Last week a colleague from my days at Microsoft wrote an excellent analysis of the current functionality using ChatGPT and a basic question in HTML. I had been dabbling with questions from time to time around accessibility and ChatGPT myself but hadn’t written anything formal up. Now I have.
This evening with a bit of time, I decided to see what the AI technology knew about reading web pages with several leading screen readers. The results confirmed a comment I had made in response to my colleague’s article where I said there was a world of difference between writing well, which ChatGPT does, and writing accurately, which ChatGPT certainly does not due when it comes to reading web pages with a screen reader.
My basic question was as follows:
How do you read a web page using the <name of screen reader>?
Part of what’s key with any AI model is the content that is used for training. I suspect ChatGPT has scoured more content talking about the JAWS screen reader than any other. Answers for not only JAWS but also NVDA and Narrator asserted I should use the Virtual Cursor to read the web page. Clearly there is some model training bias going on in these answers.
ChatGPT was 100% incorrect in instructions for starting some screen readers. The technology instructed me to press the Insert key on my keyboard to start NVDA and Insert+j for JAWS. It was equally wrong on stopping these screen readers also.
In the case of VoiceOver on the iPhone and TalkBack, I guess ChatGPT wanted to make sure I really had the screen reader running. One of the first steps in both cases was to go into settings and then turn the screen reader on. Then further on, after loading the web page, I was given another set of instructions to start the screen reader.
Using the gestures you are told to use for VoiceOver, you wouldn’t be doing much reading but rather a lot of nothing. ChatGPT instructs you to use three-finger swipes to read the web page. This is in fact VoiceOver’s scroll gesture and there is nothing to scroll horizontally on web pages and it is certainly not how you would read.
It is also interesting that for the Windows screen readers, Internet Explorer is still one of the browsers mentioned. Additionally, Edge gets no attention at all.
Anyone familiar with the basic task of reading a web page with a screen reader will recognize other factual errors here.
Details matter! As these sorts of technologies are deployed for broad societal use, it is critical that those deploying the technology recognize this and work not only to eliminate factual errors from making it into the environment but also have ways for immediate correction when such are found.
I understand this is a preview and all but it is critical that such technology is scrutinized. Again, writing well is a far cry from writing accurately and in fact doing the first can compound errors in the second.
Here are the full responses for my questions to ChatGPT.
JAWS
JAWS is a popular screen reader that helps visually impaired users navigate and interact with web pages. Here are the basic steps to read a web page using JAWS
- Open the web page: Use your keyboard to navigate to the web address bar and type in the URL of the page you want to read.
- Start JAWS: Launch JAWS by pressing Insert + J on your keyboard.
- Navigate with the cursor: Use the arrow keys on your keyboard to move the virtual cursor up and down the page. JAWS will read the content of the page as you move the cursor.
- Interact with headings: Press the H key to move from one heading to the next. This can be useful for quickly navigating to different sections of a page.
- Use quick navigation keys: JAWS provides several quick navigation keys that allow you to move to specific types of content, such as headings, links, and form elements. For example, press the T key to move to the next table, or the B key to move to the next button.
- Read specific elements: Use the virtual cursor to move to a specific element, such as a link or a form field, and then press the Enter key to activate it.
- Read context-sensitive help: Press the Insert key followed by the Question Mark key to access JAWS’ context-sensitive help, which provides information on how to interact with the current element.
- Stop JAWS: Press Insert + Q to stop JAWS.
It’s important to note that JAWS provides many other features and keyboard commands that can help you navigate and interact with web pages more efficiently. You can consult the JAWS User Guide or attend a JAWS training course for more information on these advanced features.
VoiceOver
VoiceOver is the built-in screen reader for iOS devices, including the iPhone. Here’s how to use VoiceOver to read a web page:
- Enable VoiceOver: Go to Settings > General > Accessibility, and turn on the VoiceOver switch.
- Launch Safari: Open the Safari app on your iPhone.
- Navigate to a web page: Type in the URL of the web page you want to read or use a search engine to find the page.
- Start VoiceOver: Tap the Home button three times quickly to start VoiceOver.
- Read the page: Swipe left or right with three fingers to move from one item to the next on the page. VoiceOver will read the text or describe the content as you swipe.
- Interact with elements: Double-tap with one finger to activate a link or a button. VoiceOver will announce the action and navigate to the next page if you activated a link.
