Week 2: Claude

Learning objectives

By the end of this week, you will be able to:

1. Explain what Claude is and how it relates to the large language model concepts you have already learned.
2. Use reliable, general screen reader exploration techniques, such as your screen reader's list of page elements and its heading navigation command, to find your way around Claude.ai's web app.
3. Start a new conversation with Claude, send a message, and read Claude's reply using Say All, or your screen reader's continuous reading command.
4. Describe what a Project is in Claude.ai and explain why grouping related conversations together can be useful.
5. Describe what an artifact is and explain how a screen reader user can notice and open one.
6. Upload a file to a conversation with Claude and ask a question about its contents.
7. Describe what memory features are, why a blind user might choose to use them, and what privacy considerations apply.
8. Compare the general navigation experience of the Claude.ai web app with a desktop screen reader to the Claude mobile app with a mobile screen reader such as VoiceOver or TalkBack.

Lesson

What Claude is, and how it fits what you already know

Claude is a large language model, or LLM, built by a company called Anthropic. Everything you already know about how large language models work applies directly to Claude. Claude works by next-token prediction, generating its response one small piece of text, or token, at a time, based on patterns learned during training. Claude has a context window, a limited amount of text it can consider at once, including your prompt, its own earlier replies in the conversation, and anything you have uploaded. Like every large language model, Claude can hallucinate, confidently stating something false, so the verification habits you have been practicing apply just as much to Claude as to any other AI tool.

Claude is a chatbot, meaning you interact with it mainly by typing messages, called prompts, and reading its replies. What makes this week different is depth. Earlier in this course, you used AI chatbots briefly, as examples among many. This week, you will spend real time inside one specific tool, learning its layout, its features, and the keyboard patterns that let you use it efficiently. That depth is exactly what builds prompt fluency: the more comfortable you are moving around inside a real AI tool, the less mental effort you spend on the mechanics of using it, and the more effort you can spend on writing a good prompt and evaluating the reply.

Claude is available in more than one form. There is a web app, which you reach through your web browser like any other website, and there are mobile apps for iPhone and Android. This lesson covers both, but it spends more time on the web app, since that is where you will do most of your practice this week with your screen reader and keyboard. Later in this lesson, you will read about how the mobile experience differs.

Signing in and finding your way around Claude.ai's web app

Before this section goes any further, it is worth being honest about something important. AI chatbot websites, including Claude.ai, are updated often. Button names, menu labels, and even the overall layout can change between the time this lesson is written and the time you read it. This course cannot promise you an exact, permanently accurate map of every control on the page. What it can promise, and what it will teach you instead, is a small set of reliable exploration skills that work no matter how the page changes, because they rely on how the page is built, not on memorizing a specific button's exact wording.

You already have these skills. Most screen readers offer a list of page elements, sometimes called an elements list, that lets you view every heading, link, button, form field, or landmark on a page in one organized list, and jump straight to any of them; check your own screen reader's shortcut list or documentation for the exact key combination that opens it. Heading navigation, available in your screen reader's browse mode, lets you skim a page's structure quickly, one heading at a time. Landmark or region navigation jumps between major regions of the page, such as the main content area, a navigation region, or a search region. Edit field or form field navigation jumps between text boxes, which is exactly what you need to find a message box or a search field. Button navigation jumps between clickable buttons. Together, these five tools let you build an accurate mental map of almost any website's current layout in under a minute, including Claude.ai, whatever its exact wording happens to be on the day you visit. Each screen reader assigns its own keys to these commands, so check your own screen reader's shortcut list or documentation for the exact keys it uses.

To begin, open your web browser and go to the Claude.ai website. If you do not already have an account, you will need to create one; if you do, you will sign in. A sign-in or account creation page typically includes a small number of edit fields, such as an email address field, and one or more buttons, such as a button to continue or sign in. Use your screen reader's edit field and button navigation commands to move between them if Tab alone feels slow, and use Say All, or your screen reader's continuous reading command, if you want to hear the whole page read from the top first, so you understand the general shape of the page before you start interacting with it.

