentication may involve additional security controls, so complete any required verification before beginning the document task. Once the chat interface appears, use NVDA's reading commands to identify the main landmarks, headings, buttons and editable fields. The objective is to understand the page before interacting with individual controls.

Locate the attachment control with NVDA

Use Tab and Shift+Tab to move through interactive controls. Listen for a button whose name indicates file attachment, upload or adding a document. Depending on the application, the control may appear beside the message field, inside a toolbar or within a menu. NVDA's Elements List can also help locate buttons when the page contains many controls.

If the control is announced only as button, its name does not provide enough information to identify its purpose. Try examining the surrounding content with reading commands, including the previous and next lines. Some interfaces display a paperclip icon without a meaningful accessible name. If keyboard navigation reaches the icon but NVDA gives no useful announcement, the problem is likely in the application's control labeling or its relationship to surrounding text.

• Use Tab to move forward through controls.
• Use Shift+Tab to move backward.
• Use NVDA's Elements List to inspect available buttons.
• Listen for a meaningful upload or attachment announcement.

Select the document in the Windows file dialog

Activate the attachment control and wait for the file-selection dialog. This dialog belongs to Windows rather than the web page, so the navigation behavior may differ from the chat interface. Use Tab and Shift+Tab to inspect the filename field, folder navigation controls and file list. Ctrl+L can focus the location field in many standard Windows file dialogs, allowing a known folder path to be entered directly.

For example, if the PDF is in Documents, navigate to that folder, locate project-guidelines.pdf and activate it. In a standard file dialog, the filename can also be entered directly when the filename field is available. Confirm the selection with Enter or the dialog's Open button. Return attention to Firefox and listen for a document name or upload announcement.

Confirm the upload before asking a question

A selected filename and a completed upload are two different states. The application may show an attachment chip, a progress indicator or a status message. Navigate through the area around the message composer and determine whether the document is represented in the conversation. Wait for any upload activity to finish before submitting a question.

If NVDA does not announce the change, inspect the page again with reading commands or move away from the attachment control and return to the message area. Some applications update their visual interface without placing a corresponding announcement in the screen reader's speech output. A missing announcement should be treated as uncertainty about the interface state, not proof that the file was successfully attached.

• Identify the displayed document filename.
• Wait for any progress indication to finish.
• Look for an attachment confirmation or status message.
• Resolve upload errors before composing the question.

Ask a focused question and retrieve the answer

Move to the message field and enter a question that identifies the information needed. For example: What are the three eligibility conditions listed in section 2? This is more useful than asking the assistant to summarize everything without specifying a purpose. A focused question also makes it easier to identify whether the response contains the requested information.

Submit the message using the interface's Send control or its documented keyboard shortcut. After submission, navigate to the conversation area and locate the new response. Depending on the application, NVDA may announce new content automatically, or the user may need to move to the message history. Read the answer in manageable sections. If the response contains a long list, headings and paragraph boundaries can help maintain context.

For a document containing 20 pages, asking about one section at a time can reduce the amount of information that must be navigated. If an answer includes a page reference, return to the PDF and examine that passage. AI-generated summaries are not substitutes for checking the source text, especially when a precise number, condition or exception matters.

Handle common barriers and protect document information

When the attachment control cannot be identified, check whether the interface provides another keyboard-accessible route to its upload function. If the file dialog opens but focus is unclear, inspect its controls systematically instead of entering text into an unidentified field. If an upload remains pending, check the service's accepted file types and size limits. These requirements vary by provider and can change over time.

Document privacy is another part of the workflow. Before uploading, determine whether the service retains attachments, uses submitted material for product improvement, or offers organizational privacy controls. Sensitive work documents, personal records and confidential correspondence may be subject to additional restrictions. The relevant organization's data policy and the service's current terms should determine whether uploading is appropriate.

The task is complete only when the PDF is attached, the question is submitted and the answer can be read through the screen reader. If any stage fails, record the exact control name, announcement and navigation point. Next, open a nonconfidential sample PDF in Firefox and practice locating its attachment control, confirming the filename and retrieving one answer.

Questions readers ask