- Navigate headings: Swipe up or down with three fingers to move through headings, tables, or other structural elements on the page.
- Read specific elements: Touch and hold an element with one finger to hear more information about it, such as the URL of a link.
- Stop VoiceOver: Tap the Home button three times quickly to stop VoiceOver.
It’s important to note that VoiceOver provides many other features and gestures that can help you navigate and interact with web pages more efficiently. You can access the VoiceOver User Guide on your iPhone or attend a VoiceOver training course for more information on these advanced features.
Narrator
Narrator is a screen reader built into Windows operating systems. Here’s how to use Narrator to read a web page:
- Enable Narrator: Press Windows + Ctrl + Enter to turn on Narrator.
- Launch a web browser: Open Internet Explorer, Google Chrome, or another web browser on your computer.
- Navigate to a web page: Type in the URL of the web page you want to read or use a search engine to find the page.
- Start reading: Press the Caps Lock key to start reading the page. Narrator will read the text of the page as you move the cursor with your mouse or the arrow keys on your keyboard.
- Interact with elements: Use the Tab key to move from one element to the next on the page. Press the Enter key to activate a link or a button.
- Navigate headings: Press the H key to move from one heading to the next on the page.
- Read specific elements: Use the virtual cursor to move to a specific element, such as a link or a form field, and then press the Enter key to activate it.
- Stop Narrator: Press Windows + Ctrl + Enter to turn off Narrator.
It’s important to note that Narrator provides many other features and keyboard commands that can help you navigate and interact with web pages more efficiently. You can consult the Narrator User Guide or attend a Narrator training course for more information on these advanced features.
NVDA
NVDA (NonVisual Desktop Access) is a free and open-source screen reader for Windows operating systems. Here’s how to use NVDA to read a web page:
- Download and install NVDA: Download the NVDA installer from the official NVDA website and follow the instructions to install it on your computer.
- Launch a web browser: Open Internet Explorer, Google Chrome, or another web browser on your computer.
- Navigate to a web page: Type in the URL of the web page you want to read or use a search engine to find the page.
- Start NVDA: Press the Insert key on your keyboard to start NVDA.
- Read the page: Use the arrow keys on your keyboard to move the virtual cursor up and down the page. NVDA will read the text of the page as you move the cursor.
- Interact with elements: Use the Tab key to move from one element to the next on the page. Press the Enter key to activate a link or a button.
- Navigate headings: Press the H key to move from one heading to the next on the page.
- Read specific elements: Use the virtual cursor to move to a specific element, such as a link or a form field, and then press the Enter key to activate it.
- Stop NVDA: Press the Insert key on your keyboard to stop NVDA.
It’s important to note that NVDA provides many other features and keyboard commands that can help you navigate and interact with web pages more efficiently. You can consult the NVDA User Guide or attend an NVDA training course for more information on these advanced features.
TalkBack
TalkBack is a screen reader built into Android operating systems. Here’s how to use TalkBack to read a web page:
- Enable TalkBack: Go to Settings > Accessibility > TalkBack and turn on the TalkBack switch.
- Launch a web browser: Open Google Chrome or another web browser on your Android device.
- Navigate to a web page: Type in the URL of the web page you want to read or use a search engine to find the page.
- Start TalkBack: Tap the screen two times quickly to start TalkBack.
- Read the page: Swipe left or right with one finger to move from one item to the next on the page. TalkBack will read the text or describe the content as you swipe.
- Interact with elements: Double-tap with one finger to activate a link or a button. TalkBack will announce the action and navigate to the next page if you activated a link.
- Navigate headings: Swipe up or down with one finger to move through headings, tables, or other structural elements on the page.
- Read specific elements: Touch and hold an element with one finger to hear more information about it, such as the URL of a link.
- Stop TalkBack: Tap the screen two times quickly to stop TalkBack.
It’s important to note that TalkBack provides many other features and gestures that can help you navigate and interact with web pages more efficiently. You can access the TalkBack User Guide on your Android device or attend a TalkBack training course for more information on these advanced features.
This is fantastic, I think I’ve been so focused on looking at what ChatGPT can do that it’s important to take a step back and realize there’s still a lot that it cannot do. The technology is exciting, but I agree, there definitely needs to be a way to correct it when it makes mistakes like this.