Once you are signed in, you will typically land on a main conversation screen. Most AI chatbot websites, Claude included, organize this main screen into a small number of predictable regions: a place to start or continue a conversation, a place to type a new message, and often a side area listing your past conversations or your Projects, a feature covered later in this lesson. Use your screen reader's landmark or region navigation to move between these regions, and use heading navigation to explore each region's contents once you are inside it. If what you hear does not match this general description exactly, do not worry. Fall back on your screen reader's list of page elements, open it, switch to a headings or landmarks view if one is offered, and read down the list slowly until the layout makes sense. This exploration method is the durable skill. The exact words Claude's interface uses on any given day are not.

Starting a new conversation and sending a message

On most AI chatbot websites, including Claude, there is a way to begin a brand-new conversation, separate from any conversation you have already had. This is often reachable through a heading, a link, or a button described with words like "new chat" or "new conversation," findable through your screen reader's list of page elements or through heading and button navigation as described above. Once you are in a conversation, whether new or continuing, there is a text box for typing your message. On most AI chat websites, this text box sits near the end of the page's reading order, since new messages are added to the bottom of a growing conversation, and it accepts more than one line of text, so you can write a short question or a long, detailed prompt in the same box.

To find this box reliably, use your screen reader's edit field navigation to jump between edit fields until it announces something like a multi-line edit box, often labeled with a word like "message" or a phrase like "how can I help you today." If that exact label is not what you hear, trust the type of control you land on, a multi-line text entry field near the end of the page, rather than the exact wording. Once your focus is on that field, your screen reader will normally switch into its own typing or focus mode automatically, which lets you type freely without your keystrokes being interpreted as browse-mode commands. Type your message, then press Enter to send it. On some AI chat websites, plain Enter sends the message and Shift plus Enter adds a new line without sending; if pressing Enter alone does not send your message, try looking for a nearby "Send" button instead, using your screen reader's button navigation to jump to it.

Some screen readers also offer a way to move through a page item by item, or a rotor-style menu that lists just the form controls on a page, as an alternative way to reach a text field directly. Check your own screen reader's documentation to see what it offers and which keys it uses.

Reading Claude's response with your screen reader

After you send a message, Claude's reply usually appears a moment later, sometimes appearing gradually, word by word, as it is generated, rather than all at once. It is usually easiest to wait a few seconds until the reply has finished appearing before you start reading it, so your screen reader is not interrupted partway through by new text still arriving.

Once the reply has finished, use Say All, or your screen reader's continuous reading command, to hear the entire response read from your current position to the end. This is usually the fastest way to hear a full answer. If you want to move more slowly, use your screen reader's line-by-line reading commands, in browse mode, to move one line at a time, or use heading navigation to jump between headings if Claude's reply includes its own internal headings, which it sometimes does for a longer, more structured answer. If you lose track of where you are after Claude's reply finishes, opening your screen reader's list of page elements and checking a headings view is a reliable way to reorient yourself, since a new response is often marked with its own heading or region in the page's structure.

Most screen readers offer a similar continuous-reading command that starts reading from your current position, and you can usually pause it at any point, often with the Control key, if you want to stop and think about what you just heard.

What a Project is, and why it is useful

A Project, in Claude.ai, is a way to group related conversations together under one heading, instead of leaving every conversation loose and separate. Imagine you are using Claude to help with a job search: you might have several different conversations over several weeks about researching companies, drafting a resume, and preparing interview answers. Without a Project, those conversations could pile up alongside every unrelated conversation you have ever had, making the specific ones you need harder to find later. With a Project, you can keep all of them together in one place, making it far easier to return to related work later, whether that is tomorrow or next month.

Projects typically also let you add your own custom instructions or reference material that applies to every conversation inside that Project, rather than needing to repeat that same context every single time you start a new chat. For example, if you are using Claude for your job search, you might add background information about your own work history once, at the Project level, so every conversation inside that Project already has that context available, instead of you retyping your background in every new conversation. This directly supports prompt fluency: less time spent repeating yourself means more time spent on the actual task.

To explore whether a Projects area exists in the version of Claude.ai you are using, and how it is organized, use the same exploration skills from earlier in this lesson: check your screen reader's list of page elements for a heading or link mentioning "Projects," or use landmark and heading navigation to look for a distinct region of the page dedicated to it. Because this is an area of the interface most likely to change its exact wording or placement over time, this course intentionally does not describe one exact path to it. The exploration skill is what will still work months or years from now.

What an artifact is

An artifact is a separate, persistent piece of content that Claude sometimes creates alongside its normal conversation reply. Instead of only replying with plain conversational text, Claude may decide that a piece of writing, a piece of code, or a structured document is better presented as its own separate item, one that stays available and editable on its own, rather than being buried in the flow of the conversation. A written report, a piece of computer code, or a table of information are all common examples of things Claude might turn into an artifact rather than plain conversational text.

For a screen reader user, an artifact is worth understanding because it does not always sit in the same reading position as an ordinary reply. Depending on how the interface is currently built, an artifact may appear in a separate area of the page, sometimes described as its own region, panel, or heading, alongside or after Claude's regular conversational response. This is exactly the kind of situation where your exploration skills matter most. If Claude's reply mentions that it has created something, such as a document or a piece of code, but the conversational text alone seems shorter than you expected, check your screen reader's list of page elements for a new heading, region, or link that appeared after you sent your message. Landmark or region navigation is often especially useful here, since a distinct content area like an artifact is a strong candidate for its own landmark region.

Once you locate an artifact, you can generally read it the same way you read any other content, with Say All or line-by-line review commands. Some artifacts may also include their own controls, such as a way to copy the content or view different versions of it; use your screen reader's button navigation to check what controls are nearby if you want to do more than just read the content.

Uploading a file to a conversation

Claude, like many AI chatbots, lets you upload a file, such as a document, spreadsheet, or image, into a conversation, so you can ask questions about its contents. This can be genuinely useful for a blind user: instead of manually reading through a long or poorly organized document yourself, you can upload it and ask Claude to summarize it, pull out specific information, or describe its overall structure before you decide whether to read it in full yourself.

On most AI chat websites, an upload control sits near the message box, often as a button labeled with a word like "attach" or "upload," or represented by an icon with an accessible label to that effect. Use your screen reader's button navigation, or its list of page elements filtered to buttons or form controls, to locate it, since its exact wording and exact position can vary between versions of the interface. Activating that control typically opens your operating system's normal file selection dialog, the same kind of dialog you use to open a file in any other Windows or Mac program, which you can navigate with your screen reader exactly as you would in any other app.

Once a file has finished uploading, it is usually a good idea to ask Claude a direct question about it in your next message, such as asking it to summarize the file, describe its structure, or answer a specific question about its content, rather than uploading a file with no message at all. Before you upload anything, it is worth pausing to think about privacy: uploading a file sends its contents to a company's remote servers for processing, exactly like typing sensitive information directly into a chat. Avoid uploading a document containing highly sensitive personal information, someone else's private data, or confidential material from an employer, unless you have checked and are comfortable with that tool's specific privacy practices for uploaded files.

Memory features: what they are and why they matter

Some AI chatbots, including Claude, offer a memory feature: a way for the tool to retain certain information about you or your ongoing work across separate conversations, rather than starting from a completely blank slate every single time you open a new chat. In general terms, a memory feature might let Claude recall things like your general preferences, ongoing projects, or facts you have told it before, without you needing to repeat that information in every new conversation.

For a blind professional, memory features can offer a real efficiency benefit, similar to the benefit of a Project's custom instructions described earlier in this lesson. If you no longer need to retype the same background information at the start of every conversation, you save real time and real typing effort, which matters especially if you rely on a screen reader and keyboard for everything you do. That said, memory is a feature you should use deliberately, not automatically, and this is exactly the kind of decision that calls for thinking carefully about privacy.

Memory means information about you persists somewhere beyond a single conversation, which raises an important privacy consideration: information you share with an AI tool becomes data held by the company that runs that tool, subject to that company's own privacy practices, whether that information is used once in a single conversation or retained afterward through a memory feature. Before relying on a memory feature, it is worth exploring, using your screen reader's normal exploration skills, whether the tool offers a way to view what it currently remembers, and a way to edit or clear that memory if you change your mind. Look for a settings or preferences area, generally reachable through the same kind of heading, landmark, and button exploration described earlier in this lesson, and read through what you find there carefully before deciding whether and how much to rely on it.

Conversation strategies that make Claude more useful

Beyond the mechanics of typing and reading, a few simple strategies make a real difference in how useful a conversation with Claude turns out to be. The first is asking follow-up questions. You do not need to get a perfect answer on your first try. If Claude's reply is close but not quite right, or leaves something unclear, simply ask a follow-up question in the same conversation, such as asking it to explain a specific part further, or asking a related question that builds on what it just told you. Because Claude can see the earlier parts of the conversation, within its context window, a follow-up question is usually much faster than starting over from scratch.

The second strategy is asking Claude to redo something in a different format. If a response comes back as a long block of plain paragraphs and you would prefer a numbered list, or if it comes back as a list and you would prefer plain paragraphs, you can simply ask for that directly, for example: "Please rewrite that as a numbered list" or "Please turn that into plain paragraphs instead of a table." This kind of format request is a preview of a much deeper skill you will build later in this course, but you can start practicing the basic idea right now.

The third strategy connects directly back to the context window concept. Once a conversation has gone on for a very long time, covering many unrelated topics, or has drifted far off the original subject, it can be worth starting a brand-new conversation rather than continuing to pile more messages onto the old one. A very long conversation can eventually push earlier parts outside the context window, meaning Claude may no longer have those earlier details available when forming a new reply, which can lead to confusing or inconsistent answers. Starting fresh, especially when you are moving on to a genuinely new topic, keeps each conversation focused and keeps Claude's replies more reliable. If you find you are switching topics often within one ongoing area of work, that is exactly the situation a Project, described earlier in this lesson, is designed to help with.

Claude.ai web app versus the Claude mobile app

Claude is also available as a mobile app on iPhone and Android, and the experience of using it with a screen reader is genuinely different from the web app, not just visually smaller. This difference comes from how screen readers generally work on each platform, not from anything specific to Claude.

On a computer, a desktop screen reader reading a web page like Claude.ai typically uses a browse mode, which builds a structured, explorable model of the whole page: headings, landmarks, links, buttons, and edit fields you can jump between in almost any order, using the navigation commands covered earlier in this lesson. On a phone, VoiceOver on iOS and TalkBack on Android generally use a more linear model: you move through on-screen items one at a time, usually by swiping right or left with one finger, and you activate the currently selected item with a double-tap, rather than jumping freely around a structured page. Both VoiceOver and TalkBack do offer some ways to jump around more directly, such as VoiceOver's rotor, but the everyday default experience on a phone tends to be more step-by-step than the desktop browse-mode experience.

In practice, this means a task like finding the message box may feel different on each platform. On the web app with a desktop screen reader, you might jump straight to it with edit field navigation. On the mobile app with VoiceOver or TalkBack, you are more likely to swipe through the visible items on the current screen until you reach it, since a phone screen generally shows fewer items at once and the swipe-based model is built around that smaller, more linear space. Neither approach is better in every situation. Many blind Claude users find the web app more efficient for longer, more complex work, since browse mode's jump-around navigation and a full keyboard tend to be faster once you know the page's structure, while the mobile app can be convenient for a short question asked on the go, especially when combined with dictation instead of typing. As with the web app, if the exact layout of the mobile app does not match a specific description, fall back on your phone screen reader's normal exploration habits, swiping through the screen slowly and listening to what each item is labeled, rather than assuming any single fixed layout.

Key terms from this week

Use this list to review the vocabulary introduced in this lesson before you start the exercises and the test.

Project

A way to group related conversations together in Claude.ai, optionally with shared custom instructions or reference material.

Artifact

A separate, persistent piece of content, such as a document or piece of code, that Claude sometimes creates alongside its regular conversation reply.

Memory

A feature that lets an AI tool retain certain information about you or your ongoing work across separate conversations, instead of starting from a blank slate each time.

Conversation

A single, ongoing exchange of messages between you and an AI tool, made up of your prompts and the AI's replies, sharing one context window.

Keyboard-only exercises

These exercises use only your keyboard. General screen reader navigation concepts are described below; check your own screen reader's documentation for its exact keys, since these vary between products. Remember that Claude's exact interface can change over time; if any step below does not match exactly what you hear, fall back on your screen reader's list of page elements, heading navigation, and button navigation from this week's lesson to find the equivalent control.

Exercise 1: Start a conversation and read the reply

1. Open your web browser and go to the Claude.ai website. Sign in, or create an account if you do not already have one.
2. Open your screen reader's list of page elements (check your shortcut list for the exact key), switch to a headings view if one is offered, and arrow through it to get a general sense of the page's structure before doing anything else.
3. Close that list, then use your screen reader's edit field navigation repeatedly until you reach the multi-line message box used to start a conversation.
4. Type this exact message: "In two or three sentences, explain what a context window is, for someone who has never heard the term before."
5. Press Enter to send your message. If Enter does not send it, use your screen reader's button navigation to find a nearby Send button and activate it.
6. Wait a few seconds for the reply to finish appearing, then use Say All, or your screen reader's continuous reading command, to read the whole reply.
7. Compare Claude's explanation to the definition of context window you learned earlier in this course. Notice any differences in wording or emphasis.

Exercise 2: Explore the Projects area

1. From the main Claude.ai screen, open your screen reader's list of page elements again.
2. In the headings or links view, look for anything related to "Projects." If you find it, activate it to open that area of the site.
3. If you cannot find it in that list, try landmark or region navigation instead, moving between the page's major regions, listening for one that sounds related to organizing or grouping conversations.
4. Once you reach the Projects area, use heading navigation to explore what is there, such as a way to create a new Project or a list of existing ones.
5. If a way to create a new Project is available, create one and give it a name related to a real goal, such as "Job Search" or a personal interest.
6. Use Say All, or your screen reader's continuous reading command, to confirm what your screen reader announces once the new Project is created, so you know what to listen for next time.

Exercise 3: Upload a file and ask about it

1. Prepare a short text file in Notepad containing two or three sentences about any topic you like, and save it somewhere you can find it easily, such as your Desktop.
2. Start a new conversation in Claude.ai, or continue an existing one.
3. Use your screen reader's button navigation, or its list of page elements filtered to buttons, to find the control for uploading or attaching a file, near the message box.
4. Activate that control, then use your operating system's normal file dialog navigation to locate and select the text file you prepared.
5. Once the file has finished uploading, type a message asking Claude to summarize the file in one sentence, and send it.
6. Read Claude's reply with Say All, and confirm the summary accurately reflects what you wrote in the file.

Portfolio project: My First Claude Conversation

This week's portfolio piece is a real, practiced conversation with Claude, plus a short written reflection you keep for your records. This project puts your new tool skills directly to work on a topic from your own AI Skills Roadmap.

1. Open your AI Skills Roadmap file, and choose one of the six skill areas you marked as "new to me" or "some experience."
2. Open Claude.ai and start a new conversation, or start it inside a Project if you created one during this week's exercises.
3. Have a real conversation with Claude about that skill area: ask it to explain the skill in more depth, ask for a concrete example relevant to your own field or interests, and ask at least one follow-up question based on its answer, practicing the conversation strategies from this week's lesson.
4. Read the full conversation back using Say All, or your screen reader's continuous reading command, and apply the verification habits you've been practicing to at least one specific factual claim Claude made.
5. Open Notepad or your word processor of choice, and write a short reflection, three to five sentences, describing what you asked, what you learned, and how comfortable you felt navigating Claude's interface with your screen reader.
6. Save the reflection file with the name my-first-claude-conversation, in the same portfolio folder you have been using throughout this course.

Weekly test

Answer key

Each answer below includes a one-sentence explanation drawn from this week's lesson.